Syndicated Loans and their Contribution to Systemic Risk of European Banks

(1)

Syndicated Loans and their Contribution to

Systemic Risk of European Banks

Master Thesis

Amsterdam Business School

MSc Finance

Track Corporate Finance

Submitted by: Tzarevska, Maria

Thesis supervisor: Dr. Tolga Caskurlu

(2)

Statement of Originality

This document is written by student Maria Tzarevska who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

Syndicated loans increase portfolio overlap of European banks on an industry level and lead to spillover effects and an overall increase in systemic risk in Europe. I estimate the similarity in banks’ portfolios through their investments in syndicated loans in a particular borrower’s industry or region. Additionally, I include controls for the size of the financial institutions. Results confirm the hypothesis that large banks are the main contributors to systemic risk: Systemic risk is increased when interconnections between banks measured by their portfolio similarity on an industry level are stronger. The effect is mainly pronounced during times of overall economic stagnation.

(4)

1. Introduction

In wake of the financial crisis of 2007 important questions such as contagion effects leading to risk spillovers and their implications for policy and regulators need to be considered. One of the main causes for the emergence of the Great Recession were structured products and the collapse of the interbank lending market. Nevertheless, there is another crucial point to consider regarding the expansion of the crisis which initially started in the US and was transmitted to European markets. This is the first research to analyse the interconnections between banks through loan syndications and their implication for systemic risk in Europe. It is also the first to incorporate stress tests into a simple model to estimate non-diversifiable risk in an economy-wide context.

There are several channels through which systemic risk propagates through the whole economy. One of them is directly through the interbank lending market (Allen and Babus (2009)). Allen and Babus (2009) describe a direct network between financial institutions in the interbank market and the resulting exposures. They argue that networks between individual financial institutions are important for regulators and may be particularly useful for estimating the overall risk in an economy. Hence, analysis of financial networks might be reasonable for designing macro prudential policies to supervise individual financial institutions in order to increase the overall stability in an economy.

A second channel through which systemic risk spreads through financial systems is the similarity resulting from common asset holdings in the balance sheet of financial institutions. Hence, overlapping asset portfolios may induce risks and a systemic propagation of shocks when a common drop in prices devalues asset holdings of different banks leading to fire sales (Caccioli et al. (2014)). One asset on a bank’s balance sheet which represents a possible source of contagion is the syndicated loan. A loan syndication represents a mutual contract by two or more banks which engage in underwriting a loan to a single borrower together. The market for syndicated loans evolved in the 70s, but experienced its highest growth in the 90s. The syndicated loan market in the Euro area has been growing rapidly since the 2000s, reaching a peak of approximately EUR 1,200 billion in 2011. The number of contracts has grown from 600 in the year 2009

(6)

to more than 800 in 2011 (ECB, Syndicated loan market, (2012)). Nevertheless, Popov and Horen (2015) provide evidence for a substantial decrease in syndicated lending during the global financial crisis with a recovery in 2011 which was less pronounced for European banks. This leads to the conclusion that there may have been some negative externalities related to the decrease in European syndicated loans.

There have been various reasons for the substantial drop in lending during the financial crisis. This research argues that syndicated loans might have had a contribution in the increase of overall risk in the economy during the financial crisis of 2008. It is the first to estimate the relationship between syndications in the European market and their impact on systemic risk. Systemic risk is defined as the possibility of collapse of a whole economy due to an individual bank’s default. In the Euro area the most wide-spread approach to measuring systemic risk is through stress tests. This method exposes banks to hypothetical shocks while the impact on particular balance sheet items is measured. If there is a substantial drop in key items the bank is assumed to be more vulnerable and prone to failure when overall economic conditions are unfavourable.

Another approach is to measure market values of various indicators, e.g. debt, market capitalization, and their response to an overall drop in market indices. The advantage of this approach is that data is easily available and hence not difficult to measure. However, it has the drawback that market prices are sometimes not very representative for overall bank conditions, especially during periods of recessions. Before the most recent financial crisis stress tests were conducted by individual banks to assess their vulnerability to shocks whereas after the crisis the European Banking Authority (EBA) has started performing stress tests for systemically important financial institutions on a regular basis as well as additionally disclosing the results from them to identify fragile financial institutions and stimulate the repair of balance sheet items to increase resilience of the banking sector.

This article is the first to observe the variation in systemic risk induced by interconnections through syndicated lending for both market and balance sheet based items. Market models such as LRMES and SRISK measure the capital shortfall of an individual bank conditional on a simultaneous drop in equity prices of a market index. Similarly, results from stress tests obtained from the EBA website measure the

(7)

reaction of particular parameters to a shock in the financial system. However in this case they rely on balance sheet items instead of market values.

Europe as a bank-oriented economy represents a particularly suitable area for research in a syndicated lending context. A large part of the corporations in Europe obtain capital through bank loans instead of purchasing debt or equity through capital markets. Hence, loans represent a substantial amount of a corporation’s assets and financial intermediaries are of primary importance for businesses and their growth.

Overall, the existing literature on syndicated loans focuses mainly on the US. Considering the fact that syndicated loans represent a main source of external financing for corporations in the Euro area, this research intends to test if there is a positive relationship between interconnectedness measured by syndicated loans and the risk exposures of the financial system in Europe. An important channel through which syndicated lending might have increased systemic risk during the financial crisis of 2008 is the portfolio overlap. Consequently, this research quantifies commonality of asset holdings through participation of banks in syndications with other banks in the same industry or region. I argue that underwriting a loan to the same industry measured by the primary borrower SIC code on the one hand and to multiple different regions through the headquarter nation of the borrower on the other leads to a high exposure of different banks to the same shocks which increases vulnerability in crisis times. Data on syndicated loans in Europe between 2000 and 2017 has been obtained from the Thomson One database whereas systemic risk measures have been hand-collected from the NYU V-Lab website. To my knowledge this is the first research to analyse the syndicated loan market in Europe for a broad timeframe to capture three crises – the Dotcom bubble and the Great Recession in 2008 followed by the sovereign debt crisis in 2010. The timeframe provides a unique opportunity to compare the expansion with recession periods through time series and estimate the impact of syndicated lending, or commonality in asset holdings measured by syndicated loans on systemic risk.

The relationship between syndicated loans and systemic risk is particularly important for regulators and the design of guiding principles for regulatory interventions and political actions in order to ensure the

(8)

resilience of European banks in periods of economic downturn. Banks as a main source of financing for businesses and private households in the Euro area require a detailed analysis and monitoring to forecast and accurately design the actions to be taken. Hence, this research provides an insight into a possible channel of contagion on the asset side of a bank’s balance sheet – the syndicated lending. By measuring interconnections between banks in Europe, it aims to capture the contagion of risk and its expansion through the European economy. If syndicated loans are positively affecting systemic risk they should be taken into account when designing macroprudential policies to ensure the viability of the financial system – especially in times of economic downturns.

The thesis proceeds as follows: in section 2 the most crucial research related to syndicated loans and the systemic risk of banks is discussed. This is followed by a description of the applied methodology in section 3. Section 4 presents the data and its characteristics. The results obtained are summarized in section 5 followed by robustness checks in part 6. A conclusion supported by some policy implications and propositions for future research is provided in section 7.

2. Literature Review

The research literature of the syndicated loan market in Europe has been scarce and the field is not yet fully explored. Characterized as a bank-oriented economy in terms of financing, Europe provides unique opportunity to analyse syndicated loans in a broad macroeconomic perspective. The market for syndicated loans has various advantages. The most popular in the empirical literature is the diversification motive. Simons (1993) argues that by reducing exposure to one borrower, a bank can expand into different industries and geographic areas. Hence, small and midsized banks with limited investments in local assets can easily diversify their portfolio by participating in a syndication to a foreign corporation without bearing any additional costs. Gadanecz (2004) also suggests the risk diversification in an individual banks’ portfolio as a reason to take part in a syndication.

(9)

In a syndicated loan different banks are sharing the credit and market risk of the loan. As many lenders are involved in syndicating a loan to a single borrower, there is no excessive exposure of a bank to a customer which decreases the individual risk in case of default on the loan. An additional advantage of this hybrid type of instrument as mentioned by Gadanescz (2004) is the lack of disclosure requirements supporting public type of debt such as bonds. Consequently, syndicated loans are not subject to regulations that require banks to reveal the loans in their portfolio to the public.

Another reason for a bank to engage in a syndication may be the low capitalisation in case the bank cannot comply with capital requirement restrictions and therefore has the option to form a syndicate with other banks. Thus, if a financial institution has a capital to assets ratio which nears the regulatory standards, it may not be able to originate any large loans, but it may be capable of sharing them with other banks. Basel III requires banks to maintain a capital to risk-weighted assets of 8% which is difficult to retain for undercapitalized banks close to the minimum. Therefore, banks can engage in a syndication in order to facilitate loans without exposing a large portion of their assets to extensive risk while adhering to capital requirements. In the same line of thought, if a small bank wants to lend to a large borrower, a syndicated loan is the only alternative to do so by keeping a smaller portion of the loan while maintaining a stable capital adequacy ratio (Simons (1993)).

There are various benefits from the syndication of loans also for participating banks pointed out by Allen (1990), e.g. lower origination costs and establishment of a relationship with borrowers. Generally, banks with former relationship to borrowers will mandate the loan and depending on the size of the loan would invite other banks to participate. This represents a unique opportunity for smaller lenders to generate relationships with corporations and improve relationship lending. Nevertheless, syndicated loans do not represent a typical lending relationship - unlike unilateral types of lending. Whereas relationships in lending were found to reduce information asymmetry between a bank and its clients by ensuring monitoring and screening of the borrower and by providing the opportunity to take advantage of a repeated interaction with the same customer without increasing the costs of gathering new information (Greenbaum and Thakor (1995)), syndicated lending is different due to the role of lead arrangers. They take the main responsibility

(10)

for arranging the loan. In this line of thought, participant banks have a more passive role in a syndication which may induce various problems arising from the separation between the responsible side which is in control (the lead arranger) and the participant banks taking a passive role in the mutual agreement. The existing literature identifies a moral hazard problem stemming from information asymmetry between the arranger or senior bank and the participating parties. This is the case because the leading bank or multiple leading banks are responsible for managing the process – from due diligence and monitoring of the borrower to finding other lenders to participate in the loan. Simons (1993), Jones et al. (2005) and Sufi (2007) examine whether agent banks engage in opportunistic behaviour in case of a problem with a borrower and decide not to share all the available information, but find no evidence of such pattern in a lead arranger’s actions.

While the majority of articles has focused on the benefits and motivation of banks to participate in a syndicate, there is barely any evidence on the costs of this specific type of loans. Hence, there is a paucity of literature on the disadvantages of syndicated loans and particularly on systemic risk that similarity in banks’ portfolios may impose through this type of credit.

Systemic risk is defined as the risk of a collapse of the whole economy rather than solely a particular entity. There are several reasons discussed in the literature which urge banks to engage in similar risk exposures thus increasing the probability of systemic breakdown. The first theoretical framework builds upon the claim that banks have similar exposures to risk through correlated investments by investing in the same assets. There are several explanations for this. Acharya (2009) argues that a failure in one bank’s assets decreases the deposit rate for surviving banks and hence imposes a drop in returns. Consequently, financial institutions have an incentive to build similar portfolios and engage in herding behaviour by either failing or surviving together. Similarly, Acharya and Yorulmazer (2008b) argue that regulation causes banks to engage in the same risk by undertaking similar investments. In this case the government has no other option than to organise a bail out, as too many institutions are encountering the risk of failing together.

Another type of risk which connects financial institutions is liquidity risk. Bhattacharya and Gale (1987) state that banks have the incentive to invest in illiquid assets, as they can obtain liquidity in the

(11)

interbank market whenever they face a shortage. The authors present a free-rider problem in a model where the interbank market is an insurance measure against illiquidity that has to be supported by every intermediary. The additional security that banks gain from its existence incentivises the individual bank to hold less liquid assets, which raises the risk for the bank and effectively also the risk for the whole system. Therefore, in case of a crisis, all banks are expecting to borrow from others in the interbank market which can lead to an overall illiquidity shock.

The third underlying reason for banks’ excessive risk taking is related to capital requirements under Basel III. Perotti et al. (2011) argue that in the presence of capital regulations, banks might choose more tail risks. Investment in tail risks, characterised by a high potential payoff and infrequent but extremely high losses in proportion to the initial capital, might be considered necessary by banks to compensate for the unused capital they are required to hold in reserve. Additionally, if the costs of failing the requirements (called capital adjustment costs in reference to the bank having to adjust its equity) are considered, those institutions with a borderline capital ratio would refrain from taking the risk. However, highly capitalised banks would feel secure in not incurring the adjustment costs due to their buffer and have more incentive to take the tail risk investments. The high regulation requirements thus lead to high risk-taking of well capitalised banks.

Financial innovation and the emergence of new financial products such as derivatives, loan sales and structured products have certainly possibly to the individual diversification of a bank’s portfolio, but at the same time have led to increased similarities in the portfolio of different financial institutions which imposes risk of a systemic failure in case of an individual financial distress.

Another motivation for further research in this area is the decreasing pattern in the syndicated loan market during the recent financial crisis. Howcroft et al. (2014) analyse the market for syndicated loans in Europe over time and conclude that whereas syndicated loans constitute a large part of funding, after the Great Recession banks have significantly reduced the portion of syndicated loans in their portfolio and have turned to traditional unilateral bank-oriented lending practices. Hence, banks must have experienced some form of a negative externality from syndicated lending in their portfolios during the financial crisis. The

(12)

authors discuss the syndicated lending problem from a participant bank point of view and create a link between syndicated loans, their demand and supply effects during the financial crisis of 2008 and the transmission of shocks within financial systems internationally. They argue that if the sharp drop of syndicated lending in Europe immediately after the crisis was due to supply side shocks, regulatory requirements such as Basel III should have an impact on the syndicated loan market.

Drapeau et al. (2015) examine the impact of syndicated loans on systemic risk in Canada. They test the effect of homogenization of national loan portfolios on risk in the financial system for the 6 largest Canadian banks but find an ambiguous relationship between the homogeneous portfolio of banks and systemic risk, depending on the measurement of risk. Their methodology covers the period between 1995 and 2012 and consists in simulating fictitious loans to analyse the portfolio of banks in Canada in absence of syndicated loans by transforming them into bilateral loans to decrease endogeneity in the choice between the two types of loans. They find that syndicated loans affect diversity on an individual level in terms of industry, but not geography. Therefore, loan syndications represent an opportunity to enter new industries but there is no evidence of a positive relationship with geographical expansion.

Cai et al. (2017) analyse the portfolio connections between banks and the risk of systemic crises. They investigate the syndicated loan market in the US between January 1988 and June 2011 to measure interconnectedness of banks and find a positive relationship between bank connections and systemic risk during recessions. The authors argue that similarities between banks’ portfolios are one cause of contagion effects among banks during crises. Furthermore, they conclude that the more interconnected a bank is, the higher its exposure to systemic risks. The investigated banks in the US also show a tendency to collaborate in loan syndication with partners who have a closely resembling portfolio, thus exacerbating the problem. My research aims to investigate if there is a similar impact in Europe. As a bank-oriented system, Europe represents a reasonable opportunity to analyse the effect of syndicated loans on systemic risk. Syndicated loans are a hybrid between relationship banking and arm’s length banking, wherein the lead arranger maintains a relationship with the borrower, while participant banks are less likely to do so (Dennis, Mullineaux (2000)). Furthermore, Champagne and Kryzanowski (2007) argue that banks involved in a

(13)

syndicate tend to repeatedly engage in loan participation with the same lenders, both as participant or lead arranger. Consequently, the portfolio overlap between various financial institutions is increasingly homogenous over time. The main focus of this research is on portfolio similarities between lead arrangers in the European area, which is likely to be affected also by relationship lending.

The timeframe of my research 2000-2017 captures the integration of the European markets and the role of syndicated loans during this period, as well as the most recent financial crisis which resulted in contraction in lending internationally. The last stage of European monetary and economic integration starts in January 1999 and introduces the euro as a common currency with a transition period of three years. Hence, this research captures the elimination of exchange rate risk between the countries who adopted the Euro as a common currency. This period of time is especially crucial because of the disadvantages a monetary integration may impose, namely the transmission of monetary shocks due to the higher sensitivity of member states to global financial shocks. In this line of thought, this research represents the opportunity to test if systemic risk was increased after the European Monetary Union (EMU) because of the highly integrated markets.

Some theories conclude that there is an optimal level of integration of financial institutions within various asset classes which decrease idiosyncratic risk. Caccioli et al. (2014) develop a framework in which diversification is beneficial up until a certain threshold, which if crossed results in financial contagion between financial institutions. They identify three channels of risk, namely counterparty risk, roll-over risk and common asset holdings (i.e. overlapping portfolios). Counterparty risk occurs when an institution is in distress and cannot pay its debt which leads to the other institution failing as well. Roll-over risk on the other hand is related to the interbank lending. Banks generally finance long-term assets in their balance sheet by borrowing short-term, or in other words loans with low liquidity are financed by short-term uninsured debt. Roll-over risk takes place when banks are unable to receive short-term borrowing and fail to fund long-term investments consequently because their debtholders are in distress. Acharya et al. (2011) refer to this inability to borrow short-term if no asymmetric information or other problems are present as “market freeze” and relate it to the recent financial crisis. In this period the maturity mismatch of assets and

(14)

liabilities of banks led to a collapse of the system after the interbank lending stopped functioning and funding liquidity dropped dramatically as a result.

Whereas the first two channels of risk have been extensively discussed in the literature, my research is directed at the third one. The underlying theory assumes that when a bank is in distress or goes bankrupt, the decision to sell its assets results in devaluation of other banks’ portfolios which include identical assets. I argue that this process may be facilitated to some extent by loan syndicates. Hence, similar exposures to assets (e.g. syndicated loans) in a certain industry may lead to an overall decline in asset prices and transmit financial distress in case of economic shocks to the system. This is also related to the model of Caccioli et al. (2014) who rely on the assumption that a bank holds a particular asset portfolio which is liquidated in case of bankruptcy. Therefore, through a simple market mechanism, the price of the asset drops after a fire sale which leads to a devaluation of the same asset in other banks’ portfolios and may in turn result in additional defaults. The authors test how the stability of a financial system is impacted by a network of banks and assets and find that in a stable system shocks are unlikely to propagate, whereas an instability imposes contagion risk and multiple bankruptcies, increased through leverage in financial institutions. Consequently, we would expect economic shocks to spread faster in times of crisis and instability and moreover, systemic risk to be positively affected by the portfolio similarities of banks especially when economies are unstable (e.g. during the financial crisis of 2008). This is due to the interconnected banks’ vulnerability to transmission of shocks.

3. Methodology

3.1 Dataset and weights

The following section describes the raw data set for this research and how the interconnection measures between banks have been constructed. Syndicated loan data in Europe for the time period between 2000 and 2017 has been obtained from the Thomson One database. Similar to the methodology of Cai et

(15)

al. (2017) the measurement depends on the total loan syndicated by lead arrangers. The focus in this research is on lead arrangers, as they are generally holding a larger share in a syndication. Ivashina (2009) examines the information asymmetry between lead and participant banks in a loan syndication and concludes that the lead bank’s ownership has a negative effect on asymmetric information in a syndication. Hence, if a lead arranger owns a higher share of the total loan in a syndication, it reduces the compensation required by other participants to engage in the loan since they are convinced that the borrower is creditworthy. This leads to the conclusion that the syndicated loan market is run profoundly by large financial institutions. For each loan in the database, there are several agents mandating the lending procedure and they have an incentive to own a higher share of the loan in order to signal to other participants that the borrower has a sound credit rating by simultaneously decreasing the cost of borrowing.

As there is more than one lead arranger engaging in syndicating a loan to a single borrower in the dataset, I have initially assigned equal weights to each lead agent in the syndication. In another composition of the model I have assigned size weights depending on book values measured by the portion of the loan as a percentage of the whole loan market in a particular year. There are 14,395 unique month - loan syndications for the observed period of 17 years. These loans have been sorted by year and industry of the borrower in order to calculate the weights for each year in which every lead agent had exposure to a particular industry through a loan syndication. If wa, j, t is the weight of bank a in industry j during year t,

then ∑𝐽𝑗=1𝑤𝑎,𝑗,𝑡 = 1, where J is the number of possible industries per year in which a lender can be

exposed to by participating in a syndication to the borrower in that industry. Consequently, all the weights for a lead arranger in a year within an industry should sum up to 1.

Another methodology used in this research seeks to capture the regional aspect and measures interconnections between lead arrangers by the headquarter country of each borrower in a syndication. Hence, the portfolio overlap between regions is measured as a proxy for interconnections between banks. Champagne and Kryzanowski (2007) argue that in addition to the fact that past relationships between lenders impact their future decision to form a syndication, there is a certain home bias in the selection of

(16)

loans to participate in. Thus, lenders prefer to form syndications with domestic instead of foreign counterparts. There are various reasons impacting this decision, e.g. legal or cultural environment, risk aversion, development of the country etc. Intuitively, we would expect the results from the regional aggregation to be more concentrated and to lead to higher interconnections. In other words, interconnections between lead arrangers measured by overlap on a regional level of the borrower should be stronger than interconnections proxied by industry commonality if a home bias effect is present. If banks prefer to invest in their domestic country, their interconnections will be stronger within lead arrangers in the same country and weaker with foreign counterparts. In addition, banks granting loans to many borrowers in the same country are more likely to segment themselves from other financial institutions and should be less exposed to contagion effects from abroad. Another reason for a higher concentration on a regional level in comparison to the industrial aggregation is the fact that the dataset consists of only 45 headquarter nations and 738 industries in contrast. Hence, a higher concentration of loans on a regional level during recessions is expected to be more pronounced and should prove beneficial in an economy-wide context since it is less prone to spill over effects of risk.

3.2 Distance between banks

The next step is calculating the difference in portfolio overlap measured by borrower industry or region between each bank and every other bank in a particular year. Following the methodology of Cai et al. (2017), the similarity in a bank’s portfolio is measured by the Euclidean distance of the weights:

𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑎, 𝑘, 𝑡= 1

√2∗ √∑ (𝑤𝑎,𝑗,𝑡− 𝑤𝑏,𝑗,𝑡 𝐽

𝑗=1 ) (1)

The Euclidean distance is a mathematical concept estimating the distance between two points in a plane, or in our case – between each two variables. Consequently, the Euclidean distance measures the absolute difference between each pairwise combination of portfolio weights of banks within a year. Therefore a low value will be associated with higher portfolio overlap of two banks within an industry in a

(17)

particular year, whereas a high value would mean both institutions have low weights invested in the same industry and therefore a low portfolio overlap in terms of syndicated loans. On the other hand, a higher portfolio similarity on a regional level would induce more segmented markets in a particular country which is associated with lower risk in the European economy overall.

An example of how distance between lead arrangers on an industry level has been calculated is showed in Appendix B.It represents three lead arrangers engaging in syndicated loans to different borrowers

for the year 2007. The primary SIC code enables classification per borrower industry and measures the extent of portfolio overlap. In 2007 Standard Chartered has borrowed 3.77% to a corporation which specialises in crude petroleum and natural gas, 3.56% in wood products, 86.96% in radiotelephone communications and 5.71% in depository institutions. In contrast, Svenska Handelsbanken AB and Swedbank have invested among others 17.93% and 30.72% in depository institutions respectively. Thus, we would expect a small portfolio overlap between these lenders, as this is the only industry all three banks have specialised jointly into. The distance measures calculated as in equation (1) are relatively high respectively, capturing the difference in the portfolios of the institutions. Similarly, the nation headquarter of the borrower represents the homogeneity of portfolios within the same country. However, different banks investing in the same country are expected to have lower distance and higher portfolio similarity based on exposure to the same borrower headquarter country. In the same manner, banks participating in syndications to borrowers from different countries are supposed to have a higher distance from one another and a lower overlap in assets. The regional aspect of interconnections is expected to decrease systemic risk, whereas higher industry overlap results in increased probability of contagion and systemic risk respectively.

On the other hand, there is a certain flaw in the computation of regional weights since this methodology does not account for the individual diversification of a lead arranger. Since concentration on a regional level is calculated by the distance between each pair of banks, this is not informative for the concentration of the individual portfolio. Thus, a bank might invest in only one country throughout a year and will have a complete portfolio overlap with another bank investing in that country. Similarly, if a lead

(18)

arranger invests in several regions, it will have strong similarity to other banks investing in the same regions. This two contradictory effects might lead to ambiguous results.

3.3 Interconnections between banks

In order to measure interconnections between banks’ portfolios of syndicated loans, the sum of the distance of each lead arranger to all other lead arrangers has been calculated. As a higher value of Euclidean distance corresponds to a lower interconnectedness, the measurement has been normalized between 0 and 1 with 0 representing the lowest level of interconnection and 1 the highest interconnections between banks. The equation looks as follows:

𝐼𝑛𝑡𝑒𝑟𝑐𝑜𝑛𝑛𝑒𝑐𝑡𝑒𝑑𝑛𝑒𝑠𝑠 = (1 − ∑𝑎≠𝑏𝑤 𝑎,𝑏,𝑡∗ 𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑎,𝑏,𝑡) (2)

where Distance a,b,t measures the Euclidean distance as in equation (1) and w a,b,t are the weights assigned

to each bank. Two types of weighting have been used - equal weights, and weights based on size in order to take into account the fact that larger banks holding a higher share in the syndication lending market may have a higher impact on systemic risk. Consequently, larger banks are assigned higher weights depending on the percentage contribution of an individual bank from a single loan participation to the overall syndicated lending market. We would expect the size-weighted interconnectedness to derive more accurate results, as systemically important financial institutions are generally the largest ones and systemic risk should therefore be more substantially affected by larger banks. Size weighting measured by syndicated loans captures the specific contribution of each lead arranger to the loan market. A larger bank investing the same industry or region as another bank will hence induce stronger interconnectedness between both lead arrangers.

3.4 Measuring systemic risk

The first two systemic risk measures are obtained from the NYU V-Lab database and are based on the assumption that efficient markets provide sufficient information about pricing and future events as news

(19)

are already incorporated in the price of securities. In this line of thoughts, since prices are derived from easily available public market data they provide a more accurate representation of timely occurrences. Nevertheless, Benoit et al. (2017) point out at one disadvantage of the global systemic risk measures related to their lack of theoretical background which complicates the process of identifying the origination of risk. They argue that instead of identifying a particular channel through which systemic risk propagates through the whole economy, the market-based estimations identify risk on a global scale through multiple channels without determining its source.

The first measurement of risk was initially developed by Acharya et al. (2012) and is called Marginal Expected Shortfall (MES). It represents a derivation of the Expected Shortfall (ES). ES is obtained from the common measure of firm risk, namely VaR (Value at Risk). VaR estimates the potential loss in the presence of an unlikely event, or a shock, or how much an institution will lose with probability 1-α, where α is typically 1% or 5%. Hence, if α=1%, VaR estimates the bank’s losses with 99% probability. ES is an estimation of the expected loss provided that it is higher than the VaR prediction. Hence, the expected shortfall measures the average return when the loss is in excess of VaR. MES is the partial derivative of ES and measures the rise in systemic risk by a simultaneous increase of a weight of a firm (wi) as per below:

𝜕𝐸𝑆𝛼

𝜕𝑤𝑖 = −𝐸 [𝑟𝑖|𝑅 ≤ −𝑉𝑎𝑅𝛼] ≡ 𝑀𝐸𝑆𝛼 (3)

W is the sensitivity of firm i to systemic risk. Hence, the MES captures the loss in equity of a bank α when market i losses fall below the VaR. Thus, the above equation measures the marginal contribution of a bank’s shortfall to the financial system’s losses measured by their expected return - (R). Acharya et al. (2014) expand the concept of MES to the long-run MES (LRMES) which measures the decrease in equity if the market returns experience a substantial fall in a six-month period of time. The V-Lab authors have calculated LRMES by taking into account a 40% drop in the market index, as the general threshold for market crash in a crisis is considered to be 40%. As a proxy for the market index they have used the MSCI

(20)

World Index. Hence, LRMES captures the loss in bank’s equity returns conditional on a market crash and measures the capitalisation of financial institutions required to survive a severe financial crisis. This estimation is highly sensitive to market capitalisation of the financial institution and will affect to a substantial extent the long-run MES of banks with high equity and low leverage in the sample. The calculation of LRMES is especially important in a crisis context as it gives insight into the question of how much equity a particular institution will need in case of a decline in overall returns. As such, it captures the connection between individual returns of a bank and the market returns. The more correlated a bank’s return to the overall system, the higher its shortfall will be.

The second measure is an extension of the LRMES model and is called SRISK (Systemic Risk). It was first introduced by Acharya et al. (2012) and takes into consideration not only the fall in equity but also the size, leverage and interconnectedness which are mentioned as crucial in determining systemic risk (Murphy et al. (2011)).

SRISK estimates the expected loss in capital in case of another financial crisis. Systemic risk is calculated as below:

𝑆𝑅𝐼𝑆𝐾 = 𝑘. 𝐷𝑒𝑏𝑡𝑖,𝑡− (1 − 𝑘) ∗ 𝐸𝑞𝑢𝑖𝑡𝑦𝑖,𝑡∗ (1 − 𝐿𝑅𝑀𝐸𝑆𝑖,𝑡) (4)

Where k is the capital requirement as per regulation (5.5% for Europe and 8% for the US). The inventors of the SRISK variable have chosen 5.5% as prudential capital ratio for European banks and 8% for US banks to capture the different accounting standards which impose difference in leverage. As European banks are obliged to report under IFRS they are not allowed to use netting of derivatives which substantially extends the balance sheet. On the contrary, US GAAP allows for reporting derivatives on a net basis, which decreases total assets.

Equity in the above formula denotes market capitalisation of bank i in year t, or stock price multiplied by the number of shares outstanding, and debt is calculated by subtracting the book value of assets from the book value of equity. SRISK is reported in million USD. Hence, if the above formula derives

(21)

negative results, they are interpreted as no risk. In this specification 0 is the point where no systemic risk is present. Consequently, values under the zero threshold represent no risk at all. The above formula leads to the conclusion that an increase in debt is positively related to the systemic risk estimation. In contrast, an increase in current market capitalisation leads to a decrease in systemic risk. Similarly, the SRISK ratio is the share of a firm’s loss in an overall market shortfall and is derived by the following formula:

𝑆𝑅𝐼𝑆𝐾 𝑅𝑎𝑡𝑖𝑜𝑖,𝑡=

𝑆𝑅𝐼𝑆𝐾_𝑖,_𝑡

∑_𝑗∈𝐽𝑆𝑅𝐼𝑆𝐾_𝑗,_𝑡 (5)

Where J is the sum of all the firms with positive systemic risk for the respective month t. Hence, the formula captures the capital shortfall of a single institution scaled to the whole economy’s risk, or its contribution to the overall risk in the economy. The introduced by Acharya et al. (2013) model of systemic risk is an alternative to the risk weights based on the Basel accord but using market values instead. The authors propose regulatory minimum of SRISK equal to 0 in order for a firm to stay sufficiently capitalised in times of crisis. The calculation of SRISK in relative terms allows for scaling and limits the values to positive.

The third measurement is related to stress tests implemented by the European Banking Authority (EBA) for the countries in the European Union. Stress tests are generally similar to the above described measure of systemic risk – SRISK but they are slightly less severe. The reason is that V-Lab captures stress test losses over 6 months without taking into account stress revenues whereas the stress tests of the EBA have a projection of 2 years and account additionally for revenues (Acharya et al. (2013)). Stress tests in Europe are conducted by regulators every few years since 2009. The EBA aims to increase financial stability in the European Union and to strengthen the banking sector accordingly. Hence, the institution is responsible for designing prudential regulation policies to ensure efficient functioning according to common and comprehensive supervision standards. Stress tests consist in exposing the individual bank’s balance sheet to a hypothetical economic shock and assessing its impact on different indicators. The most

(22)

important one is the CET1 ratio which is the common equity tier 1 defined in the Basel Capital Accord from 1998 as common stock and retained earnings. In addition, preferred stock (redeemable and non-cumulative) may be included in the tier 1 capital. It measures the capitalisation of a bank which is crucial in times of crisis, as it decreases leverage which in turn decreases volatility and risk. Higher capital requirements generate safety in financial institutions without imposing additional costs of capital (Admati and Hellwig (2014)). Demirguc-Kunt et al. (2010) find that better capitalised banks during the recent financial crisis had higher stock prices and the effect is larger for high quality capital such as tier 1 capital. Therefore, the CET1 ratio is a reasonable indicator when estimating the financial viability of an institution and its resilience to systemic risk due to shocks in the economy.

Another crucial parameter included in stress tests is the total risk exposure of an individual bank. It includes credit, market and operational risk of a specific financial institution. The majority is represented by credit risk defined as the deterioration in financial assets and sovereign exposures. Market risk represents changes of the values of net trading income, whereas operational risks are related to impaired processes and systems within the bank (EU-Wide Stress Test Results (2016)). According to the EBA Until 2016 stress tests were performed on a pass-fail basis by adding a specific threshold, whereas from 2016 onwards there is no threshold and the test is designed to be informative and support the review of financial institutions in improving their balance sheet.

In general, the distinction between the above mentioned models of risk lies in its definition. There is a slight difference between systemic and systematic risk. Hansen (2012) provides a comprehensive interpretation and compares both concepts. Systematic risk has been extensively researched in the empirical literature and represents economic risk of the market which is not subject to diversification. Hence, investors financing securities which incorporate this type of risk are rewarded by receiving a higher return on their investment. In contrast, systemic risk measures the breakdown of a financial system as a whole. As such, it may be triggered by liquidity shocks or internal shocks which propagate through networks between institutions (Hansen (2012)). In this line of thoughts LRMES represents systematic risk, or the correlation with the market – beta. Hence, the higher a bank’s correlation with the market, the higher its systemic risk

(23)

importance. SRISK on the other hand is dependent on leverage, beta of the firm and market capitalisation. Thus, SRISK increases the importance of large banks for systemic risk by including leverage and market capitalisation into the measurement. This corresponds to the too big to fail concept whereas LRMES is derived from interconnections through beta and is more in accord with the too interconnected to fail hypothesis of Markose et al. (2012) which states that high concentration between financial institutions may induce contagion and impose a thread to the financial system, especially through financial derivatives.

Generally, this research aims to provide separately an insight into the impact of syndicated lending on systematic risk through interconnections of banks and on systemic risk, accounting for leverage and size of the firm.

Stress tests are similarly in nature to the first two measures of risk – LRMES and SRISK, but they take into consideration the effect on balance sheet items and not market values. Hence, they capture the effect on book values but are more difficult to obtain, as data for calculation is not easily available. An advantage is that the EBA discloses the results from hypothetical shocks to the balance sheet items of banks on their website. Therefore, in case of doubt in the market efficiency hypothesis, stress tests should be a more reliable method for estimating systemic risk.

Stress tests are performed every few years by the European Banking Authority and aim to estimate how systemically important financial institutions react to shocks in the economy. The EBA is responsible for testing the viability of banks in the European Union. In order to analyse this relationship the EBA measures how capital positions change in a timeframe of 3 years under two scenarios, namely a baseline and an adverse one. The baseline scenario is assessed by the European Commission whereas the European Systemic Risk Board (ESRB) is responsible for providing the results from the common adverse scenario. Results are published and disclosed at the EBA website. Simultaneously, the ECB conducts the same tests for additional banks which are not provided to the public and every individual bank can decide to publish the resulting estimations. The methodology includes stress tests using a static balance sheet which in other words means that assets and liabilities do not change in respond to the worsened conditions under the stress scenario. This restriction in the assumption of the model implies that any measures undertaken to prevent

(24)

the collapse of balance sheet items are not taken into account. Hence, the estimated results might be biased upwards as banks might be able to exert higher resilience to shocks in reality by undertaking measures against them. To quantify the outcomes of stress tests for banks in the European Union the model forecasts a hypothetical drop in total risk exposures in the following three years.

This research attempts to find the relationship between interconnections of banks measured by syndicated loans and the change in risk exposures under a stress test scenario. Interconnections are estimated once again by the pairwise combination of banks and their portfolio overlap. However, the systemic risk variable (SRISK and LRMES in the former model) is substituted by the actual risk exposure in the year before the stress tests were performed and the projections of risk exposure for the following 3 years. This methodology allows for integrating the data from stress test results into the time series data of syndicated loans. The total risk exposure includes defaulted and non-defaulted assets under the baseline and adverse scenario using regulatory risk specifications. My methodology of incorporating stress tests into the model could be explained by the following example: For the stress tests of 2014 the actual figures in year 2013 are matched to the variable representing interconnections for the same year for each individual bank. Subsequently, projections for the following years – 2014, 2015 and 2016 under the adverse scenario for each financial institution individually are matched with the interconnections of banks estimated by syndicated loans for the respective years. Hereby, the risk exposure of banks is regressed on their interconnections for 4 subsequent years.

One disadvantage of the model is that it only captures a limited timeframe and it is not capable of including crisis years into the analysis in order to distinguish between risk levels in recession and expansion periods. Another flaw of stress tests is that they only measure risk of individual banks which does not correspond to systemic risk in the economy. Stress tests, on the other hand, provide a more precise analysis on different balance sheet indicators such as various assets and exposure levels.

This is the first research to incorporate stress tests of the EBA into a framework to examine the impact of overlap in banks’ portfolios through loan syndication on risk and to compare the results with market-based measures of risk.

(25)

Overall, the main disadvantage of the above described methodology is the lack of randomization or experimental design due to the individual choice of each bank as to whether to participate in a syndication or not. Thus, further research could be directed at the assessment of what impacts the decision to form a syndicate, or quantitative analysis of all assets on a bank’s balance sheet separately to estimate the contribution of each of them to systemic risk.

4. Data and Descriptive Statistics

Data on syndicated loans have been obtained from the Thomson One database and systemic risk measures in Europe can be found on the NYU V-Lab website. Results and information regarding stress tests in Europe have been collected from the website of the European Banking Authority (EBA). Syndicated loans and systemic risk data based on market values capture the period between 2000 and 2017. The syndicated loans data include borrowers from 45 country headquarters in Europe – a list of the countries is provided in the Appendix Section (Appendix A). The data cover 738 unique industries in which borrowers of syndicated loans are operating internationally. Thomson One provides monthly data on syndicated lending with a loan package amount measured in million USD with all lead agents participating in a single loan. The financial crisis of 2008 led to a substantial drop in syndicated loans and marked the beginning of a recession leading to a consequential drop in the lending market during the following years. This pattern captures the economy-wide stagnation leading to deterioration in overall lending and slow-down in investments, consumption and growth. The drop continues reflecting the following sovereign debt crisis in Europe and reaches a minimum in 2012. The structure of the syndicated loan market exhibits a pro-cyclical behaviour which is in line with the economic conditions and the overall financial stability of the economy within the last 17 years. Following the most recent financial crisis, banks were exposed substantially to illiquidity after the collapse of the mortgage-backed securities market and the subsequent drop in stock markets worldwide. As government funds were used to bailout large and systemically important financial institutions, fiscal deficits resulted in a period called the Great Recession of 2008-2012. Hence, many

(26)

countries were exposed to risk of default due to their inability to repay the massive increase in debt (Reinhart and Rogoff (2009a)).

The syndicated loans data is used to measure interconnectedness between financial institutions based on participation of lead arrangers. The lead arranger in a syndication is responsible for negotiating the loan and finding other participant members. As mentioned earlier, this role is especially crucial because the largest part of the loan is usually taken by the lead of the syndication (Sufi (2007)). Therefore, interconnectedness is measured by participation in a syndicate on a lead agent level. Other participants providing funds are neglected since they play a passive role in a syndication whereas lead agents are not only involved in syndicating loans with other participants but are also participants in other syndications where the former participants are now lead agents. This reciprocal arrangement leads to reduced moral hazard due to the benefits of mutual monitoring among parties but is also a source of increasing interconnectedness between lead agents and turns them into potential contributors to contagion of systemic risk in a syndicated loan context (Cai (2010)). Consequently, participants in a syndication who have no role as lead agents in another loan do not increase interconnections between banks as they are only involved in funding a single loan but do not interact with other lenders to increase exposure to risk.

After merging the two datasets there are 159 unique lead arrangers including their branches in different countries. Interconnections are measured as described in the methodology section by generating all possible pairwise combinations of lead agents in an industry or region yearly. For the construction of this variable I have initially used equal weighting - per industry and per region on a bank level. Hence, every lead agent or bank is assumed to have equal contribution to interconnectedness. The limitation of this assumption is that it does not account for size. Intuitively, a higher participation of institutions is more likely to contribute to higher interconnections which is not captured here. In another setting of the model, I have accounted for differences in size by including the weights of an individual institution measured by their individual contribution to the total syndicated loan market per year.

Figure 1in the Appendix sectiongives some insights into the time series pattern of the

(27)

seems that European banks are highly interconnected and volatile based on the country they are investing in. A maximum is reached in 2006 when the interconnections between banks in a syndication reaches nearly 60.5, or slightly more than 60%. Another crucial observation is that interconnections undergo a sharp drop after 2007 until reaching a minimum in 2009. This pattern can be explained by the contraction in lending after the beginning of the financial crisis which naturally led to a decrease in syndicated loans as a result of the overall downturn in the economy. After 2009 a gradual decrease follows and interconnections continue to fall during the sovereign debt crisis thereafter.

The time series of interconnections on a regional level represent a similar pattern depicted in the same figure, however in this case size weights account for the share of an individual bank in the syndication. A minimum of interconnections is reached in year 2005. There is a gradual increase in 2008 after the beginning of the financial crisis which apart from some business cycle volatility continues to follow an increasing overall trend until present. Hence, according to theory, thisFigure 2 is a more accurate representation of interconnections on a regional level. Intuitively, during economic downturns banks will be more likely to invest in their own countries and defer from funding risky loans abroad.

Figure 2 represents interconnections between banks estimated on an industry level. Since the

dataset contains more than 700 industries, there is no such clear pattern as described above within regions. A yearly maximum of around 60 on average is reached for the equally-weighted interconnectedness which starts decreasing thereafter until 2005 and remains stable during the following years. Accounting for size in the second specification leads to stronger interconnections on an industry level on average but fluctuates around the mean of 60 without revealing any clear trend.

The systemic risk data was manually collected from the NYU V-Lab website and covers 9,676 unique monthly bank measures of risk after merging with the syndicated loans data from Thomson One.

Figure 4 measuring the long-run marginal expected shortfall (MES) of banks in the dataset is

stable and shows no clear pattern but there is a slight increase in the time around the Great Recession until 2009 with a minor decrease thereafter. A peak is reached in 2012 when LRMES accounted for 46%.

(28)

The next figure (Figure 3) is a graphic depiction of systemic risk over time estimated by SRISK in million USD. As expected, there is a clear increase in systemic risk in the period of the financial crisis of 2008 and 2009. Systemic risk of European banks in the database starts decreasing after July 2009 and increases once again until reaching a peak of USD 12,541 million in 2012. After that it gradually decreases again. This pattern leads to the conclusion that financial institutions might have attempted to decrease their exposures by reducing debt and strengthening their capital levels straight after the beginning of the financial crisis in 2010. The decrease in systemic risk may be explained by the introduction of Basel III which called for increase in capital buffers, quantification of liquidity risk and decrease in bank leverage (Penikas, 2015).

Figure 5outlines time series of another crucial variable in the context of overall risk for the last 17

years, namely leverage. Leverage is measured by debt over total equity and is an increasing function of risk. Hence, if the leverage ratio increases, this might lead to overall increase in exposure and a whole system which is highly indebted is more likely to experience a collapse. In this sense leverage is an indicator of risk and we would expect it to be highly correlated with the systemic risk measures of banks in the sample. Starting from July 2000, there is a sudden increase in leverage around 2002 and 2003 which gradually decreases until the end of 2003. This pattern is followed by a smooth low degree of leverage thereafter up until the financial crisis of 2008. Leverage starts to increase in 2008, undergoes a slight drop between July 2008 and July 2009 and experiences a substantial growth in 2009, peaking in 2010 – the sovereign debt crisis and decreasing completely in 2012. Thus, leverage moves with systemic risk (SRISK) from Figure 3on average. Since leverage enters the formula for SRISK with a positive sign, this observation is not surprising. Therefore, overall increasing leverage of banks may be interpreted as a sign of high exposure to risk in the economy.

The summary statistics graph Appendix

Table Vshows all the above described variables with their mean, median, lower and upper quantiles

as well as their maximum, minimum and standard deviation to obtain some intuition about the data. The systemic risk measures in absolute terms are highly volatile from a minimum of -30,270 and to a maximum of 133,267, a mean of 8,149 and a median of 599 in million USD. The relative measurement of systemic

(29)

risk has lower volatility due to the limitation to only positive values of SRISK, or a minimum of zero respectively. Hence, SRISK as a ratio has a median of 0.001, a mean of 0.021 and a maximum of 0.436 out of a sample of 9,676 unique lead arranger months. The high variation of risk over the sample period from 2000 to 2017 is reasonable, considering the fact that it captures three periods of crisis – the Dotcom bubble, the Great Recession and the subsequent sovereign debt crisis.

LRMES has a mean of 0.409, close to the median of 0.418, a minimum of 0.0512 and a maximum of 0.762. Leverage is measured as debt over equityin % and reports a mean of 21.3%, a median of 14.2% and a standard deviation of 24.3%. The Euclidean distance measures 942 unique summed distances for every individual bank in each industry with a median of 1.22, mean of 1.33 and a standard deviation of approximately 0.6. The Euclidean distance measure is further transformed to a scale of 0 to 1 where 0 captures the least interconnected and 1 the most interconnected banks. Out of 2,113 unique equally-weighted observations of interconnections on a regional level, there is a median of 0.539, mean of 0.522 and a standard deviation of 0.27. Size-weighted values exert even higher concentration on a regional level with a median of 0.638 and lower quantile of 0.367. This leads to the conclusion that banks in the sample are highly interconnected measured by borrower’s country headquarter. On an industry aggregation level, there are even higher connections. The median in the equally-weighted distance measures is 0.597 with a mean of 0.511. Size-weighted connections are also more concentrated similar to the previous specification based on regional weighting. With a median of 0.767 they are highly concentrated in similar industries.

Overall, the summary statistics of interconnections between banks do not derive surprising results considering the fact that they measure dependence between lead arrangers and according to theory consist to a great extent of large banks which take a higher share in a syndication. Hence, although smaller banks benefit from a loan syndication by receiving access to borrowers, accounting for size-weights captures the higher participation of large banks as lead arrangers in a syndication. Their portfolios estimated by syndicated loans represent higher than 50% overlap which may signal that they are profoundly exposed to the same assets and are highly likely to be affected by a change in asset prices in the same way. Due to the fact that these are mainly large banks which are classified as systemically important, in bad times, this

(30)

might lead to a simultaneous increase in non-performing loans and a subsequent systemic crisis for European banks.

5. Results

5.1 Interconnectedness and LRMES

5.1.1 Industry Aggregation

Appendix Cgives a first impression on the impact of portfolio overlap on systemic risk in Europe.

In order to measure interconnections between banks on a syndicated loan level, distance between each lead arranger to every other lead arranger has been calculated by measuring the portfolio overlap on basis of a borrower’s industry. There are 738 unique industries in the dataset captured by a borrower’s primary SIC code. The interconnectedness measure is constructed as described in the methodology section and suggests that whenever two arrangers of a separate syndication invest in the same industry, or lend money to a borrower in that specific industry respectively, they have a high portfolio overlap and stronger connections between the weights invested measured by the scaled Euclidean distance from equation (2). This part of the research investigates if stronger interconnections on an industry level increase systemic risk of European banks.

Appendix Canalyses the impact of equally-weighted interconnections on industry level on LRMES

with dummy equal to 1 if a year represents a period of recession or expansion and 0 otherwise. Identified as recession periods are the year 2001 – straight after the Dotcom bubble, as well as the years after and around the Great Recession and the sovereign debt crisis in Europe – from 2008 to 2012. Expansion periods on the other hand are determined through peaks in GDP growth in the European Union and correspond to the years 2005 and 2016. The overall effect of the interconnectedness measure on systemic risk in most regressions is positive and significant in some specifications supporting the hypothesis that connections between financial institutions built upon the syndication of loans increase overall risk in the economy. Nevertheless, the interaction dummy for connections during times of recession leads to negative and

(31)

insignificant results. Likewise, higher interconnections during overall economic expansion do not lead to consistent results.

The next step of the analysis is to account for loan share in the participation of each bank in the syndicated market. Hence, predicted by theory, lead arrangers with higher share in a syndication are also supposed to have higher weights as they provide more funding to a particular industry by granting a higher portion of their finance to a borrower operating in a particular field. To account for this, I multiply the weights invested by a financial institution in a certain industry by the individual contribution of a bank in a syndication. Columns 2 to 5 include leverage, market capitalisation and individual share in the stock market into the regression, as well as entity and time specific fixed effects. In this specification, and when adding controls for market size and time and bank fixed effects in Table I below,column 5 derives a positive and significant relationship between interconnections and LRMES – a 1 standard deviation increase in portfolio overlap by industry increases LRMES by 0.012 during a recession which corresponds to an economically significant growth in LRMES of approximately 8.7% of its standard deviation. Surprisingly, also the coefficient of interaction with expansion is highly significant and positive in the specifications leading to the conclusion that similarly in periods specified as prosperity, higher portfolio overlap of banks leads to increased systemic risk in the economy. However, this is possibly reflecting the smaller part of the sample defined as expansion period and is subject to changes in the identification of economic booms which are also different for the various countries in my sample. Columns 6 to 9 estimate the equivalent relationship but controlling for size of the bank by substituting the market based variables with book values based on participation in the syndicated lending market and the logarithm of the whole syndication market. This leads to a positive significant result in column 7 with book value controls for size but the effect disappears after accounting for fixed effects.

(32)

Table I: Size-weighted Interconnectedness and LRMES, Industry Aggregation.

The below table reports coefficients estimated from regressions analysing the relationship between interconnections of European banks on an industry level and LRMES. LRMES is measured as a ratio as described in the methodology section. Interconnections are measured by the weights of syndicated lending of each lead arranger to a borrower in a particular industry. Additionally, weights for size of the loan are added to the calculation. As a result, larger banks in a syndication are granted higher weights and have stronger interconnections respectively. The first two coefficients represent the interaction dummy between interconnections and a period of recession, or expansion. Crisis years during the timeframe between 2000-2017 fall under the first category, periods of economic booms are categorised as expansion. Further controls such as equity share in the market and participation in syndicated lending of the sample are included. Leverage controls for the impact of debt on systemic risk. Bank and year fixed effects are also included in some of the specifications and are indicated at the bottom of the regressions. Standard errors are reported in parentheses.

LRMES (1) (2) (3) (4) (5) (6) (7) (8) (9) Size-weighted Interconnections*Recession 0.002 0.035 0.045** 0.035* 0.032* 0.011 0.053** -0.016 0.015 (0.028) (0.022) (0.019) (0.019) (0.018) (0.027) (0.020) (0.026) (0.021) Size-weighted Interconnections*Expansion 0.142*** 0.105*** 0.126*** 0.132*** 0.152*** 0.130*** 0.158*** 0.113*** 0.149*** (0.028) (0.023) (0.021) (0.026) (0.021) (0.027) (0.020) (0.029) (0.023) Size-weighted Interconnections -0.137*** 0.001 0.017 -0.009 0.014 -0.045 -0.157*** 0.214*** 0.077 (0.020) (0.017) (0.025) (0.017) (0.022) (0.055) (0.048) (0.061) (0.051) Recession 0.015 -0.033* -0.047*** -0.015 -0.052*** (0.025) (0.019) (0.018) (0.024) (0.019) Expansion -0.112*** -0.084*** -0.108*** -0.114*** -0.144*** (0.023) (0.019) (0.018) (0.023) (0.017) Market Share -0.308*** -0.230*** -0.094** 0.010 (0.035) (0.041) (0.040) (0.040) Log(Market Capitalisation) 0.054*** 0.048*** 0.049*** 0.044*** (0.002) (0.003) (0.002) (0.003) Leverage 0.002*** 0.002*** 0.001*** 0.001*** 0.001*** 0.002*** 0.001*** 0.001*** (0.0001) (0.0003) (0.0002) (0.0001) (0.0004) (0.0002) (0.0002) (0.0001) Loan Share 0.152 -0.446*** 0.737*** 0.161 (0.103) (0.114) (0.114) (0.110) Log(Loan Market) 0.009 -0.005 -0.133*** -0.153*** (0.007) (0.006) (0.031) (0.025)

Bank Fixed Effects Yes Yes Yes Yes

Year Fixed Effects Yes Yes Yes Yes

Number of Observations 2,566 2,566 2,566 2,566 2,566 2,566 2,566 2,566 2,566 Adjusted R2 _0.022 _0.300 _0.569 _0.442 _0.685 _0.106 _0.522 _0.271 _0.646

Syndicated Loans and their Contribution to Systemic Risk of European Banks