WPS6783


Policy Research Working Paper                       6783




  The Impact of Funding Models and Foreign
   Bank Ownership on Bank Credit Growth
             Is Central and Eastern Europe Different?

                                  Erik Feyen
                                Raquel Letelier
                                  Inessa Love
                            Samuel Munzele Maimbo
                                Roberto Rocha




The World Bank
Europe and Central Asia Region
Finance and Private Sector Department
February 2014
Policy Research Working Paper 6783


  Abstract
  This paper provides new evidence on the factors                                   ownership per se does not appear to explain credit
  affecting protracted credit contraction in the wake                               response differences to foreign funding shocks. Rather,
  of the global financial crisis. The paper applies panel                           there is a stronger response in countries that exhibit high
  vector autoregressions to a global panel that consists of                         loan-to-deposit ratios and a high reliance on foreign
  quarterly data for 41 countries for the period 2000–2011                          funding relative to local deposits. The results suggest that
  and documents that domestic private credit growth is                              funding model differences were at the heart of the post-
  highly sensitive to cross-border funding shocks around                            crisis credit contraction in several Central and Eastern
  the world. This relationship is significantly stronger in                         European countries. These findings have important
  Central and Eastern Europe, a region with considerably                            regulatory and supervisory implications for emerging
  stronger foreign presence, higher cross-border funding,                           countries in Central and Eastern Europe as well as for
  and elevated loan-to-deposit ratios compared with the                             other countries.
  rest of the world. The paper shows that high foreign




  This paper is a product of the Finance and Private Sector Department, Europe and Central Asia Region. It is part of a
  larger effort by the World Bank to provide open access to its research and make a contribution to development policy
  discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org.
  The authors may be contacted at efeijen@worldbank.org.




         The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
         issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
         names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
         of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
         its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                       Produced by the Research Support Team
    The Impact of Funding Models and Foreign Bank Ownership on Bank Credit Growth:

                           Is Central and Eastern Europe Different?



     Erik Feyen, Raquel Letelier, Inessa Love, Samuel Munzele Maimbo, and Roberto Rocha 1

                                         The World Bank

                                          Washington DC




JEL Classification: G01, G21, F36.

Key words: Global financial crisis, bank credit, foreign banks, funding models.


1
  Inessa Love is an Associate Professor at the University of Hawaii and a Senior Economist in the World
Bank’s Research Department (ilove@hawaii.edu), Erik Feyen is a Senior Financial Economist in the
World Bank’s Financial Systems Department (efeijen@worldbank.org), Raquel Letelier is a Financial
Analyst in the World Bank’s Europe and Central Asia Department (rletelier@worldbank.org), Samuel
Munzele Maimbo is a Lead Financial Sector Specialist in the World Bank's Europe and Central Asia
Department (smaimbo@worldbank.org), and Roberto Rocha is a Senior Advisor in the World Bank’s
Financial System Department (rrocha@worldbank.org). We thank Katia D’Hulster, Maria Soledad
Martinez Peria, Haocong Ren and participants of the June 2013 World Bank Central European region
retreat for comments and suggestions.
I.   Introduction

By the turn of the new century the financial sector landscape in Central and Eastern Europe and
Central Asia (ECA) had changed dramatically. Most countries in the region were finalizing
comprehensive banking restructuring programs and restoring the solvency of their banking
systems. In many cases these restructuring programs also included extensive transfers of bank
ownership to foreign strategic investors, especially Western European banks. In the years
preceding the global financial crisis these parent banks expanded their operations in the ECA
region through new subsidiaries, and in many cases also provided extensive funding support to
these subsidiaries, taking advantage of easy funding conditions in Western European and global
markets. In several ECA countries domestically-owned banks also took advantage of easy
international funding conditions and borrowed extensively abroad.


This period was characterized by strong bank credit growth to the private sector in most ECA
countries. As shown in Figure 1, credit growth in Central and Southern Eastern Europe (the new
European Union (EU) member states and the Western Balkans) and in the former Confederation
of Independent States (CIS) increased dramatically in the years preceding the crisis, much more
than in other emerging countries. Moreover, credit grew at much higher rates than deposits as
reflected in increasing loan-to-deposit ratios and the increasing share of foreign liabilities—
typically a less stable funding source—in total bank liabilities. The dramatic gains in access to
finance for households and corporates contributed to an impressive expansion of output and
consumption, raising the prospects of rapid convergence of standards of living and integration
with Western Europe. These prospects, combined with a prolonged period of abundant liquidity,
generally outweighed regulatory concerns with excessive credit growth and over-reliance on
cross-border wholesale funding.


The weaknesses of this business model and the cross-border supervisory context in which it
operated were exposed when the fall-out of the US sub-prime crisis reached Europe and both
bank funding and solvency conditions deteriorated rapidly. Although most countries across the
world experienced a severe contraction of credit during the crisis, the ECA region was
particularly affected (Figure 1) given the closer economic and financial links with Western
                                              2
Europe which faced prolonged episodes of financial market stress, including a high degree of
foreign ownership and greater reliance of cross-border funding in many countries. As a result,
the slowdown of credit growth in the ECA region was generally driven by foreign banks, as they
initiated efforts to repair their balance sheets and retrench to home markets.


Concerns with a dramatic credit crunch in the ECA region led to a concerted effort by regulators
and multilateral institutions to preserve minimum levels of cross-border funding, particularly
funding by (Western European) parent banks. This effort – labeled as the Vienna Initiative –
appears to have succeeded in avoiding a very severe credit contraction in the participating
countries (de Haas et al, 2012). Yet, the mere fact that this outcome required a concerted effort
by many actors revealed the potential downside risks associated with foreign banks and cross-
border funding, and triggered a renewed debate about the benefits and costs associated with
foreign banks, cross-border finance, and financial integration.


The crisis triggered an increasing research effort to disentangle the factors that drove the more
pronounced and protracted fall in post-crisis credit growth in most ECA countries. This research
effort is warranted, as some ECA countries with a high degree of foreign ownership but less
reliance on foreign funding seem to have been less affected (e.g. Czech Republic, Poland,
Slovakia), while other ECA countries with a lower degree of foreign ownership but strong
reliance on cross-border funding were very affected (e.g. Kazakhstan, Slovenia, Ukraine).


Research based on bank-level data confirms that foreign banks did contract local credit provision
more than domestic banks in emerging countries, but also that balance sheet structures matter,
including funding structures. Countries that had a strong presence of foreign banks but also
relied more on their domestic deposit base suffered a much less severe credit slowdown. The
balance sheet strength of parent banks also seems to matter—host countries with stronger parent
banks also experienced a less severe credit slowdown.


These results suggest the need to put more emphasis on the discussion of business models,
regulatory standards, and supervisory arrangements, rather than foreign ownership per se. One

                                                 3
potential weakness of this research, however, is the reliance on bank-level data that does not
provide sufficient granularity on funding structures, including the levels of foreign funding. The
exposure to cross-border funding—arguably one of the main transmission channels of
international financial crises—is generally proxied by the ratio of deposits to assets. The fact that
recent research work has found that this variable is significant is reassuring, confirming the
importance of funding structures, but the lack of a variable measuring more accurately the
exposure to foreign funding raises questions about the relevance of these results for the
regulatory debate on cross-border banking.


Our paper contributes to the literature by examining the relative importance of foreign ownership
and foreign funding from a different angle. First, we use country-level data instead of bank-level
data, which allows us to explicitly measure the reliance on foreign funding. Second, we follow a
different methodology—panel vector autoregressions (PVARs)—to examine the relationship
between funding structures and credit growth.


There are several advantages in using the PVAR methodology to study the importance of
different factors in explaining credit growth. First, in VARs, all variables are treated as
endogenous and interdependent so all the feedback effects are explicitly included in the model.
Thus, VARs are designed to explicitly address the endogeneity problem, which is a serious
challenge in studying the empirical relationships between credit growth, economic growth, and
funding sources. Second, unlike single country VARs that need long time series for efficiency,
PVARs can be used with relatively short time-series. This is important for our study since the
quarterly data on foreign liabilities are only available for about the past 10 years. Third, the
PVAR methodology can distinguish between the short-term impacts of each of the factors based
on the impulse-response functions and the long-term cumulative impacts of shocks based on
variance decompositions. In addition, the impulse-response functions based on PVARs can
register any delayed impacts on (and of) the variables under consideration. Fourth, as in any
panel-based method, PVARs allow us to control for country- and year-fixed effects. Country-
fixed effects will capture time-invariant country characteristics that can explain credit growth,
such as institutions, rule of law, credit information and other relatively static features of the

                                                 4
business environment. Year-fixed effects will capture global shocks affecting finance and
growth, such as the effect of the recent financial crisis that is common for all countries in the
same quarter. Thus, PVAR is a methodology that is well-suited to the questions this study aims
to address.


Our main findings can be summarized as follows. Applying PVARs to a global panel that
consists of quarterly data for 41 countries in ECA, Asia, Latin America, and the Middle East and
Africa for the period 2000-2011, we document that private credit growth is highly sensitive to
cross-border funding shocks around the world, and particularly sensitive in the average ECA
country. Moreover, we show that countries with high loan-to-deposit ratios or high reliance on
foreign funding exhibit a stronger response of private credit to foreign funding shocks. Higher
loan-to-deposit ratios make banks more vulnerable to general wholesale funding shocks while
high reliance on foreign funding specifically signifies sensitivity to foreign shocks. At the same
time, we also show that foreign ownership per se does not explain the different credit responses
during the crisis in our sample of countries. These findings are robust to additional sample splits
designed to disentangle the differential impact of ownership and funding structures on credit
growth in ECA and non-ECA countries. Taken together, our findings suggest that funding model
differences with the rest of the world were at the heart of ECA’s protracted post-crisis credit
growth contraction, not its high prevalence of foreign bank ownership.


These findings have important policy implications. Rather than trying to curtail and scale back
foreign bank presence, this paper suggests regulators ought to focus on business models of local
affiliates of international banks and the regulation of cross-border funding. Indeed, Western
European banks are already adjusting their business models by deleveraging, derisking, shifting
to local, more stable sources of funding, and reducing loan-to-deposit ratios (e.g. Impavido,
Rudolph and Ruggerone (2013), Raiffeisen (2013), and Unicredit (2013)). However, eradicating
cross-border funding altogether and promoting a fully domestically funded subsidiary model also
appears suboptimal as it may lead to costly pockets of inert liquidity and capital within the
banking group. This raises various regulatory issues which we will discuss in our concluding



                                                5
comments. Moreover, significant impediments to cross-border funding in countries with low
domestic savings but with the desire to fund productive investments, could curtail growth.


The remainder of our paper is organized as follows. Section II summarizes the recent literature.
Section III describes our data. Section IV outlines our econometric methodology: panel vector
autoregressions (PVARs). Section V presents our main empirical results. Section VI concludes
with policy recommendations. The annex provides an additional discussion of the PVAR
methodology.


II.   Overview of the Literature

It has long been acknowledged that foreign banks may contribute to financial sector development
in emerging and developing markets. This positive contribution may take place by lowering the
costs of financial intermediation and increasing the quality of financial services; increasing
access to financial services for some households and firms, and enhancing the performance of
borrowers. By increasing competition in the host country; bringing in up-to-date technology to
the market; introducing new, more diversified products and services, and pressing regulators to
reform and modernize the regulation and supervision of financial systems, foreign banks can
help countries rapidly accelerate the development of their financial systems (see e.g. Giannetti
and Ongena (2012), and Claessens and Van Horen (2013) for a comprehensive review of the
literature).


At the same time, the crisis has renewed interest in examining the downside risks of foreign
banks. Practically all the research work conducted immediately after the crisis showed that
foreign banks drove the credit boom before the crisis and the credit slowdown after the crisis
(e.g. de Haas Van Lelyveld (2011), de Haas et al (2012), Cull and Martinez Peria (2012), Popov
and Udell (2012)). De Haas et al (2012) also explore the impact of the Vienna Initiative and
show that the initiative succeeded in softening the inevitable deleveraging process in the ECA
region and preventing an uncoordinated “rush to the exit” that would have generated dramatic
consequences for the region. However, they also add that a far better solution (than the ad hoc
Vienna Initiative effort) would entail the creation of a pan-European supervisor for large groups
                                                  6
supplemented by adequate capital and liquidity regulation as well as host-country
macroprudential supervision to curb externally funded credit booms.


More recent research work has included an effort to disentangle more clearly the factors
underlying the relationship between foreign banks and credit in times of crisis. Thus, the recent
literature has tried to distinguish between varying parent bank characteristics, home and host
country profiles and balance sheet structures. In this regard, Claessens and Van Horen (2012)
use a database on bank ownership for 137 countries over 1995-2009 to review foreign bank
behavior and impact and conclude that in the recent financial crisis foreign banks reduced credit
more compared to domestic banks. However, this was not a uniform finding. They also observed
that foreign banks that generated a relative large part of their funding from deposits continued to
lend relatively more in 2009, showing the importance of funding structures.


Kapan and Minoiu (2013) extend this line of investigation by drawing attention to the balance
sheet strength of foreign banks in determining the transmission of financial sector shocks. Using
data from the syndicated loan market, they review the variations in banks reliance on wholesale
funding and their structural liquidity position just prior to the crisis (2007Q2) to estimate the
impact of exposures to market freezes during 2007-08 on the supply of credit. They conclude
that banks with strong balance sheets were better able to maintain lending during the crisis.
Specifically, banks that were more dependent on market funding, had lower structural liquidity,
and were less capitalized reduced the supply of credit more than other banks. They conclude that
their results provide support for the recent regulatory proposals under Basel III.


Choi, Gutierrez, and Martinez-Peria (2013) reach similar conclusions based on an extensive
bank-level panel dataset that allows them to explore the role of subsidiary and parent financial
characteristics, as well as the impact of parent origin and geographic reach. Overall, the analysis
finds robust evidence that foreign banks curtailed the growth of credit relative to other banks,
independently of the host region. However, this was less so for foreign banks in countries that
signed the Vienna Initiative and for those with higher deposit funding ratios and better
capitalized parents. In other words, the funding structure of foreign bank affiliates and the

                                                 7
capitalization of parent banks proved to be significant and distinct factors driving the lending
behavior of foreign banks during the global crisis.


Although there is recognition that both demand and supply factors contributed to the credit
contraction most of the empirical literature does not examine in more detail this critical issue.
Takats (2010) and Feyen and Gonzalez del Mazo (2013) are an exception in this regard by
making an effort to identify supply and demand factors. By using a panel regression framework
to analyze the key drivers for cross-border bank lending to 21 emerging economies between
1995 and 2009, Takats (2010) argues that while both supply and demand factors contributed to
the credit drop, the former that was more important during the crisis. Finding that a 1% increase
in output is associated with an increase of around 0.2% in credit (the most important demand
factor in the analysis is GDP), the paper argues that while demand and supply factors tend to be
more balanced during non-crisis periods (e.g. between 2003 and 2007), during the crisis the
stress experienced by major internationally active banks limited the supply of cross-border
lending and the overall pace of credit creation, i.e., the impact of supply was stronger.


Using a similar framework to disentangle demand from supply effects, Feyen and Gonzalez del
Mazo (2013) argue that the decline in foreign lending activity of Western European banks in a
large sample of emerging economies across the world—consisting of cross-border claims and
claims of local affiliates—was supply driven during periods of high European financial stress.
Similarly, using ECB lending survey data, Feyen and Gonzalez del Mazo (2013) also find that
while falling demand was mostly responsible for the reduction in longer-term credit flows in
Europe after the onset of the global financial crisis, supply factors played a significant role
during periods of financial stress.

III. Data

To model the interaction between our variables of interest using a PVAR approach, we compile a
panel database with quarterly data covering 41 countries in various regions with about 11 years
of data available for most countries. Table 1 provides a description of the variables and the
sources, while Appendix Table A1 shows the availability of each of the key variables by country.
                                                 8
A.    Panel VAR Variables

The variables in our PVAR model are taken from the IMF’s International Financial Statistics
(IFS): PRIVATE CREDIT, FOREIGN LIABILITIES, GDP, and DEPOSITS. We construct our
variables as follows. First, to ensure stationary, we calculate quarter-on-quarter nominal growth
rates in local currencies for all variables in the system. Second, we construct real growth rates by
subtracting the quarter-on-quarter percent change in the CPI index. 2 Third, to remove significant
country-specific seasonality in GDP growth rates, we regress them on country-specific quarter
dummies and subtract the predicted values. Table 2A provides the key descriptive statistics of
the variables in the PVAR model.


PRIVATE CREDIT is the quarter-on-quarter real growth rate of private credit denominated in
local currency. Private credit equals deposit money bank credit to the domestic private sector.
FOREIGN LIABILITIES is the quarter-on-quarter real growth rate of foreign liabilities of
deposit money banks denominated in local currency. These liabilities consist of claims of non-
residents (both financial and non-financial) on the domestic banking system, and include
deposits, securities, loans, financial derivatives and other liabilities. Note that these liabilities
include the claims of foreign banks on their subsidiaries, in the form of loans, deposits, or other
instruments. These foreign liabilities constitute the cross-border linkages between home and host
countries and the channels through which financial shocks are transmitted. GDP is the quarter-
on-quarter, seasonality-adjusted growth rate of real GDP in local currency. DEPOSITS is the
quarter-on-quarter real growth rate of bank deposits. Bank deposits consist of the sum of
domestic demand, time, and savings deposits of deposit money banks.


Table 2A shows that the mean real quarterly growth rates of FOREIGN LIABILITIES were
higher than those of PRIVATE CREDIT AND DEPOSITS, and its standard deviation was
considerably higher, by a factor of three or larger. This result reveals the major weakness of



2
  We also used GDP deflators but this variable limited the sample. All our results hold using the GDP deflator
although they become slightly less significant.
                                                         9
foreign funding – its volatility in periods of crisis. The mean growth rates of FOREIGN
LIABILITIES AND PRIVATE CREDIT foreign were much higher in the case of ECA
countries, relative to Non-ECA countries, reflecting the greater reliance of the ECA region on
foreign funding and the key role played by cross-border finance in ECA’s pre-crisis credit boom.
Interestingly, the standard deviation of FOREIGN LIABILITIES in the ECA and non-ECA
regions are similar, revealing that countries in the two regions are potentially exposed to the
same volatility in periods of crisis, but ECA countries are much more exposed, given their
greater reliance on foreign funding. This result will be further explored in Section V.


B.   Sample Split Variables

One of the objectives of this paper is to understand the factors that drive the response of
PRIVATE CREDIT to other PVAR variables in ECA countries and other emerging countries,
and identify the differences across regions. In doing so, we use various variables hypothesized to
explain the difference to split the sample further. These variables are the loan-to-deposit ratio
(LDR), the share of foreign ownership of the banking system (FOREIGN OWNERSHIP), and
the share of foreign liabilities in total bank liabilities (FOREIGN FUNDING). To gauge the
impact of these variables, we split the sample based on their pre-crisis values, right before the
banking system weaknesses in ECA became apparent. Table 2B reports the pre-crisis country
averages for our three sample-split variables broken for ECA and non-ECA countries. 3
Appendix Table A2 reports individual country-level data.


More specifically, LDR is the loan-to-deposit ratio calculated as the ratio of PRIVATE CREDIT
to DEPOSITS, defined as above from the IFS. LDR proxies the extent to which domestic bank
credit is funded by domestic deposits. An LDR in excess of 100 percent implies that credit is
financed with other sources of funding such as parent funding, unsecured wholesale funding in
international or domestic markets, or other instruments such as covered bonds. Some of these



3
 We use 2007:Q4 values for LDR and FOREIGN FUNDING. We use the closest available observation for
FOREIGN OWNERSHIP, which is 2008 for most countries.

                                                10
funding sources are typically short-term and highly sensitive to market conditions, in contrast
with retail deposits which are usually more stable. FOREIGN FUNDING is a proxy for the
banking system’s reliance of foreign sources of funding relative to total funding, calculated as
the ratio of FOREIGN LIABILITIES to the sum of FOREIGN LIABILITIES AND DEPOSITS.
FOREIGN FUNDING complements LDR and provides information on the extent to which the
deposit funding gap captured by the LDR measures is filled with foreign sources of funding.
Lastly, FOREIGN OWNERSHIP measures the fraction of banking system assets that are
majority-owned by non-residents. These data are taken from the World Bank’s Bank Regulation
and Supervision Survey.


Table 2B shows that LDR, FOREIGN FUNDING and FOREIGN OWNERSHIP are
significantly higher in ECA countries, relative to non-ECA countries, although the standard
deviations also suggest that there is substantial variation across countries within each of the two
regions. This is an important result that will also be further explored in Section V.

C.   Correlations

Panel A of Table 3 presents the correlations between our PVAR variables for the full sample, as
well as for ECA and the rest of the sample (“non-ECA”). As expected, the correlation between
DEPOSITS and PRIVATE CREDIT is high in all samples. We also observe that FOREIGN
LIABILITIES are correlated with PRIVATE CREDIT (0.40). However, this correlation is almost
twice as high in ECA compared to non-ECA (0.50 vs. 0.23). This suggests that credit in the ECA
region is more vulnerable to foreign funding shocks than in the rest of the world, a hypothesis we
formally test with PVARs in Section V.


Panel B of Table 3 displays the correlations for the sample split variables and clearly shows the
different funding model in ECA, reflected in the high correlation between FOREIGN FUNDING
and LDR in that region and the absence of a correlation between these two variables outside the
region. Indeed many banks in ECA relied on foreign funding to supplement domestic deposits
and sustain high credit growth, resulting in high loan-to-deposit ratios (see also Figure 1). Some
countries in the non-ECA region have high LDR ratios but these reflect primarily domestic
                                                 11
liabilities. We also find that FOREIGN FUNDING and FOREIGN OWNERSHIP are
significantly and positively correlated in the full sample implying that foreign bank ownership is
generally associated with higher cross-border funding. Interestingly, the correlation between
these two variables is not significant in the two sub-samples. This reflects the fact that both
FOREIGN OWNERSHIP and FOREIGN FUNDING are generally higher in ECA compared to
non-ECA countries, but also that within each sub-sample there is a greater variety of
combinations (see Table 4, Panel B and Table A2). More specifically, Central and Southern
European countries tend to have high levels of foreign ownership but different levels of parent
funding, while CIS countries tend to have lower levels of foreign ownership but domestic banks
in some of these countries have borrowed significantly from wholesale markets abroad.

IV. Methodology 4

A.       Panel Vector Autoregressions (PVARs)

PVARs combine the advantages of the traditional Vector Autoregression (VAR) approach with
the panel-data approach. Our PVAR methodology follows Love and Zicchino (2006). Basic
VARs allow for the simultaneous analysis of the evolution of a system of endogenous variables.
The prime benefit of VARs is that the dynamic impact of orthogonal shocks can be evaluated—
i.e. the isolated impact of a shock of one variable on the system over time, keeping the shocks of
the other variables equal to zero. At the same time, the panel-data dimension of PVARs allows
for unobserved, time-invariant heterogeneity of the cross-sectional units (e.g. countries) to be
controlled for with fixed effects.


The PVAR technique is therefore particularly suitable for our purposes because we seek to
model the evolution of a system of four variables of interest—PRIVATE CREDIT, FOREIGN
LIABILITIES, GDP, AND DEPOSITS—in a set of countries which significantly differ along
various dimensions such as the level of financial and economic development and the quality of
regulatory and institutional frameworks. Moreover, these countries are faced with common
exposures such as global financial and economic shocks and extraordinary measures of large


4
    See Annex for further technical details on PVARs.
                                                        12
central banks. Controlling for these common exposures is particularly relevant because our
sample includes the onset of the global financial crisis during which financial stress rapidly
spilled over to other parts of the world significantly increasing the size of common risk factors.


The structure of our baseline PVAR model can be written in reduced form as follows:
                                        Ζit = Γ0 + Γ1 Ζit-1 + fi + dt +eit (1)

where Ζit is a vector of our four variables for quarter t and country i and modeled as a function of
the first-order lags of all variables in the system. The fi and dt terms are dummy variables for
country i and quarter t which capture time-invariant country-specific effects and time-varying
common effects, respectively.


To minimize the influence of outliers, we winsorize all variables by replacing the data below the
1st percentile with the value of the 1st percentile. Similarly, we replace the data above the 99th
percentile with the value of the 99th percentile. 5

B.    Impulse-response Functions and Variable Orderings

One of the advantages of PVARs is their ability to model the impact of an isolated shock or
innovation of a single variable on the whole system over time, while setting the innovations of
all other variables equal to zero. This produces so-called impulse-response functions which
display a point estimate of the response and a corresponding confidence interval for several post-
shock periods. The response can be said to be statistically significant if the confidence interval
does not include the horizontal axis (i.e. the zero line) of the impulse-response function. Impulse-
responses take into consideration the estimated coefficients matrix, as well as the correlation of
residuals across equations.


However, the errors across the variables in the system are typically correlated, which inhibits the
attribution of the impact of an innovation of a single variable to that variable only. To isolate the


5
 We choose to winsorize the data to avoid a significant reduction in our sample size by dropping these outliers.
However, our results are robust to excluding these observations from the system (results available upon request).
                                                         13
impact of innovations, it is necessary to decompose the residuals in such a way that they become
orthogonal. The common solution is to adopt a particular variable ordering which assumes that
variables that come earlier in the order affect all following variables contemporaneously, while
variables that come later affect previous variables only with a lag. That is, in the VAR model all
the variables are treated as endogenous, the ordering only affects the timing of the responses and
the variables that come later in the ordering have a delayed response on the variables that come
earlier in the ordering. As a result of this process, the correlation between the residuals of two
variables is allocated to the variable that comes first in the ordering. The ordering can therefore
have implications for the shape of impulse-response functions and for variance decompositions.


In our baseline model, we assume the following ordering: FOREIGN LIABILITIES, GDP,
DEPOSITS and PRIVATE CREDIT. We place FOREIGN LIABILITIES first since it is to a
large extent driven by external supply factors such as global risk appetite, parent bank health,
economic home conditions, and global funding markets. This assumption implies that FOREIGN
LIABILITIES affects all other variables contemporaneously. In contrast, the other variables can
only affect FOREIGN LIABILITIES with a 1-quarter lag. This is a reasonable assumption since
reversing the flow of foreign liabilities is likely to take some time.


We place PRIVATE CREDIT last in the order because arguably PRIVATE CREDIT can react to
all other factors quickly, i.e. in the same quarter; however, the PRIVATE CREDIT only affects
other variables with a 1-quarter lag. Indeed, typically there is a delay between loan origination
and loan deployment, so an impact on other variables can only be expected with a lag.


Having established FOREIGN LIABILITIES and PRIVATE CREDIT as first and last in the
order, respectively, in our baseline model we put GDP second, followed by DEPOSITS. Our
rationale is that GDP has a more immediate impact on DEPOSITS as money demand responds
more quickly to changes in GDP, while changes in DEPOSITS are likely to affect GDP only




                                                  14
with a lag. However, the alternative ordering in which DEPOSITS enters second and GDP third
produces qualitatively similar results. 6


In this paper, we aim to understand the differences in responses of PRIVATE CREDIT to shocks
in various sub-samples to gauge whether i) there are differences between country groupings and
ii) what factors can explain those differences. In our first sample split, we explore whether there
are differences between the ECA region and the rest of the sample. Second, we split the full
sample based on the three sample split variables that highlight the differences in ECA’s banking
model: LDR, FOREIGN OWNERSHIP, and FOREIGN FUNDING. We use the pre-crisis
median value of each of these variables to split the full sample into high and low subsamples of
equal size. After each split, we produce separate impulse-response functions for each subsample
for visual inspection. To test for statistical significance between the subsamples we also produce
impulse-response functions of the response difference of the two subsamples and use the
associated confidence interval to determine statistical significance.

C.    Variance Decompositions

Besides impulse-response functions, we also employ variance decompositions to understand the
cumulative impact of the shock of a particular variable on the system. More specifically,
variance decompositions show the percent of the variation in one variable after a certain amount
of time that is explained by the shock of another variable and therefore provide an indication of
the magnitude of the total effect a variable exerts on another. We report the total effect
accumulated over 10 quarters, but longer time horizons produced equivalent results since after 10
quarters the effect of a shock has mostly worked its way through the system.

V.    Empirical Results

A.    T-tests: Is ECA Different from the Rest of the World?



6
  The order for our alternative model is: FOREIGN LIABILITIES, DEPOSITS, GDP and PRIVATE CREDIT. The
main results are not affected. However, in the variance decomposition, the impact of GDP on PRIVATE credit is
slightly smaller and the impact of DEPOSITS is slightly larger compared to the baseline model (results available
upon request).
                                                       15
Before presenting our PVAR results, we first ascertain the presence of differences between ECA
and the rest of the world. In doing so, we split our sample into ECA and non-ECA subsamples
and run basic mean comparison t-tests to document statistically significant differences in the
subsample means of an indicator of interest. By running t-tests for different periods, we are able
to track the evolution of our variables in ECA and non-ECA before (2005-07), during (2008-
2009), and after (2010-2011) the global financial crisis.


Table 4 presents the results. Panel A documents the ECA vs. non-ECA differences for the PVAR
variables: PRIVATE CREDIT, FOREIGN LIABILITIES, GDP, and DEPOSITS. We observe
that in the pre-crisis period, ECA was significantly different from the rest of the sample in terms
of the growth of foreign liabilities, private credit, and deposits. The quarterly real rate of growth
of foreign liabilities and private credit for ECA was more than double the respective rates of
growth in non-ECA countries: 10.06% vs. 3.64% for foreign liabilities and 7.87% vs. 3.15% for
private credit. These differences are statistically significant and reveal the distinctiveness of
ECA’s business model, which consisted of rapid private credit growth sustained by large inflows
of foreign funding. The resulting ECA credit boom contributed to higher average real deposit
growth (5.42% vs. 2.65%) and in real GDP growth (0.95% vs. 0.51%) compared to non-ECA.


Next, we study the PVAR variables during the peak of the crisis in 2008-2009. During that
period, credit growth declined in both regions, but relatively more in ECA, due to the sharp drop
in foreign liabilities. Private credit growth remained higher in ECA (2.15% vs. 1.45%), but the
difference is only marginally significant in this period, in contrast with the preceding period.
Foreign liabilities growth declined significantly but remained higher in ECA (2.33% vs. 0.74%),
reflecting the impact of the Vienna Initiative. However, the difference of foreign liabilities was
not statistically significant in this period, reflecting substantial variations in the sample, including
the ECA subsample. Note that economic activity contracted significantly in ECA during this
period (-1.86% vs. -0.51%), driven by the sharper slowdown in credit and other factors (exports,
remittances, tourism).




                                                  16
In the immediate post-crisis years (2010-2011), many non-ECA countries started their recovery
as evidenced by the acceleration of foreign liabilities, private credit, and deposit growth. In
contrast, the growth of foreign liabilities and credit in ECA were significantly slower, reflecting
the onset of the EU’s sovereign and banking crisis, its protracted effects on cross-border lending,
and ECA’s greater dependency on foreign funding. The continuation of the crisis in the EU is
also reflected in the lower GDP and deposit growth rates in ECA, although the differences are
not statistically significant.


Given such distinctive differences in the impact of the global crisis and the path of recovery in
ECA vs. non-ECA economies, we seek to understand what key pre-crisis features of financial
systems in ECA could be responsible for such differences. As discussed, we hypothesize the
differences could be attributed to dissimilarities in the funding model and/or the extent of foreign
ownership. Therefore, to gauge business model differences we use t-tests on LDR, FOREIGN
OWNERSHIP and FOREIGN FUNDING. Table 4, Panel B shows the results.


The average LDR in ECA is significantly higher than in non-ECA (127% vs. 101%). The higher
LDR shows that ECA banks relied much more on non-deposit, external funding and had
significantly less liquid assets to deal with a liquidity shock compared to non-ECA banks.
Moreover, since ECA’s LDR exceeds 100% by a wide margin, ECA had to fund its liquidity gap
from other sources, especially of foreign nature, as evidenced by the fact that FOREIGN
FUNDING was twice as high in ECA compared to non-ECA (32% vs. 11%). While these
foreign sources of funding can act as stabilizers when stress originates from the domestic
financial sector (e.g. De Haas and Van Lelyveld, 2006), they can also function like conduits of
foreign financial stress (e.g. Cetorelli and Goldberg, 2010). The ECA funding model contrasts
sharply with the one of foreign bank affiliates in Latin America which typically source their
funding in domestic markets—mostly retail deposits—which impedes excessive credit expansion
and the transmission of foreign stress.


At the same time, we find that foreign ownership is also very high in ECA. In the average non-
ECA country, 33% of banking assets are majority-owned by non-residents compared to 65% in

                                                17
the average ECA country. Therefore, many ECA countries have relatively high foreign bank
ownership, high foreign funding dependence, and high loan-to-deposit ratios. As such, it is
difficult to disentangle the impact of foreign ownership from the impact of foreign funding based
on sample averages. In addition, such sample averages cannot distinguish whether demand (e.g.
GDP) or supply (e.g. foreign liabilities and deposits) factors drove the decline of the private
credit because all these variables deteriorated simultaneously during the crisis. To address these
issues we now turn to PVARs.

B.   The Baseline Model for the Entire Sample

Our baseline PVAR models the interaction between PRIVATE CREDIT, FOREIGN
LIABILITIES, GDP, and DEPOSITS using Generalized Method of Moments (GMM)
estimations for our entire sample of 41 countries. As discussed, the PVAR approach allows us to
exploit the panel structure of our sample while at the same time modeling the dynamic
interactions between our variables and how they behave in response to shocks. Importantly, the
PVAR allows us to separately study demand and supply factors to better understand the
evolution of PRIVATE CREDIT. Because we include all countries, the baseline model will
provide a good idea of the basic interactions between the variables for the average country and
help us interpret the findings when we start splitting the sample.


For completeness, we report the coefficient estimates of our reduced form PVAR models for the
full sample, and the ECA and non-ECA subsamples in Table 5. Since our main focus is to isolate
how shocks to one variable affect another variable, we don’t focus on the coefficient estimates
and instead turn our attention to the impulse-response functions.


Figure 2 shows 4×4=16 impulse-response functions for our baseline model. Each row
corresponds to a particular variable and shows 4 impulse-response functions which display how
this variable responds to an isolated shock of each of the variables in the system (i.e. including
the variable itself). Each response is traced for 6 periods (1.5 years) after which the shock has
mostly worked its way through the system and any residual impact is minimal. For example, the
graph in row 1, column 4 reports the response of FOREIGN LIABILITIES to a shock in
                                                18
PRIVATE CREDIT and the graph in row 4, column 1 reports the response of PRIVATE
CREDIT to a shock in FOREIGN LIABILITIES. Because PRIVATE CREDIT comes later in the
ordering, its impact on FOREIGN LIABILITIES is delayed by one period (and hence there is no
effect at time zero), while the impact of FOREIGN LIABILITIES on PRIVATE CREDIT is
immediate and positive at time zero. Each graph shows the point estimates of the impulse-
response function (the middle line) as well as the 5% and 95% confidence bounds (the top and
the bottom lines) which are based on Monte Carlo simulations (see Annex). The confidence
interval allows us to gauge whether a shock triggers a statistically significant response by
inspecting whether the interval excludes the x axis (i.e. the zero line).


Given our research objective, we focus on row 4 which displays how PRIVATE CREDIT
responds to various shocks. The upper part of Table 6 presents the point estimates of the
impulse-response of PRIVATE CREDIT to each of the factors for the 6 periods which are
plotted in Figure 2 in row 4. We start with the impact of our supply factors. We document a
significantly positive response of PRIVATE CREDIT to a shock in FOREIGN LIABILITIES.
This suggests that FOREIGN LIABILITIES is a significant driver of PRIVATE CREDIT in the
entire sample: a one standard deviation shock in FOREIGN LIABILITIES results in a 1.24%
increase in PRIVATE CREDIT at time zero which is large given that average PRIVATE
CREDIT in our entire sample is 3.2% and its standard deviation is 5.1% (see Table 2A). This is a
key result which we will study further in order to understand response differences in ECA vs.
non-ECA countries. Next, we find a positive response of PRIVATE CREDIT to GDP, which
captures the demand for credit. This response is also statistically significant, but somewhat
smaller in magnitude: 0.59% at time zero. Finally, we observe a positive and significant response
of PRIVATE CREDIT to a DEPOSITS shock: 1.53% at time zero. Thus, we conclude that all
our supply and demand factors are significant drivers of PRIVATE CREDIT.


Various other impulse-response functions in Figure 2 show interesting results that are
statistically significant. FOREIGN LIABILITIES responds positively to a PRIVATE CREDIT
shock (row 1, column 4) suggesting that a sudden increase in credit can be funded by FOREIGN
LIABILITIES which can typically be attracted on short notice. These results are clearly caputing

                                                 19
the mobilization of foreign funding from both parent banks and wholesale markets abroad during
the credit boom period. At the same time, DEPOSITS also respond positively to a PRIVATE
CREDIT shock (row 3, column 4) but the magnitude is smaller than for foreign liabilities,
arguably because it is more difficult to significantly raise deposits in the short term. Together,
these findings imply that foreign sources of funding can be useful to temporarily fill domestic
funding gaps. We also confirm that a PRIVATE CREDIT shock boosts GDP, which is expected
since private credit typically expands consumption and investment. This corroborates with
results of the finance and growth literature (e.g. Levine, 1997). Finally, all variables have a
positive and significant response to their own shocks (the charts on the diagonal of Figure 2), but
this is simply a mechanical result.


Impulse-response functions are useful to study the short-term response of each variable to a
shock in another variable. To examine the long-term cumulative impact of a shock we compute
variance decompositions. Table 7, Panel A displays the variance decomposition for our baseline
model. Each cell in the table shows what percent of variation in the row variable is explained by
the column variable after the 10 quarters, i.e. 2.5 years. Note that by construction shocks to a
variable of its own lag explains most of its variance (i.e. the diagonal of Table 7, Panel A
contains the largest values).


Row 4 of Table 7, Panel A reports that a FOREIGN LIABILITIES shock explains a large part of
the variance in PRIVATE CREDIT (9.7%). DEPOSITS shocks explain an even larger portion of
the variance (14.2%), but this is expected, given that typically DEPOSITS account for the largest
share of total bank liabilities in all countries, including countries that rely more on foreign
funding. The fact that FOREIGN LIABILITIES explain a large part of the variance of PRIVATE
CREDIT, almost equal to that of DEPOSITS, despite representing typically a much smaller share
of total bank liabilities, reveals again the downside risks associated with cross-border borrowing.
By contrast, shocks to GDP do not seem to contribute to the variance in PRIVATE CREDIT
(2.4%). This could be due to longer response lags or noise in the quarterly GDP data in some
countries in the sample.



                                                20
Taken together, our PVAR model explains a substantial portion of the variation in PRIVATE
CREDIT (26.2% is explained by other variables and 73.8% is explained by its own shocks). A
smaller portion of the variation in DEPOSITS and FOREIGN LIABILITIES is explained by
other variables: 7.4% and 3.3%, respectively. The explanatory power of GDP is substantially
lower (less than 1% is explained by non-GDP factors). However, our objective is to build a
predictive model for PRIVATE CREDIT and our model fulfills this objective. Our next goal is to
compare the impact of various shocks to private credit growth in different country samples.
Therefore, we will use these findings as our reference point for our subsample results.


Before we proceed with our sample splits, we conduct various robustness checks to our baseline
model. First, our baseline impulse-response functions and variance decompositions do not
change significantly when we i) include an additional lag in the model 7 or ii) change the ordering
of the variables where we interchange DEPOSITS and GDP (results available on request).
Second, we find that dropping large outliers (i.e. extreme observations above 99th percentile and
below 1st percentile) rather than winsorizing the data do not affect the results (results available
on request). Therefore, in the remainder of this paper, we use the ordering of the baseline model
and include 1 lag only.

C.    Sample Splits: Is ECA Different from the Rest of the World?

After having established the baseline impulse-response functions and variance decompositions
for the whole sample, we set out to explain whether PRIVATE CREDIT in the ECA region
responded differently compared to the rest of the world. Indeed, our basic sample mean analysis
suggested that ECA experienced a deeper PRIVATE CREDIT contraction which persisted for
several more years relative to the rest of the world.




7
  Adding an additional lag to the model significantly increases the number of coefficients that have to be estimated –
i.e. from 16 in our baseline model to 32 – and reduces the number of observations from 1,591 in the baseline model
to 1,554 because one year of data is lost. The loss in degrees of freedom is particularly relevant for models for
subsamples.
                                                         21
We therefore split our full sample into ECA and non-ECA subsamples and run separate PVARs
for each subsample. The two subsamples are roughly equal in size: (ECA: 22 countries, 947
observations; non-ECA: 19 countries, 820 observations). Figure 3 presents the results. For space
considerations, we only present impulse-response functions and variance decompositions for
PRIVATE CREDIT, our key variable. The first and second rows in Figure 3 show the PRIVATE
CREDIT impulse-response functions to shocks of all four variables for the ECA and non-ECA
subsamples, respectively (“Sample: ECA” and “Sample: Non-ECA”). Both rows exhibit very
similar patterns to our baseline findings and confirm our key result: the significant and positive
response of PRIVATE CREDIT to a FOREIGN LIABILITIES shock. We also observe a positive
and significant response in both samples to GDP and DEPOSITS shocks.


The key question, however, is whether the magnitude of the PRIVATE CREDIT response to a
FOREIGN LIABILITIES shock is significantly different in a statistical sense between the ECA
and non-ECA subsamples. Visual inspection of the 95th percentile bound of the first two rows
shows that response to a FOREIGN LIABILITIES shock is substantially larger in the ECA
sample: 1.71% vs. 1.10% in non-ECA. Table 6 reports the actual point estimates for the impulse-
response functions in period zero: 1.53 and 0.89 in ECA and non-ECA, respectively. The ECA
response is thus 0.64 percentage points higher (72%), which is an economically relevant
difference. The difference is even more pronounced after 1 quarter: the response in period 1 in
ECA is 0.74, while in non-ECA it is 0.23, which is about 3 times lower. This pattern continues
for several quarters – even in quarter 3 the ECA response is at 0.28, while the non ECA is at
0.07. Thus, the impact of FOREIGN LIABILITIES shocks in ECA is not only stronger in
magnitude, but also lasts longer.


To assess whether this difference is also statistically significant, we calculate the impulse-
response functions of the difference between ECA and non-ECA. Figure 3, row 3 presents the
result (“Sample: Difference”). Row 3, column 1 confirms the difference is indeed significant (i.e.
the zero line does not fall within the confidence interval). The other impulse-response functions
in row 3 show that PRIVATE CREDIT does not behave differently in ECA in response to a GDP
shock, while there is a slightly larger response to a DEPOSIT shock after 1 period.

                                               22
Lastly, we compute the variance decompositions in the ECA and non-ECA subsamples to study
the cumulative longer-term impact of various shocks on PRIVATE CREDIT. Table 7, Panel B
reports the results. To save space, we only report the decompositions for the PRIVATE CREDIT
variable. The first row is the baseline decomposition for the whole sample, replicated from Panel
A. The second and third rows show the decompositions for the ECA and non-ECA subsamples,
respectively. A FOREIGN LIABILITIES shock explains 6.0% of the PRIVATE CREDIT
variation in non-ECA countries which is substantially less than the baseline of 9.7% (see Table
7, Panel A). In contrast, a FOREIGN LIABILITIES shock explains 12.9% of the variation in the
ECA subsample, more than twice as large as in the non-ECA group. These findings establish the
second of our key results: PRIVATE CREDIT in ECA has been more heavily influenced by
shocks to FOREIGN LIABILITIES. Unlike the case of Non-ECA countries, shocks to
FOREIGN LIABILITIES in ECA countries explain almost the same share of the variations of
PRIVATE CREDIT as DEPOSITS, despite the fact that FOREIGN LIABILITIES account for a
smaller share of bank liabilities than DEPOSITS, even in ECA. This result reflects both the
larger exposure of ECA countries to cross-border finance (larger FOREIGN FUNDING) and the
much higher volatility of FOREIGN LIABILITIES in periods of crisis, as reported in Table 2A.

D.   Further Sample Splits: Explaining Differences in Private Credit Responses

After having established that PRIVATE CREDIT is significantly more responsive to FOREIGN
LIABILITIES shocks in ECA compared to non-ECA countries, this section seeks to identify the
factors that drive the difference—different factors will imply different policy implications. As
discussed above, the banking sector in ECA is markedly different from those in non-ECA
countries along our sample split variables: ECA countries exhibit significantly higher foreign
ownership (FOREIGN OWNERSHIP), higher reliance on foreign funding (FOREIGN
FUNDING), and higher LDR ratios (LDR).


To further investigate which of these factors is driving the observed difference between the ECA
and non-ECA subsamples, we perform sample splits using each of the three sample-split
variables one at a time. In each case we split the whole sample in two subsamples of equal size
                                               23
based on the median value of the variable in question. The sample-split approach is similar to
interacting each of the factors with the responsiveness of private credit to foreign liabilities. 8


First, we test whether LDR is driving the response differences. We calculate the pre-crisis
median LDR ratio in our full sample using data from the fourth quarter of 2007. We then split
the full sample into high and low LDR subsamples of equal size based on the median and rerun
our baseline model for each sample. The impulse-response results are presented in Figure 4 and
the variance decompositions are presented in Table 7, Panel B. We find that the difference
between the response of PRIVATE CREDIT to a FOREIGN LIABILITIES shock in both high
LDR and low LDR samples is positive and statistically significant. 9 In addition, the responses
differ significantly in magnitude. The variance decomposition shows that in the high LDR
subsample, a FOREIGN LIABILITIES shock explains 14.1% of the variation in PRIVATE
CREDIT, while this is only 5.6% in the low LDR sample. This finding suggests that high LDRs
are associated with a stronger response of PRIVATE CREDIT to FOREIGN FUNDING shocks
and consistent with the notion that high LDRs are at least partially responsible for the ECAs
weak crisis and recovery experience while the rest of the world proved more resilient.


Next, we perform another sample split and run new PVARs to test whether foreign funding
dependence is driving the difference. In doing so, we use pre-crisis FOREIGN FUNDING values
for all countries from the fourth quarter of 2007 and split the sample into high and low
FOREIGN FUNDING subsamples. The impulse-response results are presented in Figure 5 and
the variance decompositions are presented in Table 7, Panel B. We find that the difference
between the response of PRIVATE CREDIT to a FOREIGN LIABILITIES shock in high and
low FOREIGN FUNDING samples is positive and statistically significant. Again, we find a
strong difference in the magnitude of response to shocks. The variance decompositions show that
in the high FOREIGN FUNDING subsample a FOREIGN LIABILITIES shock explains 12.9%
of the variation in PRIVATE CREDIT, while it only explains 6.9% in the low FOREIGN



8
  Note that such interactions cannot be modeled directly in a VAR setting.
9
  The significance is at about 5% in period 1 because the bottom 5th percentile line is touching the zero line, while it
is stronger in periods 0 and period 2 and 3.
                                                           24
FUNDING subsample. These results suggest that high reliance of countries on foreign funding is
associated with stronger response of PRIVATE CREDIT to FOREIGN FUNDING. Again, the
results are in line with the hypothesis that high reliance on foreign funding is responsible for
explaining the difference between the ECA and non-ECA samples.


Lastly, we split our full sample using FOREIGN OWNERSHIP. Based on the latest pre-crisis
data, we create high and low FOREIGN OWNERSHIP subsamples. The impulse-response
results are presented in Figure 6 and the variance decompositions are presented in Table 7, Panel
B. We find that the difference between the responses of PRIVATE CREDIT to a FOREIGN
LIABILITIES shock in the high and low FOREIGN OWNERSHIP samples are not statistically
significant (row 3). The variance decompositions show that in the high FOREIGN OWNERSHIP
subsample a FOREIGN LIABILITIES shock explains 10.0% of the variation in PRIVATE
CREDIT while it explains 9.1% in the low FOREIGN OWNERSHIP subsample, which is not
materially different. These results suggest foreign ownership itself is not associated with a
stronger response of PRIVATE CREDIT to FOREIGN LIABILITIES, in contrast with high
LDR and high FOREIGN FUNDING. Therefore, we find no evidence that foreign ownership per
se is responsible for the observed differences between ECA and non-ECA countries in their
impact of a FOREIGN LIABILITIES shock on PRIVATE CREDIT.


Taken together, our findings suggest that high LDR and high FOREIGN FUNDING in ECA
have significantly contributed to the difference in responses of PRIVATE CREDIT to FOREIGN
LIABILITY shocks in ECA vs. non-ECA countries. In contrast, we do not find that high
FOREIGN OWNERSHIP drives the difference. This implies that the funding model of banks in
ECA is at the heart of ECA’s vulnerability to external financial shocks rather than the strong
presence of foreign banks in the region per se.

E.   What factors explain differences of ECA with the rest of the world?

Our previous results suggest that the differences in the impact of FOREIGN LIABILITY shocks
on PRIVATE CREDIT between ECA and other regions are not driven by differences in
FOREIGN OWNERSHIP. Instead, these differences seem to be driven primarily by the lack of a
                                                  25
more stable funding model, reflected in high loan-to-deposit ratios (LDR) and much greater
reliance on foreign funding (FOREIGN FUNDING).


In this section we provide further evidence on the factors that drive the differences in ECA and
the rest of the world. While all three factors – LDR, FOREIGN FUNDING and FOREIGN
OWNERSHIP – are higher in ECA, not all factors are high in all countries at the same time, as
we discussed above. This allows us to investigate which of the three factors is driving the
difference between ECA and non-ECA in terms of the response of private credit to foreign
funding shocks. In doing so, we adopt the following procedure. First, we create “truncated ECA
samples” by removing countries that are high on one of the characteristics that we are interested
in—FOREIGN OWNERSHIP, FOREIGN FUNDING, and LDR. Such a “truncated ECA
sample” is more similar to the rest of the world in terms of the selected characteristic (i.e.
FOREIGN FUNDING), but can be different in terms of the other two characteristics (i.e. LDR
and FOREIGN OWNERSHIP). Then we run PVARs on both samples and compare the results of
the “truncated ECA sample” with the rest of the world. We do this subsequently for each of our
three variables of interest. If we still find a difference between ECA and the rest of the world, we
conclude that the specific characteristic appears not to be driving such difference (and vice
versa). Table 4, Panel C provides the summary statistics of the various truncated ECA samples.


First, we remove from the ECA sample countries that are high on FOREIGN FUNDING. We use
the pre-crisis median of FOREIGN FUNDING to construct this truncated sample. The resulting
sample contains ECA countries with relatively low FOREIGN FUNDING. However, even the
relatively low FOREIGN FUNDING countries in ECA still have significantly higher FOREIGN
FUNDING (20% in the truncated ECA sample, vs. 11% in the non-ECA sample). This stacks the
cards towards confirming the differential impact in the truncated ECA sample and the non-ECA
sample. We then compare the truncated ECA sample with the non-ECA sample. Figure 7
presents the differences between the samples in the impulse-responses of the response of
PRIVATE CREDIT to a FOREIGN LIABILITIES shock. Graph in row 1, column 1 shows that
the difference between the truncated ECA sample and non-ECA sample is not significantly



                                                26
different from zero. This suggests that FOREIGN FUNDING is indeed a factor that explains the
response difference between the (full) ECA sample and the non-ECA sample.


Second, we remove from the ECA sample countries that are high on LDR. Again, we use the
pre-crisis median LDR in ECA to create the truncated sample. The resulting truncated ECA
sample is not significantly different from non-ECA sample in terms of LDR. The graph in Figure
7 row 1, column 2 shows that again, there is no longer a significant difference between ECA and
non-ECA samples. Thus, LDR appears also a factor responsible for the observed response
differences between the (full) ECA sample and non-ECA sample.


Third, we remove from the ECA sample countries that are high on FOREIGN OWNERSHIP.
Again, we use the pre-crisis median FOREIGN OWNERSHIP in ECA to make the truncated
sample. The results in row 2, column 1 in Figure 7 show that there still a significant response
difference between the truncated ECA sample and the non-ECA sample.


Before drawing conclusions, Table 4, Panel 4 shows that even the truncated ECA sample still has
a relatively high level of FOREIGN OWNERSHIP. Average FOREIGN OWNERSHIP in the
truncated sample is 39%, while it is 33% in the rest of the world. However, this difference in not
statistically significant. Nevertheless, to ascertain the finding, we perform a stricter test. We drop
four remaining high FOREIGN OWNERSHIP countries (Serbia (75%), Latvia (68%), Poland
(68%) and Armenia (60%)).            This further truncated ECA sample has average FOREIGN
OWNERSHIP of 23%, which is below the average FOREIGN OWNERSHIP in the non-ECA
sample (i.e. 33%), but not significantly different. We then proceed to compare the behavior of
this ECA sub-sample with the non-ECA sample. Again, we find the two samples show
statistically significantly different responses of PRIVATE CREDIT to FOREIGN OWNERSHIP
(row 2, column 2). 10 In other words, even when the ECA looks statistically the same in terms of



10
  We have performed this test after truncating the sample even further – by removing all countries with foreign
ownership above 52%, which removed 3 additional countries: Armenia (60%), Latvia (68%) and Poland (67%). The
resulting truncated sample has only 7 ECA countries and has the average foreign ownership of 23%, which is way
below the non-ECA average of 33% (but not statistically different). Nevertheless, we still find significant
                                                      27
average FOREIGN OWNERSHIP as our non-ECA sample, there is still a significant difference
in the response of PRIVATE CREDIT to FOREIGN FUNDING shocks. Therefore, we conclude
that foreign ownership does not appear to be the factor that is driving the difference of between
the (full) ECA and non-ECA samples.


         To summarize our findings, we show that removing from the full ECA sample countries
with very high FOREIGN FUNDING or very high LDR makes the “truncated
ECA” sample similar to the rest of the world in terms of the response of PRIVATE CREDIT to
FOREIGN FUNDING shocks. This suggests that high FOREIGN FUNDING and high LDR are
indeed factors that are driving the observed differences between the (full) ECA and non-ECA
samples. In contrast, high FOREIGN OWNERSHIP is not driving these differences because
even when high FOREIGN OWNERSHIP countries are removed from the ECA sample and the
truncated ECA sample has lower average FOREIGN OWNERSHIP than the non-ECA sample,
there is still a significant difference in the response of PRIVATE CREDIT to FOREIGN
FUNDING shocks. These results provide more direct evidence that high FOREIGN FUNDING
and high LDR ratios were key factors in explaining the higher sensitivity of PRIVATE CREDIT
to FOREIGN FUNDING shocks in ECA, while high FOREIGN OWNERSHIP per se did not
account for this.

VI. Summary of Conclusions and Policy Implications

By applying PVARs to a global country panel database we show that bank credit growth to the
private sector is highly sensitive to cross-border funding shocks around the world. We find that
this relationship is significantly stronger in the average ECA country where the response is 72%
larger compared to the average country in the rest of the world. At the same time, we show that
foreign ownership per se does not explain the different credit responses in our sample of
countries. Instead, our results indicate stronger responses in countries with high loan-to-deposit
ratios and stronger reliance on foreign funding. Higher loan-to-deposit ratios make banks more



differences between responses of private credit to foreign liabilities between this very small sample and the non-
ECA sample.
                                                         28
vulnerable to general wholesale funding shocks while high reliance on foreign funding
specifically implies exposure to more volatile cross-border financing flows and sensitivity to
foreign shocks. Taken together, our findings therefore suggest that funding model differences
with the rest of the world were at the heart of ECA’s post-crisis credit growth contraction, and
that this contraction was not simply due to the high prevalence of foreign bank ownership in the
region.


Our findings provide yet another illustration of the potential downside risks of financial opening
or financial integration in the absence of adequate regulatory and supervisory frameworks. The
easy access to parent funding and direct wholesale borrowings abroad exposed many ECA
countries to substantial funding risks which materialized when cross-border flows were
interrupted by the global crisis and parent bank health deteriorated. The sudden slowdown of the
pace of funding contributed in turn to a sharp slowdown of credit and GDP growth, as well as an
accumulation of non-performing loans in many countries (Raiffeisen (2013) and Unicredit
(2013)). 11


As discussed, the Vienna Initiative prevented a severe withdrawal of parent funding and a more
dramatic slowdown of credit in the participating countries, but all the countries that were
overextended before the crisis have been adjusting their funding structures, both the countries
that relied more on parent funding and those that relied more on direct borrowings abroad. This
ongoing adjustment is reflected in the decline in loan-to-deposit ratios and the share of foreign
liabilities to total liabilities. This adjustment was necessary to curb the excess of the pre-crisis
period, and reflect both market and regulatory pressures. The central question being faced by
regulators in the region is the extent to which this adjustment will continue to take place.


Recent market research by one of the major international banks operating in the region suggests
that European banks will continue adjusting their loan-to-deposit ratios and levels of foreign
funding         and        may         eventually         move          towards         a       “Spanish”             or


11
   The credit boom exposed ECA countries to other risks as well, including credit, interest rate, and exchange rate
risks in their consumer, mortgage and SME portfolios.
                                                         29
“Santander” model of standalone subsidiaries which are locally funded and capitalized
(Raiffeisen (2013)). This business model was adopted from the outset by Spanish parent banks
operating in Latin America and seems to have been driven by pressures from the home
supervisor itself (Impavido, Rudolph and Ruggerone (2013) and CGFS (2010b)). This model has
minimized within group contagion and mitigated the transmission of the crisis on Latin
American banks, although it also possibly sacrifices efficiency of scale, funding, and liquidity,
and may lead to costly pockets of liquidity and capital (see, e.g. Fiechter et al (2011) and
Schoenmaker (2013)).


The “Spanish model” of standalone subsidiaries is a relevant benchmark for the ECA region, but
possibly an extreme solution for a region that shares a common market and a regulatory
framework (including the European passport for licensing) and that seeks to establish a banking
union in the future. However, in the absence of a common fiscal backstop as part of such a
union, standalone subsidiaries can reduce risks to host supervisors and tax payers. Recognizing
that the banking union is a long-term project, the question is whether there are regulatory and
supervisory approaches that would allow ECA countries to reap most of the benefits of financial
integration in the coming years, while mitigating the risks.


The full implementation of Basel III may contribute to the achievement of such an objective.
Basel III is expected to boost capital and liquidity and also promote more stable funding through
the Net Stable Funding ratio (NSFR). The Basel III approach also opens room for the
introduction of additional capital charges on banks with domestic and regional systemic
importance, such as Western European banks with a network of subsidiaries in the ECA region. 12
However, the extent to which Basel III will be used to address effectively the problems identified
in the crisis remains to be seen. More specifically, the scope for curbing excessive parent funding
through the NSFR needs to be further clarified and regulated, and the calibration of capital
surcharges to the risks generated by cross-border lending by banks with domestic and regional
systemic importance also remains to be tested.


12
  The BIS (2012) discusses a framework for dealing with domestic systemically important banks, including banks
with an extensive network of subsidiaries within a region.
                                                      30
Moreover, while Basel III may enable national regulators (home and/or host authorities) to
preserve financial stability while maintaining an element of cross-border finance, an additional
question is whether further and complementary measures could also contribute to the
achievement of this objective. In this regard, some national regulatory authorities already
introduced additional measures to induce a more rapid balance sheets adjustment after the crisis
without excessively curtailing foreign funding. For example, in 2012 Austrian regulators
introduced a cap of 110 percent on the loan-to-deposit ratios of all subsidiaries of Austrian parent
banks (see Austrian Financial Market Authority and Austrian National Bank (2012)). The ratio
provided some flexibility as it was applied on a flow basis and allowed for the inclusion of long-
term domestic market funding instruments (such as covered bonds) in the definition of deposits.
Moreover, the level of the cap was determined by the actual region-wide average loan-to-deposit
ratio of Austrian subsidiaries, implying that only the more over-extended countries had to adjust
their balance sheets. Also, some exceptions to the rule were granted to small subsidiaries.


The approach of Austrian regulators does imply some flexibility relative to the “Spanish model”,
although it was still met with reservations by many host regulators and parent banks when it was
introduced. While there was a general consensus on the need for adjustment, a cap on loan-to-
deposit ratios—even when applied on a flow or marginal basis—was considered by many
regulators as an excessively blunt instrument that undermined the singleness and openness of
European financial markets and financial integration. Also, concerns about the timeliness of the
measure were raised. Moreover, smaller countries indicated that they would face greater
difficulties, given their more limited scope to develop local and alternative sources of domestic
finance. 13


Therefore, home and host regulators in Europe still face the challenge of designing an effective
regulatory framework for cross-border banking in the coming years. Such a framework may
include supplements to Basel III to prevent the funding and credit growth excesses of the past


13
  The loan-to-deposit cap was introduced suddenly without consulting host supervisors. After protest of host
supervisors, the measure was transformed into a “monitoring measure”.
                                                        31
decade and the risk of another crisis, while allowing banking institutions to benefit from the
centralized management of capital, liquidity, and funding. There are a variety of measures that
can be taken by either home or host authorities, imposed at either the parent or subsidiary level.
Beyond national measures, European authorities may also consider if specific supervisory
guidance on cross-border funding is needed. In any case, meeting this challenge will be relevant
for ECA countries, especially smaller countries that face more constraints to develop local
capital markets, and that may continue depending relatively more on parent and other cross-
border bank funding.




                                               32
                                           References

Allen, F., T. Beck, E. Carletti, P. Lane, D. Schoenmaker, W. Wagner, 2011, “Cross Border
Banking in Europe: Implications for Financial Stability and Macroeconomic Policies”, Center for
Economic Policy Research, London, UK.

Austrian Financial Monetary Authority (AFMA) and Austrian National Bank (ANB),
2012, “Background note on the strengthening of the sustainability of the business models of
large internationally active Austrian banks”. Vienna.

Brown, M., S. Ongena, A. Popov, and P. Yeşin, 2011, “Who needs credit and who gets credit
in Eastern Europe?” Economic Policy, 26, 93-130

Choi, M., E. Gutierrez, and M. Martinez-Perias, 2013, "Dissecting foreign bank lending
behavior during the 2008-2009 crisis”. World Bank Policy Research Working Paper 6674
(October).

Claessens, S., and N. van Horen, 2012, “Foreign banks: Trends, Impact, and Financial
Stability”, IMF Working Paper, WP/12/10 (January). Forthcoming in the Journal of Money,
Credit and Banking.

Claessens, S., and N. van Horen, 2013, “Impact of Foreign Banks, The Journal of Financial
Perspectives, Volume 1, Issue 1.

Coeuré, B., 2013, “International financial integration and fragmentation: Drivers and policy
responses”. Conference organized by the Banco de España and the Reinventing Bretton Woods
Committee, Madrid (March).

Committee on the Global Financial System (CGFS), 2010a, “The functioning and resilience
of cross-border funding markets”, CGFS Papers, no 37.

______, 2010b, “Funding patterns and liquidity management of internationally active banks”,
CGFS Papers, no 39.

                                               33
______, 2010c, “Long-term issues in international banking”, CGFS Papers, no 41.

Cull, R., and M. S. Martinez Peria, 2012, “Bank ownership and lending patterns during the
2008-2009 financial crisis: evidence from Latin America and Eastern Europe,” World Bank
Policy Research Working Paper, No. 6195.

De Haas, R., and I. Van Lelyveld, 2012, “Multinational banks and the global financial crisis:
weathering the perfect storm,” Journal of Money, Credit and Banking, forthcoming.

De Haas, R., and N. van Horen, 2013, “Running for the exit? International bank lending during
a financial crisis,” Review of Financial Studies, forthcoming.

De Haas, R., Y. Korniyenko, E. Loukoianova, and A. Pivovarsky, 2012, “Foreign banks and
the Vienna initiative: turning sinners into saints?” IMF Working Paper WP/12/117 (April).

Feyen, E. and I. González del Mazo, 2013, “European bank deleveraging and global credit
conditions: Implications of a multi-year process on long-term finance and beyond”, World Bank
Policy Research Working Paper 6388 (March).

Fiechter, J., I. Ötker-Robe, A. Ilyina, M. Hsu, A. Santos, and J. Surti, 2011, “Subsidiaries or
Branches: Does One Size Fits All? IMF Staff Discussion Note SDN 11/04 (March).

Giannetti, M., and S. Ongena, 2012, “Lending by example: direct and indirect effects of
foreign banks in emerging markets,” Journal of International Economics, 86, 167-180.

Impavido, G., H. Rudolph, and L. Ruggerone, 2013, “Bank funding in Central, Eastern and
Southern Europe post Lehman: A new normal?” IMF Working Paper WP/13/148 (June).

International Monetary Fund, 2013, Global Financial Stability Report.

Kapan, T., and C. Minoiu, 2013, “Balance sheet strength and bank lending during the global
financial crisis”, IMF Working Paper WP/13/102.



                                                34
Love, Inessa, and Lea Zicchino, 2006, “Financial development and dynamic investment
behavior: Evidence from panel vector autoregression.” The Quarterly Review of Economics and
Finance 46, 190-210.

Martel, M., A van Rixtel, and E. Mota, 2012, “Business models of international banks in the
wake of the 2007-2009 global financial crisis”, Banco de España Revista de Estabilidad
Financiera, No. 22, pp. 19-121.

Mishkin, F. S., 2007, “Is financial globalization beneficial?” Journal of Money, Credit and
Banking, 39, 259-294

Ongena, S., J. L. Peydro, and N. van Horen, 2012, “Shocks abroad, pain at home? Bank-firm
level evidence on financial contagion during the 2007-2009 crisis,” Mimeo, Tilburg University,
Universitat Pompeu Fabra, De Nederlandsche Bank

Popov, A., and G. Udell, 2012, “Cross-border banking, credit access and the financial crisis,”
Journal of International Economics, 87, 147-161.

Raiffeisen, 2013, CEE Banking Sector Report (May), Raiffeisen Research, Vienna.

Rixtel, A. and G. Gasperini, 2013, “Financial crisis and bank funding: recent experience in the
euro area”, BIS Working Paper No. 406 (March).

Schoenmaker, D., 2013, Governance of International Banking: The Financial Trilemma.
Oxford University Press. New York.

Takáts, E., 2010, “Was it Credit Supply? Cross-border bank lending to emerging market
economies during the financial crisis”. BIS Quarterly Review, June.

Unicredit, 2013, Banking in CEE – Looking for Sound Growth, Unicredit Group, Vienna.

Visco, I. 2013, “The impact of the crisis on financial integration in Central and Eastern Europe”.
Bank of Italy. Speech made at a conference hosted by the National Bank of Slovakia: “Twenty
years of transition – experiences and challenges”.

                                                35
Figure 1 – Credit Trends in ECA and Other Countries

                  Annual credit growth                                            Annual deposit growth
              regional averages (yoy in %)                                     regional averages (yoy in %)
80                                                               60
70
                                                                 50
60
50                                                               40
40
                                                                 30
30
20                                                               20
10
                                                                 10
 0
-10                                                               0
      2003M1
      2003M7
      2004M1
      2004M7
      2005M1
      2005M7
      2006M1
      2006M7
      2007M1
      2007M7
      2008M1
      2008M7
      2009M1
      2009M7
      2010M1
      2010M7
      2011M1
      2011M7
      2012M1
      2012M7



                                                                      2003M1
                                                                      2003M7
                                                                      2004M1
                                                                      2004M7
                                                                      2005M1
                                                                      2005M7
                                                                      2006M1
                                                                      2006M7
                                                                      2007M1
                                                                      2007M7
                                                                      2008M1
                                                                      2008M7
                                                                      2009M1
                                                                      2009M7
                                                                      2010M1
                                                                      2010M7
                                                                      2011M1
                                                                      2011M7
                                                                      2012M1
                                                                      2012M7
      ECA - CESEE        ECA - CIS    Other Emerging Countries        ECA - CESEE      ECA - CIS      Other Emerging Countries


                      Loan to deposit ratio,                               Foreign liabilities in total liabilities,
                    regional averages (in %)                                    regional averages (in %)
200                                                              50

180                                                              45

                                                                 40
160
                                                                 35
140
                                                                 30
120
                                                                 25
100
                                                                 20
 80                                                              15

 60                                                              10
      2002M1
      2002M7
      2003M1
      2003M7
      2004M1
      2004M7
      2005M1
      2005M7
      2006M1
      2006M7
      2007M1
      2007M7
      2008M1
      2008M7
      2009M1
      2009M7
      2010M1
      2010M7
      2011M1
      2011M7
      2012M1
      2012M7



                                                                      2002M1
                                                                      2002M7
                                                                      2003M1
                                                                      2003M7
                                                                      2004M1
                                                                      2004M7
                                                                      2005M1
                                                                      2005M7
                                                                      2006M1
                                                                      2006M7
                                                                      2007M1
                                                                      2007M7
                                                                      2008M1
                                                                      2008M7
                                                                      2009M1
                                                                      2009M7
                                                                      2010M1
                                                                      2010M7
                                                                      2011M1
                                                                      2011M7
                                                                      2012M1
                                                                      2012M7
      ECA - CESEE        ECA - CIS    Other Emerging Countries        ECA - CESEE      ECA - CIS      Other Emerging Countries

Source: IFS




                                                             36
Figure 2 - Baseline Panel Vector Autoregression Model

The figure displays impulse-response functions of the baseline 1-lag PVAR model which is
based on 4 variables: FOREIGN LIABILITIES (foreign), GDP (gdp), DEPOSITS (deposits), and
PRIVATE CREDIT (priv cred).

Impulse-responses for 1 lag VAR of foreign gdp deposits privcred
               (p 5) foreign        foreign                      (p 5) gdp        gdp                        (p 5) deposits        deposits                     (p 5) privcred        privcred
               (p 95) foreign                                    (p 95) gdp                                  (p 95) deposits                                    (p 95) privcred
 13.0499                                          0.7748                                       0.9718                                             2.6618




  0.0000                                          -0.2578                                      -0.0958                                            0.0000
           0                                  6              0                             6             0                                    6             0                                    6
                                s                                             s                                                s                                                  s
    response of foreign to foreign shock              response of foreign to gdp shock          response of foreign to deposits shock              response of foreign to privcred shock
               (p 5) foreign        foreign                      (p 5) gdp        gdp                        (p 5) deposits        deposits                     (p 5) privcred        privcred
               (p 95) foreign                                    (p 95) gdp                                  (p 95) deposits                                    (p 95) privcred
  0.3698                                          3.5696                                       0.3324                                             0.1805




 -0.0386                                          -0.7137                                      -0.0598                                            -0.1227
           0                                  6              0                             6             0                                    6             0                                    6
                                s                                             s                                                s                                                  s
      response of gdp to foreign shock                      response of gdp to gdp shock          response of gdp to deposits shock                   response of gdp to privcred shock
               (p 5) foreign        foreign                      (p 5) gdp        gdp                        (p 5) deposits        deposits                     (p 5) privcred        privcred
               (p 95) foreign                                    (p 95) gdp                                  (p 95) deposits                                    (p 95) privcred
  0.9000                                          0.8376                                       4.1138                                             0.6323




 -0.1382                                          -0.0517                                      -0.2645                                            0.0000
           0                                  6              0                             6             0                                    6             0                                    6
                                s                                             s                                                s                                                  s
   response of deposits to foreign shock             response of deposits to gdp shock         response of deposits to deposits shock             response of deposits to privcred shock
               (p 5) foreign        foreign                      (p 5) gdp        gdp                        (p 5) deposits        deposits                     (p 5) privcred        privcred
               (p 95) foreign                                    (p 95) gdp                                  (p 95) deposits                                    (p 95) privcred
  1.4147                                          0.7420                                       1.6568                                             3.6137




  0.0000                                          0.0000                                       0.0000                                             0.0000
           0                                  6              0                             6             0                                    6             0                                    6
                                s                                             s                                                s                                                  s
   response of privcred to foreign shock              response of privcred to gdp shock        response of privcred to deposits shock              response of privcred to privcred shock

                                    Errors are 5% on each side generated by Monte-Carlo with 200 reps




                                                                                                             37
Figure 3 - ECA vs. Non-ECA Sample Split

The figure shows impulse-response functions of PRIVATE CREDIT due to various shocks. The
first and second rows provide the functions for the ECA and non-ECA subsamples, respectively.
The third row provides the difference functions between these subsamples. The shocks are as
follows: “foreign” is FOREIGN LIABILITIES, “gdp” is GDP, “deposits” is DEPOSITS, and
“privcred” denotes PRIVATE CREDIT itself.


                                                   Split by Region: ECA vs. non ECA
                                                                          Sample: ECA
   1.7141




                                         0.9046




                                                                                   1.8292




                                                                                                                          3.8968
                                         -0.0383
   0.0014




                                                                                   0.0015




                                                                                                                          0.0034
             0      2       4       6              0         2       4         6             0       2       4      6              0       2       4      6
                        s                                        s                                       s                                     s
             response to foreign shock                 response to gdp shock                 response to deposits shock            response to privcred shock

                                                                         Sample: nonECA
   1.0991




                                         0.6474




                                                                                   1.6221




                                                                                                                          3.2902
   -0.0386




                                         0.0000




                                                                                   0.0001




                                                                                                                          0.0002
             0      2       4       6              0         2       4         6             0       2       4      6              0       2       4      6
                        s                                        s                                       s                                     s
             response to foreign shock                 response to gdp shock                 response to deposits shock            response to privcred shock

                                                                     Sample: Difference
   0.9092




                                         0.4762




                                                                                   0.7759




                                                                                                                          0.9332
                                         -0.5024




                                                                                   -0.0956
   0.0006




                                                                                                                          0.0004




             0      2       4       6              0         2       4         6             0       2       4      6              0       2       4      6
                        s                                        s                                       s                                     s
             response to foreign shock                 response to gdp shock                 response to deposits shock            response to privcred shock




                                                                                   38
Figure 4- Sample Split by High and Low Loan-to-Deposit Ratios

The figure shows impulse-response functions of PRIVATE CREDIT due to various shocks. The
first and second rows provide the functions for the high LDR and low LDR subsamples,
respectively. The split is based on the median value (see text for details). The third row provides
the difference functions between these subsamples. The shocks are as follows: “foreign” is
FOREIGN LIABILITIES, “gdp” is GDP, “deposits” is DEPOSITS, and “privcred” denotes
PRIVATE CREDIT itself.

                                                                         Split by LDR
                                                                          Sample: High
   1.8003




                                         0.7093




                                                                                   1.8362




                                                                                                                          3.7621
                                         -0.0657
   0.0012




                                                                                   0.0010




                                                                                                                          0.0030
             0      2       4       6              0         2       4         6             0       2        4     6               0       2       4      6
                        s                                        s                                       s                                      s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock

                                                                          Sample: Low
   1.0837




                                         0.9716




                                                                                   1.5592




                                                                                                                          3.4692
   0.0000




                                         0.0001




                                                                                   0.0001




                                                                                                                          0.0002
             0      2       4       6              0         2       4         6             0       2        4     6               0       2       4      6
                        s                                        s                                       s                                      s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock

                                                                     Sample: Difference
   1.0253




                                         0.2321




                                                                                   0.5899




                                                                                                                          0.9152
   -0.1082




                                         -0.5235




                                                                                   -0.2453




                                                                                                                          -0.0050




             0      2       4       6              0         2       4         6             0       2        4     6               0       2       4      6
                        s                                        s                                       s                                      s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock




                                                                                                         39
Figure 5 - Sample Split by High and Low Foreign Funding Dependence

The figure shows impulse-response functions of PRIVATE CREDIT due to various shocks. The
first and second rows provide the functions for the high FOREIGN FUNDING and low
FOREIGN FUNDING subsamples, respectively. The split is based on the median value (see text
for details). The third row provides the difference functions between these subsamples. The
shocks are as follows: “foreign” is FOREIGN LIABILITIES, “gdp” is GDP, “deposits” is
DEPOSITS, and “privcred” denotes PRIVATE CREDIT itself.

                                                                 Split by Foreign Share
                                                                          Sample: High
   1.8338




                                         0.8894




                                                                                   1.9543




                                                                                                                          4.0131
                                         -0.1518
   0.0010




                                                                                   0.0013




                                                                                                                          0.0027
             0      2       4       6              0         2        4        6             0       2       4      6               0       2       4      6
                        s                                         s                                      s                                      s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock

                                                                          Sample: Low
   1.1422




                                         0.7132




                                                                                   1.5497




                                                                                                                          3.2686
   0.0001




                                         0.0002




                                                                                   0.0002




                                                                                                                          0.0005
             0      2       4       6              0         2        4        6             0       2       4      6               0       2       4      6
                        s                                         s                                      s                                      s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock

                                                                      Sample: Difference
   0.9120




                                         0.4017




                                                                                   0.8195




                                                                                                                          0.8941
   -0.0062




                                         -0.6834




                                                                                   -0.0005




                                                                                                                          -0.0223




             0      2       4       6              0         2        4        6             0       2       4      6               0       2       4      6
                        s                                         s                                      s                                      s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock




                                                                                              40
Figure 6 - Sample Split by High and Low Foreign Ownership

The figure shows impulse-response functions of PRIVATE CREDIT due to various shocks. The
first and second rows provide the functions for the high FOREIGN OWNERSHIP and low
FOREIGN OWNERSHIP subsamples, respectively. The split is based on the median value (see
text for details). The third row provides the difference functions between these subsamples. The
shocks are as follows: “foreign” is FOREIGN LIABILITIES, “gdp” is GDP, “deposits” is
DEPOSITS, and “privcred” denotes PRIVATE CREDIT itself.

                                                           Split by Foreign ownership
                                                                         Sample: High
   1.4061




                                         0.9503




                                                                                   1.7090




                                                                                                                          3.6034
                                         -0.3446
   0.0006




                                                                                   0.0006




                                                                                                                          0.0017
             0      2       4       6              0         2       4         6             0       2        4     6               0       2       4      6
                        s                                        s                                        s                                     s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock

                                                                         Sample: Low
   1.4548




                                         0.8628




                                                                                   1.6842




                                                                                                                          3.7035
   0.0003




                                         0.0006




                                                                                   0.0004




             0      2       4       6              0         2       4         6             0       2        4     6     0.0013    0       2       4      6
                        s                                        s                                        s                                     s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock

                                                                     Sample: Difference
   0.6124




                                         0.5788




                                                                                   0.5498




                                                                                                                          0.2933
   -0.3223




                                         -1.0378




                                                                                   -0.2392




                                                                                                                          -0.5414




             0      2       4       6              0         2       4         6             0       2        4     6               0       2       4      6
                        s                                        s                                        s                                     s
             response to foreign shock                 response to gdp shock                 response to deposits shock             response to privcred shock




                                                                                                         41
Figure 7 - Truncated ECA sample vs. non-ECA

The graphs show differences in responses of PRIVATE CREDIT to FOREIGN LIABILITIES in
truncated ECA samples vs. the rest of the sample. Each truncated ECA sample is constructed by
removing, one at a time, from the ECA sample countries above the median for each of the three
characteristics: FOREIGN OWNERSHIP, FOREIGN FUNDING and LDR. The graph in
column 2, row 2 removes an additional 4 countries which have FOREIGN OWNERSHIP above
51% (see text for details).

ECA   without   high   FOREIGN                           ECA without high LDR versus non-ECA
FUNDING versus non-ECA
0.80                                                     1.00
                                   5th                                                    5th
                                   percentile            0.80                             percentile
0.60                                                                                      impulse
                                   impulse
                                                         0.60
0.40
                                                         0.40
0.20
                                                         0.20
0.00
                                                         0.00
                                                                 0   1   2     3      4       5        6
-0.20                                                    -0.20

-0.40                                                    -0.40
        0    1    2     3      4        5       6                            Period
                      Period


ECA without high FOREIGN OWNERSHIP                       ECA without high FOREIGN
(above the median) versus non-ECA                        OWNERSHIP (> 51%) versus non-ECA
1.60                                                     2.00
                                   5th                                                5th
1.40                               percentile                                         percentile
                                   impulse               1.50                         impulse
1.20

1.00                                                     1.00
0.80

0.60                                                     0.50

0.40
                                                         0.00
0.20                                                             0   1   2     3      4       5        6

0.00                                                     -0.50
        0   1    2      3      4       5        6                            Period
                      Period




                                                    42
Table 1 - Variable Definitions and Sources

Variable
               Full Name          Definition                           Source
Name
                                                                        IMF, International Financial
                                                                        Statistics. For Albania,
                                  Gross Domestic Product, Nominal.
                                                                        Colombia, Sri Lanka, Egypt,
GDP            GDP                Quarterly growth rates from figures
                                                                        Tunisia and Nigeria, the
                                  in national currency (in percentage).
                                                                        sources are respective
                                                                        central bank websites.
                                                                        IMF, International Financial
                                  Other Depository Corporations,        Statistics. For Bulgaria,
                                  Line 22d (Claims on Private           Czech Republic, Hungary,
PRIVATE
               Private Credit     Sector). Quarterly growth rates       Latvia, Lithuania and
CREDIT
                                  from figures in national currency     Poland, the sources are
                                  (in percentage).                      respective central bank
                                                                        websites.
                                  Other Depository Corporations,
                                  Line 26c (Liabilities to Non-
FOREIGN        Foreign                                                  IMF, International Financial
                                  residents). Quarterly growth rates
LIABILITIES    Liabilities                                              Statistics
                                  from figures in national currency
                                  (in percentage).
                                  Other Depository Corporations,
                                  Line 24 (Transferable Deposits
                                  Included In Broad Money) + line 25
                                                                        IMF, International Financial
DEPOSITS       Deposits           (Other Deposits Included In Broad
                                                                        Statistics
                                  Money). Quarterly growth rates
                                  from figures in national currency
                                  (in percentage).
                                  Share of banking system's assets
          Foreign                                                       World Bank, Bank
FOREIGN                           where foreigners own 50% or more
          Ownership of                                                  Regulation and Supervision
OWNERSHIP                         equity (in percentage). End-of-year
          Banks                                                         Survey
                                  figures.
               Loan-to-deposit                                          IMF, International Financial
LDR                               Private Credit/Deposits*100.
               ratio                                                    Statistics
                                                                        Authors calculation based on
FOREIGN        Share of foreign   Foreign liabilities/(foreign
                                                                        the original data from above
FUNDING        liabilities        liabilities + Deposits)*100.
                                                                        sources




                                               43
Table 2A - Summary Statistics for Quarterly PVAR Variables

All variables are expressed as quarterly real growth rates (%).

i) Full sample

                FOREIGN
                                  PRIVATE CREDIT                  GDP      DEPOSITS
               LIABILITIES
N                  1,767                  1,767                    1,767     1,767
mean                3.86                   3.23                    -0.01      2.62
Sd                 13.66                   5.12                     3.53      4.67
p1                -28.10                 -10.72                   -11.05     -9.80
p25                -4.42                   0.11                    -1.55     -0.10
p50                 2.36                   2.66                     0.07      2.08
p75                 9.97                   6.00                     1.71      4.85
p99                52.40                  19.42                     9.21     16.70

ii) ECA

                FOREIGN
                                  PRIVATE CREDIT                  GDP      DEPOSITS
               LIABILITIES
N                   947                    947                      947       947
mean                5.75                   4.55                    -0.05      3.41
Sd                 14.07                   5.78                     4.18      5.42
p1                -29.13                 -10.79                   -13.67    -10.61
p25                -2.99                   0.61                    -1.82     -0.07
p50                 4.37                   4.56                     0.06      2.82
p75                12.43                   8.18                     2.05      6.59
p99                55.20                  20.31                    10.55     20.50

iii) Non-ECA

                FOREIGN
                                  PRIVATE CREDIT                  GDP      DEPOSITS
               LIABILITIES
N                   820                   820                      820        820
mean                1.68                  1.71                     0.03       1.70
Sd                 12.84                  3.70                     2.60       3.40
p1                -27.09                 -9.55                    -7.31      -7.26
p25                -5.72                 -0.33                    -1.28      -0.18
p50                -0.13                  1.61                     0.08       1.67
p75                 7.12                  3.60                     1.40       3.56
p99                43.67                 11.36                     6.48      10.30



                                                  44
Table 2B – Pre-Crisis Summary Statistics for Split Variables

All variables are expressed as percentages (%) and reflect 2007:Q4 values for LDR and
FOREIGN FUNDING and the closest available observation for FOREIGN OWNERSHIP,
which is 2008 for most countries.

i) Full sample

                                   FOREIGN             FOREIGN
                 LDR
                                   FUNDING            OWNERSHIP
N                  41                 41                  41
mean             115.10              0.22                50.02
p50               99.58              0.17                 51
Sd                48.96              0.16                31.47

ii) ECA

                                   FOREIGN             FOREIGN
                 LDR
                                   FUNDING            OWNERSHIP
N                  22                 22                  22
mean             127.22              0.32                64.64
p50               115.5              0.28                79.61
Sd                47.30              0.16                31.48

iii) Non-ECA

                                   FOREIGN             FOREIGN
                 LDR
                                   FUNDING            OWNERSHIP
N                  19                 19                  19
mean             101.07              0.11                33.09
p50               83.53              0.10                29.93
Sd                48.27              0.06                21.94




                                            45
Table 3 - Correlations

Panel A. PVAR Variables

                                    FOREIGN                                       PRIVATE
                                   LIABILITIES       GDP               DEPOSITS    CREDIT
                                             Full Sample
 FOREIGN LIABILITIES                    1
 GDP                                  0.05             1
 DEPOSITS                             0.24           0.18                  1
 PRIVATE CREDIT                       0.40           0.16                0.51          1
                                                 ECA
 FOREIGN LIABILITIES                    1
 GDP                                  0.05             1
 DEPOSITS                             0.30           0.20                  1
 PRIVATE CREDIT                       0.50           0.18                0.54          1
                                               non ECA
 FOREIGN LIABILITIES                    1
 GDP                                  0.06             1
 DEPOSITS                             0.13           0.11                  1
 PRIVATE CREDIT                       0.23           0.11                0.42          1
Bold numbers indicate statistical significance at 5%.

Panel B. Sample Split Variables

                                        FOREIGN               FOREIGN
                                       OWNERSHIP              FUNDING             LDR
                                                Full Sample
 FOREIGN OWNERSHIP                          1
 FOREIGN FUNDING                           0.35                  1
 LDR                                       0.08                 0.54               1
                                                   ECA
 FOREIGN OWNERSHIP                          1
 FOREIGN FUNDING                          -0.06                  1
 LDR                                      -0.12                 0.96               1
                                                 non ECA
 FOREIGN OWNERSHIP                          1
 FOREIGN FUNDING                           0.30                  1
 LDR                                      -0.11                -0.02               1
Bold numbers indicate statistical significance at 5%.




                                                        46
Table 4 - Mean Tests for ECA vs. Non-ECA

Reported are t-tests for differences in means between ECA and non-ECA countries. Variables
are quarterly growth rates (%). Significance is indicated by: ***, **, *, NS represent 1%, 5%,
10%, and above 10% respectively.

Panel A: Key Differences for PVAR Variables (growth rates in %)

                              Sample: Non-ECA     Sample: ECA
                             N(1) Mean(1)       N(2) Mean(2)           Significance
                                             Years 2005-2007
FOREIGN LIABILITIES          300       3.64      286     10.06         ***
GDP                          239       0.51      263      0.95         **
DEPOSITS                     312       2.65      286      5.42         ***
PRIVATE CREDIT               312       3.15      286      7.87         ***
                                                 Years 2008-2009
FOREIGN LIABILITIES          214        0.74         192      2.33     NS
GDP                          160        -0.51        176     -1.86     ***
DEPOSITS                     222        1.59         192      1.34     NS
PRIVATE CREDIT               222        1.45         192      2.15     *

                                                 Years 2010-2011
FOREIGN LIABILITIES          214        3.62         192     -0.02     ***
GDP                          159        0.09         176     -0.11     NS
DEPOSITS                     222        1.81         192      1.59     NS
PRIVATE CREDIT               222        2.19         192      0.67     ***

Panel B: Key Differences for Sample Split Variables

Variables are measured at pre-crisis levels: 2007 Q4 for LDR and FOREIGN FUNDING and
2008 for FOREIGN OWNERSHIP.

                            Sample: Non-ECA            Sample: ECA
Variable                    N(1)    Mean(1)          N(2) Mean(2)    Significance
LDR                          19      101%             24     127%    **
FOREIGN FUNDING              19       11%             24      32%    ***
FOREIGN OWNERSHIP            19       33%             24      65%    ***




                                                47
Panel C: Key Differences for Sample Split Variables for truncated ECA sample

Variables are measured at pre-crisis levels: 2007 Q4 for LDR and Foreign funding and 2008 for
foreign ownership. Truncated ECA sample is constructed by removing from ECA sample
countries above the median for each of the indicator one at a time. FOREIGN OWNERHSIP 2
sample is truncated further to remove 4 more countries with high foreign ownership: (Serbia,
75%, Latvia, 68%, Poland 68% and Armenia 60%), in effect it removes all countries with
foreign ownership above 51%.

                                                      Sample:
                             Sample: Non-ECA       truncated ECA
Variable                     N(1)    Mean(1)      N(2) Mean(2)      Significance
LDR                           19      101%         11       92%     NS
FOREIGN FUNDING               19       11%         11       20%     ***
FOREIGN OWNERSHIP             19       33%         11       39%     NS
FOREIGN OWNERSHIP 2           19       33%          7       23%     NS




                                             48
Table 5 - GMM Estimates of the Reduced-Form PVAR Models

This table reports system estimates in which each of the dependent variables is regressed on the
lags of the other variables in the system, estimated by system GMM in which the original
variables are transformed using forward orthogonal deviations. The instruments are lags of the
untransformed variables as described in text. All variables enter as real quarterly growth rates.

                                                     Dependent Variable:
                    FOREIGN                                                            PRIVATE
                  LIABILITIES                   GDP               DEPOSITS              CREDIT
                Coef    T-stat         Coef       T-stat       Coef   T-stat       Coef    T-stat
                                              Full Sample
L.FOREIGN          0.01       0.33       0.02        2.01        0.00      -0.33     0.00       0.72
L.GDP             -0.01      -0.10      -0.14       -3.28        0.05       1.38     0.03       0.78
L.DEPOSITS        -0.14      -1.44       0.04        1.41       -0.05      -1.39     0.01       0.23
L.CREDIT           0.61       5.95       0.01        0.49        0.12       3.44     0.33       8.89
                                                  ECA
L.FOREIGN          0.00       0.01       0.02       1.90         0.00      -0.24     0.01        0.89
L.GDP             -0.02      -0.13      -0.15      -2.61         0.03       0.64    -0.00       -0.06
L.DEPOSITS        -0.18      -1.54       0.03       0.76        -0.04      -0.77     0.02        0.40
L.CREDIT           0.69       4.83      -0.01      -0.38         0.09       1.74     0.36        7.10
                                                Non ECA
L.FOREIGN          0.02       0.44       0.00          0.63      0.00      -0.12     0.00       -0.08
L.GDP              0.00       0.00      -0.12         -2.56      0.11       2.41     0.10        2.03
L.DEPOSITS        -0.05      -0.28       0.07          2.52     -0.09      -1.66    -0.01       -0.18
L.CREDIT           0.47       3.30       0.05          1.87      0.19       4.14     0.27        4.93




                                                49
Table 6 - Impulse-response Functions for Private Credit Growth

Each cell shows a response of private credit to a shock in a column variable in a period specified
in the Period column. The same responses are plotted in the middle line in the last row of graphs
in Figure 2 for the full sample and Figure 3 for ECA vs. non-ECA sample split.

             FOREIGN                                                    PRIVATE
Period      LIABILITIES             GDP             DEPOSITS             CREDIT
                                     Full Sample
       0         1.24               0.59                1.53               3.53
       1         0.51               0.29                0.54               1.16
       2         0.18               0.09                0.18               0.40
       3         0.06               0.03                0.06               0.14
       4         0.02               0.01                0.02               0.05
       5         0.01               0.00                0.01               0.02
       6         0.00               0.00                0.00               0.01

                                        ECA
       0         1.53               0.68                1.63               3.76
       1         0.74               0.25                0.68               1.35
       2         0.28               0.10                0.24               0.53
       3         0.11               0.04                0.10               0.20
       4         0.04               0.01                0.04               0.08
       5         0.02               0.01                0.01               0.03
       6         0.01               0.00                0.01               0.01

                                      Non-ECA
       0         0.89               0.49                1.45               3.16
       1         0.23               0.37                0.35               0.84
       2         0.07               0.07                0.12               0.23
       3         0.02               0.03                0.03               0.07
       4         0.01               0.00                0.01               0.08
       5         0.00               0.00                0.00               0.01
       6         0.00               0.00                0.00               0.00




                                               50
Table 7. Variance Decompositions

Panel A. Baseline PVAR Model

Each cell shows the percent of variation in the row variable explained by a shock of the column
variable after 10 quarterly periods.

                              FOREIGN                                            PRIVATE
                             LIABILITIES            GDP         DEPOSITS          CREDIT
                                       Baseline Model
FOREIGN LIABILITIES                  96.7%            0.0%             0.1%             3.1%
GDP                                   0.5%          99.2%              0.2%             0.0%
DEPOSITS                              3.2%            3.0%            92.6%             1.2%
PRIVATE CREDIT                        9.7%            2.4%            14.2%            73.8%

Panel B. Variance Decomposition for Private Credit Growth in Various Sample Splits

Each cell reports the percent of variance in PRIVATE CREDIT explained after 10 quarters by a
shock to the column variable.

                                              FOREIGN                                PRIVATE
                                   Obs       LIABILITIES      GDP      DEPOSITS       CREDIT

Full Sample                         1,767          9.7%       2.4%       14.2%        73.8%

ECA                                  947           12.9%      2.3%       13.9%        70.9%
Non-ECA                              820            6.0%      2.6%       15.7%        75.6%

High LDR                             895           14.1%      1.6%       13.9%        70.5%
Low LDR                              872            5.6%      4.8%       13.2%        76.3%

High FOREIGN FUNDING                 799           12.9%      1.9%       14.7%        70.4%
Low FOREIGN FUNDING                  968            6.9%      3.8%       13.8%        75.5%

High FOREIGN OWNERSHIP               886           10.0%      3.0%       15.2%        71.8%
Low FOREIGN OWNERSHIP                881            9.1%      3.0%       12.8%        75.2%




                                              51
                                   Table A1: Data availability by Country
Region             Country         GDP           privcred         foreign         deposits         LDR
         Albania               2005Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Armenia               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Azerbaijan            2001Q2-2011Q4   2001Q2-2011Q4   2001Q2-2011Q4   2001Q2-2011Q4   2001Q2-2011Q4
         Belarus               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Bulgaria              2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Croatia               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Czech Republic        2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Estonia               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Georgia               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Hungary               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Kazakhstan            2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
ECA
         Latvia                2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Lithuania             2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Macedonia, FYR        2003Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Poland                2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Romania               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Russian Federation    2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Serbia, Republic of   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Slovak Republic       2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Slovenia              2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Turkey                2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Ukraine               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Argentina             2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Brazil                2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
LAC
         Chile                 2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
         Colombia              2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4

                                                     52
           Mexico                  2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Peru                    2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Uruguay                 2005Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           China, P.R.: Mainland   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Indonesia               2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Korea, Republic of      2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
 ASIA      Malaysia                2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Philippines             2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Sri Lanka               2002Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Thailand                2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Egypt                   2001Q4-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Jordan                  2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
MENA       Morocco                 2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
& Africa   Qatar                   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           Tunisia                 2000Q2-2011Q3   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4
           South Africa            2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4   2000Q2-2011Q4




                                                         53
                  Table A2. Country Averages for Sample Split Indicators
                             FOREIGN                                       FOREIGN
Country                     OWNERSHIP                 LDR                  FUNDING

                                      ECA Countries
Albania                         94%                    48%                    7%
Armenia                         60%                   117%                   29%
Azerbaijan                       9%                   130%                   27%
Belarus                         22%                   115%                   21%
Bulgaria                        84%                   112%                   26%
Croatia                         91%                    99%                   25%
Czech Republic                  85%                    84%                   15%
Estonia                         99%                   206%                   55%
Georgia                         91%                   160%                   41%
Hungary                         85%                   143%                   35%
Kazakhstan                      13%                   198%                   59%
Latvia                          68%                   246%                   70%
Lithuania                       86%                   166%                   48%
Macedonia, FYR                  93%                    83%                   13%
Poland                          67%                    94%                   14%
Romania                         88%                   114%                   37%
Russia                          19%                   120%                   28%
Serbia                          75%                    98%                   27%
Slovak Republic                 96%                    93%                   26%
Slovenia                        31%                   152%                   48%
Turkey                          17%                    72%                   16%
Ukraine                         51%                   150%                   34%

Average                         65%                   127%                   32%

                                   Non ECA Countries
Argentina                       31%                   62%                    10%
Brazil                          21%                   72%                     8%
Chile                           41%                  137%                     9%
China                            2%                  234%                     5%
Colombia                        22%                  174%                    10%
Egypt                           31%                   55%                     4%
Indonesia                       33%                   68%                     6%
Jordan                          51%                   84%                    27%
Korea                           77%                  168%                    18%
Malaysia                        22%                   91%                    10%
Mexico                          84%                   81%                     9%
Morocco                         21%                   67%                     2%
                                           54
Peru           51%         71%   13%
Philippines    12%         53%   11%
South Africa   30%        126%   15%
Sri Lanka      15%         95%   15%
Thailand        6%        100%    4%
Tunisia        28%        118%   17%
Uruguay        52%         65%   20%

Average        33%        101%   11%




                     55
                                             Annex 1
                Technical Details on Panel Vector Autoregressions (PVARs)


Correlation problem between fixed effects and regressors
The advantage of PVAR is the same as the advantage of any panel approach – i.e. in allowing for
explicit inclusion of fixed effects in the model, denoted fi in Equation (1), which captures all
unobservable time-invariant factors at the country level. However, fixed effects present an
estimation challenge which arises in any model which includes lags of the dependent variables:
the fixed effects are correlated with the regressors and therefore the mean-differencing procedure
commonly used to eliminate fixed effects would create biased coefficients.

To avoid this problem we use forward mean-differencing, also referred to as the `Helmert
procedure' (see Arellano and Bover, 1995). This procedure removes only the forward mean, i.e.
the mean of all the future observations available for each firm-year. This transformation
preserves the orthogonality between transformed variables and lagged regressors, which allows
us to use lagged regressors as instruments and estimate the coefficients by system Generalized
Method of Moments (GMM). 14

The time-fixed effects are implemented by time-differencing all the variables prior to inclusion
in the model, which is equivalent to putting time dummies in the system.


Recursive ordering
To isolate the impact of a shock to a single variable while computing impulse-response
functions, we need to account for the fact that the actual variance-covariance matrix of the errors
is unlikely to be diagonal. In other words, they are correlated which inhibits attribution of the
impact of the shock of a variable to itself. A common solution is to adopt a particular ordering



14
   In our case the model is “just identified” because the number of regressors equals the number of
instruments, therefore system GMM is mathematically equivalent to equation-by-equation 2SLS. Also,
there are no overidentifying restrictions because the number of instruments is equal to the
number of variables in the model.

                                                56
and allocate any correlation between the residuals of any two variables to the variable that comes
first in the ordering. This procedure is known as a Choleski decomposition of the variance-
covariance matrix of residuals and is equivalent to transforming the system into a “recursive''
VAR for identification purposes (see Hamilton (1994) for the derivations and discussion of
impulse-response functions).


Generating confidence intervals of impulse-response functions
To analyze the impulse-response functions we need an estimate of the confidence intervals of the
point estimates. Since the various impulse-response functions are constructed from the estimated
VAR coefficients, their standard errors need to be taken into account. We calculate confidence
intervals with Monte Carlo simulations. To do so, we randomly generate a draw of coefficients
Γ0 and Γ1 from Equation (1) using the model estimations and the variance-covariance matrix and
re-calculate the impulse-responses. We repeat this procedure 200 times and then generate the 5th
and 95th percentiles of this distribution which we use as confidence intervals.


Impulse–response functions of differences between two subsamples

Because the two subsamples are independent, the impulse-responses of the differences are equal
to the difference in impulse-responses. To calculate the confidence interval, we merge the
distributions of errors via Monte-Carlo simulations with 200 repetitions each for both samples
and generate a new distribution which is the difference between errors generated in each of the
repetitions (i.e. each of the 200 errors now contains the difference between the distributions).
From this new distribution we generate new 5th and 95th percentile bounds.




                                                57