The effect of banks’ risk on firms

(1)

The effect of banks’ risk on firms

Name: Jiří Píza

Student number: 10392548

26.01.2015

Bachelor Thesis

Supervisor: Dr. Tomislav Ladika

JEL classification: G12, G21

(2)

Abstract

During the Sovereign Debt Crisis, multiple banks were negatively influenced given their holdings of government bonds. This increased risk can be transferred on firms which are dependent on the banks given the strong bank-firm relationships, increased interests or decreases access to capital market. Therefore, this empirical work examines how the increased riskiness of the banks influences the firms. The study uses regression on panel data. The Fama French model is extended by a risk factor. To estimate the riskiness of the bank, returns on CDS spreads, stocks returns of the banks and idiosyncratic returns of the banks are considered. Moreover, there are added crisis dummy and interaction terms with each factor in order to capture the structural change caused by the crisis. It was found positive correlation, but the effect is stronger for bigger players. Moreover, CDS spreads were not considered before the crisis and for smaller players, compared to stock returns of the banks, which were constantly incorporated. The decreased explanatory power of CDS spreads can be attributed towards the correlation of CDS spreads and their lower explanatory power as firm-specific risk indicators. The crisis as such had negative and significant influence on the expected returns of firms, which might be due to possible mispricing. Those results reflect the reality where investors want to be rewarded for being exposed to higher risks and more dominant players are more analyzed. Also, crises help to burst bubbles and correct asset pricings.

(4)

1 Introduction

The stock returns of each traded firm are influenced by multiple factors. Apart from the firm’s own behavior and performance, inflation and money growth significantly correlate with the returns together with other macroeconomic factors (Flannery & Protopapadakis, 2002). Recently, there have been studies to identify other stakeholders who contribute to the well-being of companies. Current literature recognizes the impact of principal consumers of firms and to which extent the shocks to consumers correlate with firms’ returns (Cohen & Frazzini, 2008). Previous financial crises pointed out another possible influencer – banks associated with the firms (Ongena, Smith, & Michalsen, 2003). Bernake argues that the failure of banks worsened already the decline during the Great Depression (1983). Similar situation happened during the recent Sovereign Debt Crisis, when few countries had difficulties to meet their financial obligations (Lane, 2012). These troubles created spillovers into the banking sector causing interdependence between banks and countries. (De Bruyckere, Gerhardt, Schepens, & Vander Venner, 2013). The sovereign risk was transferred to financial institutions via asset holding, collaterals, ratings and guarantees. Consequently, for especially large banks, bonds holding correlate negatively with lending during sovereign defaults since the bonds make on average 9% of their assets (Gennaioli, Martin, & Rossi, 2014). This can cause possible financial distress for firms dependent on external financing and thus, be reflected in the stock returns of the firms. The banks can decrease the amount of available loans or increase the interest payments in order to compensate the losses caused by crisis. Similarly, decreased amount of capital reserves of banks can negatively influence the access to capital market for the firms. This was observed in previous crisis, such as Norwegian banking crisis or recent financial crisis in 2008 (Chodorow-Reich, 2014).

In context of this impact, this study examines the effect of the riskiness of banks on the stock returns of associated firms. Considering the main bank of each firm, it is observed the influence of the riskiness of the bank before and after the Sovereign Debt Crisis with the middle point in May 2010, when was provided first bailout to Greece (Lane, 2012). Given the efficient market hypothesis, all publicly known information should be fully reflected in the stock price and thus, there should not be any mispricing or influencers which are not identified. To construct the efficient price, the Fama French three factor model is used as a tool, which has higher accuracy compared to CAPM (1993). Afterwards, additional risk factor is constructed and added to the model. To quantify the risk of the main bank, three different

(5)

measures are used and compared – CDS spread, stock returns of the bank and idiosyncratic stock returns of the bank. The CDS spreads of each bank are insurance policies on the default risk of that individual bank, thus their change should also reflect the riskiness of the bank. Similarly, the stock returns of the banks should reflect all the available information about the bank’s future cash flows, hence, should also reflect the riskiness of the financial institution. Lastly, since the stock returns of the banks are also influenced by systematic risk, the

idiosyncratic returns are isolated to obtain only the firm-specific risk. To obtain the effect of the crisis, new Crisis dummy variable was introduced together with interaction dummies for each regressor. This shows the structural change on the market as a consequence of the crisis, but also the change in the influence of the added risk factor and the market.

The study considered the German stock market DAX and gathered the information about the traded firms from the same market. For each firm, the main bank was identified based on the total financing in the last eight years. Hence, it was assumed that bank with the highest proportion will have the greatest impact on the firm. It was constructed two samples, one for banks with CDS spreads and one for stock returns. These samples are similar and comparable, but given the limitations in identifying the relationships, the samples are not identical. The data was constructed as a panel data for 58 time periods, creating sample size of 5510 and 6032 observations respectively. The values were winsorized to avoid influence of outliers. Given the results of Hausman test and Breusch-Pagan test, regular regression was applied. Subsequently, multiple robustness checks are performed. Firms are divided in bond-holding and not bond-bond-holding, effect of big banks and small banks is separated and lastly, different time period is considered. In the regressions, the significance of the added risk factor is tested. Then, the effect of the crisis is tested by the joint significance of dummy variable and all crisis interaction variables.

The aim of this empirical study is to examine the effect of the riskiness of main bank of each firm to the stock returns of the firm within the context of Sovereign Debt Crisis. The study found that the risk factors are positively correlated with the stock returns as a

compensation for the added risk. Secondly, the CDS spreads are not considered for smaller players and before the crisis given the high correlation of the CDS spreads. Their effect starts to be significant for main banks or after the crisis. The bond holding firms are more

influenced by the CDS spreads. On the other hand, the stock returns of the banks were incorporated in the price already before the crisis. Lastly, the crisis had negative effect on the expected stock returns of the firms, which could partially fix asset mispricing. Those findings

(6)

can be explained by the fact that more influential banks and companies tend to be more analyzed and observed, which causes lower mispricing and incorporation of the risk factors. Moreover, the mispricing occurred after the financial crisis, when the market became more volatile and less predictable. Lastly, the CDS spreads are highly correlated, because they are strongly influenced by macroeconomic factors, thus there explanatory power as a firm-specific risk indicators is limited.

This paper adds to the current literature in multiple perspectives. Firstly, it introduces the idea of new stakeholder who can influence the stock prices. The current literature looks on the supply chain as a potential influencer, but the banking relationships hasn’t been widely used yet (Cohen & Frazzini, 2008). Secondly, it compares multiple measures of estimating the risk of the bank. Previous studies consider mostly a single factor which they believe is the most relevant (Chodorow-Reich, 2014). Thirdly, this study analyses the effects in terms of Sovereign Debt Crisis. Since this crisis happened recently and the consequences are still present, the literature on the effects of the bank-firm relationships is rare. Researchers have analyzed other events, such as Norwegian banking crisis, which happened during the period 1988-1991 (Ongena, Smith, & Michalsen, 2003). Lastly, the theory is applied to German market. The markets in the United Kingdom, the United States or Japan are popular among researchers at the expense of other still highly liquid markets.

Despite the lack of the research in this area as shown above, the research has high relevance to the current society. Firstly to investors, because it points out possible mispricing on the market which could be arbitraged. Moreover, it proves the importance and the

influence of banks on the firms. Therefore, the investors should consider also behavior of banks in terms returns on CDS spreads. Secondly to firms, because it suggests that bond holding firms are more influenced by the banks. At the same time, collaborating with Deutsche Bank and Commerzbank can be more influential on the stock price compared to other banks. Even though these results are approximate, this impact can play an important role while diversifying the risks of the company. Lastly, there is also relevance to policy makers. The results reflect the implications of the bailout policies. Even though the policy supports the moral hazard, it decreases the impact of the crisis on the banks and collaborating firms and consequently, decreases the volatility on the market.

The rest of the paper is organized as follows. Chapter 2 presents background information about the Sovereign debt crisis and the borrower-lender relationships. It also

(7)

describes prior research in the Efficient Market Hypothesis and introduces the Fama French factors and added risk factor. Chapter 3 describes the used models. Chapter 4 summarized which data were used and how they were transformed. Chapter 5 provides main results with additional robustness checks. Finally, Chapter 6 summarizes the findings and makes a conclusion.

(8)

2 Theoretical framework

2.1 Sovereign Debt Crisis and Banks

After the global financial crisis, new sovereign debt crisis emerged in Europe. The high amount of indebtedness of many European governments increased a risk of bankruptcy of states (Arghyrou & Kontonikas, 2012). The situation became severe with the 110-billion euro Greek bailout package, which was ratified in May 2010. At the same time, the European Financial Stability Facility (EFSF) was created with a capital of 440-billion euro. Afterwards, in November 2010 Ireland asked for EFSF emergency assistance, which was also done by Portugal in April 2011. The news about Greece and Greek bailout had contagion effect and the prices of sovereign debt of Portugal, Ireland and Spain react accordingly (Mink & de Haan, 2013).

According to the Bank for International Settlements (2011), there are four possible channels through which the sovereign risk is transferred on financial institutions. Firstly, through asset holdings channel, because the assets of banks can be directly weaken by losses on holdings of sovereign debt. Second is collateral channel. The risk is transferred to banks when the value of collateral, which banks hold, is decreased due to the sovereign debt. Third is a rating channel, which influences banks’ funding possibilities. The downgrade of the country can spread negatively on the banking institutions, because the banks will have less access to the money markets and deposit markets and their funding costs may increase. Lastly, the guarantee channel can increase the risk of banks, because the government might not fulfill their guarantee in case of the failure of the bank and the credibility of too-big-to-fail may not hold. Therefore, considering all those four channels, the increased risk of bankruptcy of the sovereign can create spillovers on the riskiness of banking institutions. Moreover, according to Mink and de Haan, the news about bailout of governments leads to abnormal returns of majority of banks, even though they are not exposed to the bailed state (2013).

Consequently, banks in Europe needed to restructure their capital and liquidity strategies and recreate capital buffers (De Bruyckere, Gerhardt, Schepens, & Vander Venner, 2013). The access to money market was worsened due to the lack of trust. Besides, the European Banking Authority (EBA) performed sovereign stress tests and published result to inform market participants about the riskiness of banks. In general, the sovereign debt crisis had significant negative effect on the health of European banks.

(9)

2.2 Lender Borrower relationships

Especially in Germany, there are strong bank-firm relationships, but the effects of the relationships are in question (Agarwal & Elston, 2001). On one hand, Agarwal and Elston show that stronger relationships reduce agency costs and increase the access to capital. On the other hand, in can also increase the risk and misuse of private information, as was found evidence for the German market. They found significantly higher interest payments to debt ratio for bank-influenced firms, which also shows conflicts of interests between creditors and shareholders.

Besides, the academic debate goes further if there is actually any effect. Greenspan claims that in well-developed economies, the market offers good substitute for distress banks and the final effect on access to financing is thus negligible (1999). The opponents of this idea point out the imperfections of the market, which prevent firms from financing in case of crisis, such as the firms dependent on Continental Bank who suffered significant negative stock returns in 1984 (Slovin, Sushka, & Polonchek, 1993). Similarly, Hoshi and Kashyap blame the disruptions in lending for Japan’s economic decline (1992). Also during the Norwegian banking crisis in 1988, the stocks prices of firms cooperating with distressed banks decreased, even though just by small amount (Ongena, Smith, & Michalsen, 2003). Likewise, during the Lehman bankruptcy, firms maintaining relationship with less healthy lenders had lower probability of obtaining loans and they had higher interests. (Chodorow-Reich, 2014). Therefore, it can be also assumed that influenced banks during the Sovereign Debt Crisis could contribute to changes in returns of firms, with which they cooperate.

2.3 Efficient Market Hypothesis

Eugene Fama proposed in his first papers the idea of efficient markets (1970). He argued that a market in which prices always “fully reflect” available information is called “efficient”. Since this created controversy in which information is meant, more elaborated definition of Efficient Market Hypothesis was proposed by Malkiel (Newman, Milgate, & Eatwell, 1992). He claimed that a market is efficient with respect to given information set It if no investor can make economic profits by trading on the basis of It. This idea suggests that all stocks are fairly valued at all times based on all currently available information. Therefore, if markets are efficient, the prices should reflect all available information, there should not be possibility of trading on the basis of the information. Given the current digital accessibility of information, it can be assumed that asset prices should have at least semistrong form. In other words, the information set It should include history of prices and returns as well as all public information

(10)

(Fama, 1970). Thus, the prices should reflect information about the asset and publicly known stakeholders, who can influence the value of the asset. For publicly traded firms, it is often known with which banks they cooperate, when they need to issue bonds or obtained loans. Therefore, if happens a transfer of the risk of the bank on the firm, the behavior of the bank should influence the returns on the firm, when considering the publicly known behavior or its respective measures.

2.4 Fama French Three-factor model

After implementation of the Modern Portfolio Theory, which describes the demand of financial assets, William Sharpe proposed a theory describing equilibrium in financial

markets, so that the supply equals demand in financial market (1964). Consequently, the idea of Capital Asset Pricing Model (CAPM) emerged, together from works of Jan Mossin (1966) and John Lintner (1965). In this model, there is assumed one source of systematic risk – market risk (Fama & French, 2004). The idiosyncratic risk is not considered since it can be fully diversified away. Hence, it derives anticipated rates of return on assets based on the assets risk levels compared to market risk, which goes along with the theory of security market line (Bodie, Kane, & Marcus, 2011). Given the poor performance of the single factor, multifactor models emerged, from which the three factor Fama French model became widely used for its higher explanatory power (Fama & French, 1993). Also, the model is able to explain many of market anomalies which CAPM had not, such as the value puzzle (Fama & French, 1996).

The Fama French Three Factor model (FF) shows the cross-sectional variation in expected returns with the use of three factors – excess return on the market portfolio, SMB factor and HML factor (Fama & French, 1993). The return on the market portfolio is same factor as in CAPM. It is the return on the market portfolio in excess of the risk free rate or return. It captures common variations in stocks returns and sensitivity of the stocks compared to the market, which can be also considered as macroeconomic approximation. Fama and French created the SMB and HML factors based on six value weighted portfolios sorted on size and book-to-market ratio (Fama & French, 1993).

This study is oriented on the German stock market. Therefore, there were used values which are specific for the Frankfurt Stock Exchange (Brückner R. , Lehmann, Schmidt, & Stehle, 2014). The approximations of the German values of the factors are created in the

(11)

similar way as in the original paper of Fama and French, whereas multiple factor sets available are compared and the most precise is used.

2.5 Additional risk factor

The added value of this study is adding additional factor of a systematic risk to the FF model. As shows above, the behavior of the bank associated to the firm can influence the stock returns of the firm. To quantify the risk and the behavior of the bank, there are used and compared three different measures – return on CDS spread, stocks returns of the bank and idiosyncratic stock returns of the bank.

The credit default swaps (CDS) is an insurance policy on the default risk of a corporate bond or loan (Bodie, Kane, & Marcus, 2011). The CDS purchaser (the insured) pays fee, also known as spread, to the CDS seller (the insurer), which is estimated in percentages of the insured amount. Thus, CDS spread is highly correlated with the firm-specific risk behavior and default risk of the bank or firm, for which the CDS was issued. (Galil, Shapir, Amiram, & Ben-Zion, 2014). Hence, the CDS spread can be used as an approximation of the risk of the bank in the regression as proposed and used in other studies (De Bruyckere, Gerhardt, Schepens, & Vander Venner, 2013). Unfortunately, even though high CDS quotes are mostly driven by single default factor, low CDS quotes tend to by correlated within each and their explanatory power can be misleading (Koziol, Koziol, & Schön, 2014).

Therefore, additional risk measure is introduced, which is suggested by Berndt and Obreja (2010). The equity markets show less comovement and more idiosyncratic risk compared to CDS spreads. Moreover, given the Efficient Market Hypotheses, the stock returns of the banks should also reflect all available information, including risk. As a result, the risk and behavior of the bank can be approximated by the return on stocks of the bank. To be more precise, the study also isolated the idiosyncratic risk in the stock returns by predicting the values by FF model and obtaining residuals. This should eliminate systematic risk which influences the stock returns of the bank and there should remain only firm-specific risk, which creates the approximation of the behavior more precise.

(12)

3 Models

This empirical study uses a multiple regression on panel data in order to investigate the effect of riskiness of main bank of each firm to the stock returns of the firm within the context of Sovereign Debt Crisis. As mentioned above, there are used three different risk measures to estimate the riskiness of the bank – return on CDS spread, return on stocks of the bank and idiosyncratic return on the bank. The returns are calculated in all cases as a percentage change of a price compared to previous last value. The computation can be express by following formula, where RetOfX represents corresponding return and 𝑃₀ and 𝑃₁ old and new prices respectively:

𝑅𝑒𝑡𝑂𝑓𝑋 = 𝑃1− 𝑃0 𝑃0

Consequently, there are also three main regressions to be applied: Regression 1: 𝐸𝑥𝑐𝑅𝑒𝑡𝐹𝑖𝑟𝑚 = 𝛼 + 𝛽1𝐸𝑥𝑐𝑀𝑘𝑡𝑅𝑒𝑡 + 𝛽2𝑆𝑀𝐵 + 𝛽3𝐻𝑀𝐿 + 𝛽4𝑅𝑒𝑡𝐶𝐷𝑆𝑠𝑝𝑟𝑑 + 𝛽5𝐶𝑟𝑖𝑠𝑖𝑠 + 𝛽6𝐶𝑟𝑖𝑠𝑖𝑠 ∗ 𝐸𝑥𝑐𝑀𝑘𝑡𝑅𝑒𝑡 + 𝛽7𝐶𝑟𝑖𝑠𝑖𝑠𝑆𝑀𝐵 + 𝛽8𝐶𝑟𝑖𝑠𝑖𝑠𝐻𝑀𝐿 + 𝛽9𝐶𝑟𝑖𝑠𝑖𝑠𝑅𝑒𝑡𝐶𝐷𝑆𝑠𝑝𝑟𝑑 + 𝜀 Regression 2: 𝐸𝑥𝑐𝑅𝑒𝑡𝐹𝑖𝑟𝑚 = 𝛼 + 𝛽1𝐸𝑥𝑐𝑀𝑘𝑡𝑅𝑒𝑡 + 𝛽2𝑆𝑀𝐵 + 𝛽3𝐻𝑀𝐿 + 𝛽4𝑅𝑒𝑡𝑢𝑟𝑛𝑂𝑓𝐵𝑎𝑛𝑘 + 𝛽5𝐶𝑟𝑖𝑠𝑖𝑠 + 𝛽6𝐶𝑟𝑖𝑠𝑖𝑠 ∗ 𝐸𝑥𝑐𝑀𝑘𝑡𝑅𝑒𝑡 + 𝛽7𝐶𝑟𝑖𝑠𝑖𝑠𝑆𝑀𝐵 + 𝛽8𝐶𝑟𝑖𝑠𝑖𝑠𝐻𝑀𝐿 + 𝛽9𝐶𝑟𝑖𝑠𝑖𝑠𝑅𝑒𝑡𝑢𝑟𝑛𝑂𝑓𝐵𝑎𝑛𝑘 + 𝜀 Regression 3: 𝐸𝑥𝑐𝑅𝑒𝑡𝐹𝑖𝑟𝑚 = 𝛼 + 𝛽1𝐸𝑥𝑐𝑀𝑘𝑡𝑅𝑒𝑡 + 𝛽2𝑆𝑀𝐵 + 𝛽3𝐻𝑀𝐿 + 𝛽4𝐼𝑑𝑖𝑜𝑅𝑒𝑡𝑢𝑟𝑛𝑂𝑓𝐵𝑎𝑛𝑘 + 𝛽5𝐶𝑟𝑖𝑠𝑖𝑠 + 𝛽6𝐶𝑟𝑖𝑠𝑖𝑠 ∗ 𝐸𝑥𝑐𝑀𝑘𝑡𝑅𝑒𝑡 + 𝛽7𝐶𝑟𝑖𝑠𝑖𝑠𝑆𝑀𝐵 + 𝛽8𝐶𝑟𝑖𝑠𝑖𝑠𝐻𝑀𝐿 + 𝛽9𝐶𝑟𝑖𝑠𝑖𝑠𝐼𝑑𝑖𝑜𝑅𝑒𝑡𝑢𝑟𝑛𝑂𝑓𝐵𝑎𝑛𝑘 + 𝜀

The regressions are founded on the FF model with two main modifications. Firstly, there is added new risk factor. Secondly, since we want to examine the effect in context of the Sovereign Debt Crisis, new dummy variable Crisis is inserted together with interaction terms of the dummy with each explanatory variable. This step is to control for the effects of the crisis, since there could be a significant change in the volatility of the factors or in pricing of the market as such. The Crisis dummy variable has value 1 for all dates after the selected date inclusive, which is in this case May 2010, and value 0 for period before May 2010. The dependent variable ExcRetFirm is an excess return on stocks of investigated firms during the

(13)

entire period, which was calculated by subtracting the risk free interest rate from the stock returns. Explanatory variables SMB and HML are Fama French factors, constructed for the German market according to the most precise methodology (Brückner R. , Lehmann, Schmidt, & Stehle, 2014). Lastly, there are added risk factors, RetCDSsprd, ReturnOfBank and IdioReturnOfBank, respectively. RetCDSsprd is a return of CDS spread of the main bank of each firm. ReturnOfBank is the stock return of the main bank of respective firm. Lastly,

IdioReturnOfBank is an idiosyncratic stock return of the main bank of respective firm. The

idiosyncratic part of the return was calculated by regressing each banks stock return with FF factors and isolating the residuals of the regression.

To investigate the effects of the added risk factors before the crisis, if the influence of the risk factor has changed after the crisis and if the risk factor matters as such, following hypotheses are tested respectively for each regression:

𝐻0: 𝛽4= 0 𝐻1: 𝛽4≠ 0

𝐻0: 𝛽9= 0 𝐻1: 𝛽9≠ 0

𝐻0: 𝛽4= 𝛽9= 0 𝐻1: 𝛽4≠ 0 ∨ 𝛽9≠ 0

Secondly, to examine the effects of the crisis, such as if there was change on the market as such, sensitivity towards the market and if the crisis had effect as such, respectively, following hypotheses are tested in each regression:

𝐻0: 𝛽5= 0 𝐻1: 𝛽5≠ 0

𝐻0: 𝛽6= 0 𝐻1: 𝛽6≠ 0

(14)

4 Data

This study considers the German stock market, especially the Frankfurt Stock Exchange (DAX). DAX, as one of the most liquid markets in Germany, was chosen as a representative of an exchange market with Euro currency in order to minimize the currency disturbances and to obtain the effect of the Sovereign Debt Crisis, which had greatest impact in the Euro area (Lane, 2012). Moreover, the German market is characterized by strong firm-bank

relationships which are essential to obtain the effects of banks (Agarwal & Elston, 2001). Moreover, German banks are holding significant amount of sovereign bonds within their assets (Gennaioli, Martin, & Rossi, 2014).

Afterwards, there are used values for end of each month from December 2007 to September 2012. This time period was selected because of limited access to comparable data on CDS spreads before December 2007. Moreover, in middle of the period, in May 2010, is a breakpoint, when the first Greek bailout took place and from when the crisis started to have bigger influence (Lane, 2012). Therefore, there are 58 examined time periods.

Furthermore, the study selected publicly traded firms on DAX in prime standard, which were traded during the examined period. The values of stock returns for each firm were downloaded from Datastream. Moreover, given the limitations when identifying the bank-firm relationships, we obtained two samples. First sample for bank-firms with banks, which have CDS, contains 95 firms. Second sample for firms with banks, which have traded equity, contains 104 firms. The majority of firms in both samples are same, so the samples are approximately comparable, but the difference was necessary in order to keep the sample sufficiently large. Consequently, we obtained samples formed as strongly balanced panel data, with 5510 and 6032 observations respectively. The form of panel data was chosen to

minimize omitted variable bias and to represent the data in most adequate manner.

To approximate the market return, Brückner et al. suggest using composite index of all traded stocks at Frankfurt Stock Exchange, CDAX (2014). This value has high correlation with other approximations, such as 0.93 with DAX30 performance index or 0.98 with DAX share price index. Given the high correlation, suggested measure CDAX was used. Its values were downloaded from the official website of Humboldt University in Berlin and given that the selected firms are from prime standard, TOP segment was selected (Brückner R. , Lehmann, Schmidt, & Stehle, 2014). In addition, Brückner et al. also calculated specific

(15)

values of Fama French factors for the Frankfurt Stock Exchange and investigated the most efficient ones, which were then used in this study.

The risk free rate is approximated by money market rate of one month. This value is highly liquid and represents the risk free rate most precisely (Brückner R. , Lehmann, Schmidt, & Stehle, 2014). At the same time, this value has high correlation with other used approximations. For example, the correlation with 3-months treasury bills of German Government is 0.98. The values were provided in yearly percentages, thus, to obtain excess returns of firms or of market, the value of risk free rate was multiplied by 1/12 and then subtracted from respective monthly returns.

To estimate the risk factor, firstly the CDS spread was used. The study selected mid values of 5 year senior MM CDS spreads denominated in Euros, because they are considered to be the most liquid segments of the market (Annaert, De Ceuster, Van Roy, & Vespro, 2013). The data were downloaded from the Reuters Thomson Datastream database. Furthermore, the stock returns of the banks were also downloaded from Datastream.

The vital part of the data collection was in identifying of the bank-firm relationships. For each firm, the values of publicly known loans, bonds and equity were downloaded from the Thomson One database. Afterwards, there was selected one bank which had the highest cumulative participation in terms of the total amount in the last eight years. This study

therefore assumes that this bank can have the highest influence on the firm and if the firm has need of financing, there exists high chance it might contact the same bank (Chodorow-Reich, 2014). When there were banks with equal amount of financing, the book runner was chosen. During the research, the sample of purely loans relationship was also considered, but because the sample would be too small and not fully representative, only the option of total financing was selected. Moreover, the market of purely syndicate loans is highly dominated by two banks, Deutsche Bank and Commerzbank, which would create bias in the estimations. For total financing, the participation is around 52%, which allows for variation of other banks. Moreover, the bank-firm relationships are also vital not only where there are syndicate loans, but also when firms are issuing bonds or equity. Previous studies found that bank

relationships have positive and significant effects on a firm’s underwriter choice, over and above their effects on fees (Yasuda, 2005). Also, banks with stronger bank firm relationships are more likely to issue their bonds sooner (Hale & Santos, 2008). Same authors suggest that stronger relationships decrease the information frictions between the entities. Investment

(16)

banks use often this opportunity to motivate firms to enter the public bond market. Similarly, strong main bank relationships give issuers increased access to equity capital market

(Kutsuna, Smith, & Smith, 2007). Considering all those influences, it can be assumed that considering total financing measure gives sufficient and representative approximation of the bank-firm relationship.

(17)

5 Results

5.1 Descriptive statistics

Table 1 in Appendix summarizes descriptive statistics of both samples. In the first sample, it

was possible to identify bank-firm relationships and collect data for 95 firms, each for 58 periods, which creates 5510 observations in total. The sample is formed as strongly balanced panel data. The median value of the monthly risk free rate is 0.0725, which is relatively low, because of the monetary policy of ECB in order to support the economy (Roman & Bilan, 2012). Secondly, the mean value of the market value is very close to zero and the standard deviation is high, because of the impact of the financial crisis in 2008 and sovereign debt crisis in 2010, which destabilized markets (Lane, 2012). It can be also noticed that those events caused sharp decrease and prices slowly increased afterwards, because the median of monthly market returns is 0.739, which is much bigger, compared to mean 0.019.

In order to avoid the misleading influence of outliers, the values of excess return on firms and return on CDS spread were winsorized. The bottom 1% of values were replaced by the 1st percentile value and the values of the 99th percentile were replaced by the value at 99th percentile. The explanatory power of equations slightly increased, but the total effect is marginal, so only winsorized values are used and shown. The mean value of returns on firms is higher than the mean of market return, which goes along with the theory of mean-variance utility, because the standard deviation is also higher (Markowitz, 1952). Even though the market consists of the same firms, those values are not identical, because multiple traded firms were omitted from the sample since their bank-firm relationship wasn’t possible to identify.

The returns on CDS spread have high volatility given the unstable environment during crisis. In general, the CDS spreads reflect market sentiments about the financial health of institutions (Annaert, De Ceuster, Van Roy, & Vespro, 2013). Given the high volatility, the average return is also higher, but also because of increasing default risk of banks during the crisis. Unfortunately, the CDS spread returns in the sample are highly correlated as can be shown in Figure 1. This phenomenon is also observed by other researchers. During the period of the financial crisis, correlated default factors accounted for about 80% of the default risk, after the crisis above 50% (Koziol, Koziol, & Schön, 2014). Therefore, even though the CDS spread should reflect individual default probability of the institution, the interdependence is significant and macroeconomic factors do influence the values more than individual risks.

(18)

Figure 1 - Return on CDS spreads over time for individual banks

In the sample, the actual correlation of CDS spreads with market returns is -0.42, which shows the similar influence of the macroeconomic factors and comovement. Moreover, the correlation is negative, which proves the theory that when there are unfavorable

conditions, the returns on market decrease and at the same time, the riskiness of the banks in terms of the CDS spreads increases. Similarly as in the previous studies, the correlation in the sample increased during the period of a crisis, more specifically to -0.49 (Koziol, Koziol, & Schön, 2014). Furthermore, to estimate serial correlation, it was used test proposed by Wooldridge (2002). The null hypothesis proposes that there is no first-order autocorrelation. Nevertheless, the F-value 47.91 suggests very significant rejection of this hypothesis, which proves the correlation of CDS spreads over time and thus, their minimal explanatory value as firm-specific risk indicators. If we perform same test on regression with stock returns of the banks, the F-value is only 2.961 which makes it much less significant. Lastly, we can observe differences in between effects of each added risk factor. Testing the regressions in section 3, the between effect of RetCDSsprd has coefficient 0.045, which corresponds to probability of tvalue of 0.601. When same test is performed with stock returns of banks, the coefficient is -0.348, which corresponds to probability of t-value of 0.06. Therefore, we can see the lost significance of explanatory power of RetCDSsprd as firm-specific risk estimator. In Figure 2, we can observe that stock return on banks tend to be less correlated and there is less

(19)

Figure 2 - Return on banks stocks over time for individual banks

Table 1 in Appendix also summarizes descriptive statistics of the second sample. This

sample is for 104 firms during the same 58 periods, so there are 6032 observations. The majority of the firms are same as in the first sample, but there are few different in order to keep the sample large and representative. Therefore, the comparison can be approximate. Since it is the same time period, the values of risk free rate, market return, SMB and HML are identical in both samples. The values of returns on firms, returns on banks and idiosyncratic returns on banks are also winsorized and similar logic as in the first sample is applied. Given this adjustment, the average of idiosyncratic returns is not zero as the definition of residuals in regression claims. During the crisis, majority of firms were influenced negatively, so the stock returns of firms are lower than the German average monthly return in last 40 years (Artmann, Finter, & Kempf, 2012). Furthermore, the average stock returns on bank are negative, since the banks were main victims of the crisis (Lane, 2012). Also, it can be shown the increasing risk of the banks over the time, because the CDS spreads in table 1 are positive, which decreases the stock returns on banks.

5.2 Main results

The data were formed as panel data. In order to use the adequate regression, Hausman test was performed. The Hausman test distinguishes between the fixed and random effects in panel data. Theoretically, the fixed effects could occur since there are multiple time periods and the dependent variables could be time variant (Stock & Watson, 2012). In all cases, the probability of χ2

was very close to 1, so the hypothesis that we can use random effects cannot be rejected by far. Consequently, Breusch-Pagan test was performed. The null hypothesis in the B-P test of independence is that the residuals across entities are not correlated, so regular regression can be applied. In all samples, the χ2 probability was also very close to 1. Thus, in

(20)

all regressions, regular regression was applied without considering fixed or random effects of panel data.

Table 2 summarizes beta coefficients of CAPM regression, Fama French regression

and regression with extra risk factor CDS spreads respectively. In general, the explanatory powers of regressions (adjusted R2) are relatively low – between 0.28 and 0.31. This low value can be accounted to the unstable and unpredictable environment during the crisis. Secondly, the Fama French factors have in general lower explanatory power for the German market (Artmann, Finter, & Kempf, 2012). Lastly, the CAPM and Fama French regressions are mostly designed to predict long-term expected returns and their efficiency can be judged on longer period. Since this study uses 58 months and during unstable economic conditions, the lower explanatory power is justified.

By introducing Fama French factors, adjusted R2 increased, which goes along with the theory of Fama and French (Fama & French, 1993). On the other hand, this increase was not so significant, only from 0.289 to 0.308, which can be explained by lower efficiency of the factors on the German market (Artmann, Finter, & Kempf, 2012). By including the additional risk factor return on CDS spread, the adjusted R2 marginally increased. At the same time, the constant slightly decreased, but its significance decreased. Therefore, it can be concluded that the added risk factor does not seem to help in predicting expected returns of the firms. This can be contributed to the high correlation of CDS returns, as shown above, which are more influenced by same macroeconomic factors rather than individual default probabilities. At the same time, similar economic factors are already captured by market return variable, thus by including CDS spreads, the effect on the regression is marginal.

Consequently, the coefficient of Return on CDS spreads (RetCDSsprd) is not significant and the null hypothesis cannot be rejected. The value is slightly positive, which suggests positive correlation between the firm returns and returns on CDS spreads. This is consistent with the theory of CDS spreads, because it should predict higher default risk, which should be compensated by higher returns (Annaert, De Ceuster, Van Roy, & Vespro, 2013). The value is marginally close to zero, which can be explained by the correlation of CDS spreads and influence of macroeconomic factors, which are already included in market returns (Koziol, Koziol, & Schön, 2014). On the other hand, during crisis, the return on CDS spreads (CrisisRetCDSsprd) is negative and statistically significant with 10%. Considering that all crisis factors are negative and the comovement with market returns, the added risk factor

(21)

predicts lower returns after the crisis, consistently with other factors. Moreover, the increase in significance within the crisis period shows that market started to reflect the sovereign risk via CDS spread, as was also found in other countries (Beirne & Fratzscher, 2013). The marginal effect of CDS spreads can be also seen on the F-test on all added risk factors (with and without crisis dummy). The joint contribution cannot be rejected even with 20%

significance. Even though the total influence of CDS spreads in the regression is not significant, the crisis coefficient of RetCDSsprd is greater than the factor without the crisis dummy, which makes it negative at the end. This can be firstly attributed to marginally small value of the factor before the crisis, because the absolute value created during crisis is similar as for other risk factors during crisis as is shown later. Secondly, it points out again the lack of incorporation of CDS spreads, because the higher risk given by CDS spreads, the lower are the returns. In other words, the macroeconomic factors, such as lower profits on the market and higher bankruptcy rate, which increase the CDS spreads, decrease the returns of the firms at the same time. It can be concluded that the CDS spreads on the banks are not reflected in the stock returns of firms since the CDS spreads do not fully reflect the financial health of the individual institutions.

Contrary to CDS spreads, the crisis had influence on the returns. Crisis dummy and all crisis interaction dummies have negative coefficient. In other words, having same dependent variables, the expected return on firms is smaller after May 2010. This impact is statistically significant, because the F-test of joint influence of all crisis variables shows 5.97%

probability, so the hypothesis can be rejected. The value of crisis dummy is very close to the value of a constant, so during the crisis, the possible alpha is very close to zero. This can be explained by correction of mispricing, which was created before the crisis, because when the equity is priced correctly, the alpha should be zero (Bodie, Kane, & Marcus, 2011). In addition, the market return coefficient decreases significantly during the crisis. The beta during the crisis is close to 1, which should be in a portfolio constructed from majority of firms on the market. Thus, it can be attributed to realization of the market about the

mispricing. Moreover, higher beta coefficient shows higher expected returns and sensitivity towards the returns given the same conditions (Berk & DeMarzo, 2011). Hence, if there are high possible future cash flows, as can be seen in tech-firms, the betas are higher. When an event changes the expectations of possible cash flows, this can be reflected in the change of coefficient. Consequently, given the multiple bankrupts of firms and risk of bankruptcy of countries with the same currency, the realization of this threat decreased the expectations in

(22)

equity returns, which is shown in negative coefficients of crisis factors. To sum up, the crisis had significant negative influence on the asset pricing because of the mispricing and

worsening of market sentiment and economic conditions during the crisis.

Table 3 shows the regressions on the second sample. In this case, there are two

possible risk factors – stock returns on banks and idiosyncratic stock return on banks. The change of explanatory power, the effect of Fama French factors and the effect of the crisis are similar to the previous sample and thus also their interpretations. Contrasting to the minimal effect of returns on CDS spread, the returns on banks’ stocks do contribute to the firms’ returns. Overall, the effect of all returns and idiosyncratic returns is very similar. Isolating idiosyncratic part of the variations contributes to slightly decreasing the constant and changes the coefficients, but the effect is minimal and the effects can be generalized for both risk measures. The main result is in the high statistically significant influence of the stock returns of the associated bank on the firms’ returns. In the sample, increase in the stocks returns of the banks by 1% causes an increase of 0.05% in expected returns of the associated firm. The positive correlation shows that when banks become riskier, their shareholders as well as shareholders of the firms need to be compensated by higher returns, given the mean-variance utility. The increased significance can be attributed to lower comovement on the equity markets, so the changes actually reflect the health of the institution (Berndt & Obreja, 2010). In addition, CDS spreads tend to be less precisely identified on the market than equity stocks, so their changes are not fully adequate. The magnitude of the coefficient reflects the amount of transferred risk of the institution to the firm (Ongena, Smith, & Michalsen, 2003). The factor is not higher given the multiple bank-firm relationships majority of firms have other possibilities of financing. The factor during the crisis is also negative, but not significant as was case of previous sample. This supports the theory that market had already reflected the risk via stock returns before the crisis adequately.

To recapitulate, CDS spreads do not seem to be considered as a risk factor before the crisis compared to stock returns of the banks, which contribute significantly. After the crisis, the effect of CDS spread increases and their information starts to be incorporated in the stock price of firms, contrasting to equity returns of banks, which haven’t changed. The crisis has negative influence on the expected returns of firms. Similar results were already found in previous studies, when stock returns contributed, even though by just small amount, to explaining expected stock returns (Ongena, Smith, & Michalsen, 2003). It absolute terms, it could be thought that CDS spreads have opposite effect on the firm returns than stock returns

(23)

of the banks. However, this is result of the marginally insignificant value of the coefficient before the crisis, because the measure was not considered by investors. During the crisis, the CDS spreads started to be considered and thus, the effect on them became similar as on other factors. Still, there remained high influence of macroeconomic factors which created this anomaly. It is also important to mention that the sample is not large enough to take the results fundamental.

5.3 Additional Robustness Checks

In order to validate the results, multiple robustness checks were performed. Firstly, firms were divided in bond holding and non-bond holding. Secondly, firms were divided by cooperating with two big banks, Deutsche Bank and Commerzbank, and cooperating by other banks. Lastly, different time period was selected – 29 periods before and after the collapse of

Lehman Brothers in September 2008. First two robustness checks were performed in order to identify groups for which the effect can be different and possible logical reasoning of the difference. The last robustness check verifies if the reasoning can be applied in different time period and other influential event.

Table 4 shows betas for sample with returns on CDS spreads, which was divided if the

firm had issued bonds within the last 8 years. In the sample, 56 firms have issued bonds and 39 didn’t, resulting in sample size 3248 and 2262 observations respectively. The time period remained same as during previous tests. As mentioned before, winsorized sample was used. Surprisingly, the firms with bonds, which could be less depended on loans and single-bank relationships, are significantly more influenced by CDS spreads as well as by crisis compared to firms without bonds. The joint influence of CDS spreads is significant with probability of 3% for firms with bonds compared to 90% for firms without bonds. Therefore, the null hypotheses can be rejected for firms with bonds, but it cannot be rejected for firms without bonds. This result suggests that the manager’s entrenchment incentive occurs. (Hideaki & Yasuhiro, 1999). When firms have strong firm-bank relationships, the firms are much more likely to issue public bonds than to borrow (Hardin & Wu, 2010). This difference supports also the idea that the CDS spreads are not so much observed. It shows that bigger firms, which have mostly issued bonds, are more analyzed and thus, CDS spreads of their banks are incorporated in the firms. Smaller firms tend to be less analyzed and thus, the incorporating of banks CDS spreads in not present. Therefore, firms with bonds tend to be more sensitive to changes in CDS spreads. The signs of the coefficients are same as in previous analysis. For firms without bonds, the signs are opposite, but the values are very marginal and highly not

(24)

significant, so the values can be disregarded. Similarly, firms with strong bank-firm

relationships tend to be more sensitive to the crisis as being more influenced by it compared to firms without bonds, which seem to be more independent and stable. In addition, firms with bonds have higher explanatory power than firms without bonds. Firms with bonds are mostly bigger, which makes them more observed, their information more incorporated and thus, more predictable (Reinganum, 1981). It is important to note that both groups use approximately same kind of lenders. In other words, the distribution of less and more influential banks is approximately similar given the small sample.

Table 5 shows same division for the sample with banks’ equity. Same as in previous

case, the explanatory power increased. The impact on the influence of the risk factor is similar as in table 4, but it is not so strong. This again shows that smaller firms are analyzed less but compared to the CDS spreads, the banks’ stock returns are incorporated in the firms’ stock prices. This idea is supported also by values of constants. In both tables 4 and 5, there is substantial different between firms with bonds and without, in all three regressions. As bigger firms are more observed, possible alphas are arbitraged away (Bodie, Kane, & Marcus, 2011). Similarly, smaller firms without bonds are less traded and their stocks need to be underpriced in order in attract investors. The crisis’s effects are similar as previously, but smaller.

Similarly as in main results, the returns on banks’ stocks are mostly incorporated and the prices, for both type of firms, and the crisis had negligible effect. It can be concluded that firms with bonds are more sensitive to changes in their banks’ riskiness as well as they were more influenced by the crisis. Also, smaller firms show significant premium. CDS spreads are not incorporated in the prices of smaller firms at all, stocks returns partially more.

The second robustness check is summarized in Table 6. The sample of firms with CDS spreads was divided in firms cooperating with Deutsche Bank and Commerzbank and

cooperating with other banks. Those two banks are influential in the German market and in the tested sample, they are connected to 49 out of 95 firms as the main cooperating bank. The main difference in the groups is the influence of the crisis factors. Firms cooperating with the two big banks are much less influenced by the crisis than firms cooperating with other banks. This result can be interpreted as an impact of possible bailouts. Deutsche Bank and

Commerzbank, as dominant players on the market, are more likely to receive a bailout during the crisis compared to other banks, because their collapse would be catastrophic for the German economy (Mattana, Petroni, & Sonia, 2015). Therefore, firms cooperating with those banks are aware of the possible policy and they tend to underestimate the impact of the crisis,

(25)

even when there are tendencies to re-regulate the market (Abreu & Gulamhussen, 2013). On the other hand, the returns on CDS spreads are more significant for firms cooperating with those two big banks. This supports the previous claim that market did not incorporate CDS spreads of smaller firms or smaller banks, especially before the crisis. The big players are more analyzed and therefore, their risk in terms of CDS spreads was considered also before the crisis. The event showed the investors that also other banks matter and their CDS spreads started to be incorporated. These results are consistent with explanations provided for main results. The rest of the factors do not show significant difference between the two groups.

The last robustness check is shown in Table 7. Using the sample of returns on banks’ stocks, the time period was changed. As the main event, it was selected the bankruptcy of Lehman Brothers in September 2008. Then, same time distance was chosen – 29 periods before and after the event. The period after the event is selected as Crisis period, which is summarized with crisis interactions dummies and crisis dummy variable. Therefore, the crisis period is approximately similar to the pre-crisis period of main results. Firstly, the risk factor of Return of the bank doesn’t seem to be considered before the crisis, but its influence greatly increased afterwards. This would suggest that investors before the first crisis in 2008 have not considered risks of the banks at all, but thanks to the Lehman bankruptcy, they started to observe the stock returns to approximately estimate the risk. Secondly, the effect of the crisis is significant and positive. Given that this crisis period is same as pre-crisis period in previous sample when the pre-crisis factors were positive, the results tend to be consistent. Moreover, the data show mispricing which occurred in the time period 2008 and 2010. The constant is marginally close to zero, but the crisis dummy is significant, positive and large. In the main results, this alpha is corrected after 2010. Therefore, the results consistently suggest that in this period, the asset pricing was not correct and majority of the assets were underpriced (Driessen & Van Hemert, 2012). This again points out the pessimistic environment after the first crisis.

The robustness checks have supported the results found in the main section. The CDS spreads of smaller players (firms without bonds or not dominant banks) on the market were less analyzed before the crisis, but it has been corrected afterwards. Moreover, the returns on banks’ stocks were already analyzed before the crisis in 2010 and their influence remained constant, but there incorporation happened after 2008 crisis, when the market realized the risk exists.

(26)

6 Conclusion

This study focuses on the effects of the riskiness of main bank of each firm to the stock returns of the firm within the context of Sovereign Debt Crisis. It builds on the Fama French model and adds additional risk factor – riskiness of the bank. This is measured by returns on CDS spreads, stocks returns and idiosyncratic returns on the respective bank. The study found that the CDS spreads are not reflected in the stock returns before the crisis, especially since the spreads are highly correlated and do not reflect the riskiness of the bank. The CDS spreads started to be influence for bigger players, such as firms with bonds or firms collaborating with bigger banks. After the crisis, the impact of the CDS spreads increased to remaining firms. On the other hand, the stock returns of the banks were influence also before the crisis and for firms with and without bonds. The changes in the risk factors’ measures are positively correlated with the stock returns in all significant results. In addition, the crisis might have corrected possible mispricing and decreased expected returns and alphas, holding everything else constant. The results show that smaller firms tend to be less analyzed compared to bigger players, which can lead to possible mispricing or not including all possible risks.

There are multiple limitations to this study. Firstly, it was possible to identify only part of the publically traded firms, which can create selection bias in the results. Thus, it would be advisable to select market with higher possible coverage of total financing of firms in order to increase the sample size and its representativeness. Secondly, the study assumed the risk factor of the bank as the only systematic risk influencing the firm, which creates omitted variable bias. Even though panel data was selected to minimize the bias, it would be

suggested to include also changes in other stakeholders, such as main costumers or suppliers in order to capture other systematic risks. Thirdly, the Fama French factors have in general lower explanatory power for the German market. Therefore, it can be also included

momentum factor of Carhart. Even better job would do this 4-factor model where SMB factor is substituted earnings-to-price ratio (Artmann, Finter, & Kempf, 2012). Another assumption of this study is in the bank-firm relationship. It was assumed that the bank with highest amount in total financing in last 8 years had the greatest impact. Therefore, it was disregarded if the loans were sold during the period or which bank had the biggest influence in the recent years. The main assumption was that the firm in need of financing would choose again the bank with highest contribution. To overcome this assumption as least partially, multiple banks would need to be included in the regression weighted by their contribution. Still, the past participation does not guarantee future cooperation. Lastly, the effect of reverse causality was

(27)

disregarded. It was assumed that banks are big enough to influence firms and firms are small enough to not have influence on the banks. In reality, higher risk of the firms can be

transferred on banks that hold their assets. Therefore, it would be suggested for further research to take a longer period with more precise sample in order to obtain stronger results. Also, combination of risk factors of multiple stakeholders would be highly beneficial.

(28)

Bibliography

Abreu, J., & Gulamhussen, M. (2013). The stock market reaction to the public announcement of a supranational list of too-big-to-fail banks during the financial crisis. Journal of International

Financial Markets, Institutions and Money, 25, 49-72.

Agarwal, R., & Elston, J. (2001). Bank-firm relationships, financing and firm performance in Germany. Economics Letters, 72, 225-232.

Annaert, J., De Ceuster, M., Van Roy, P., & Vespro, C. (2013). What determines Euro area bank CDS spreads? Journal of International Money and Finance, 32, 444-461.

Arghyrou, M., & Kontonikas, A. (2012). The EMU sovereign-debt crisis: Fundamentals, expectations and contagion. Journal of International Financial Markets, Institutions & Money, 22, 658-677.

Artmann, S., Finter, P., & Kempf, A. (2012). Determinants of Expected Stock Returns: Large Sample Evidence from the German Market. Journal of Business Finance & Accounting, 39(5-6), 758-784.

Beirne, J., & Fratzscher, M. (2013). The pricing of sovereign risk and contagion during the European sovereign debt crisis. Journal of International Money and Finance, 34, 60-82.

Berk, J., & DeMarzo, P. (2011). Corporate Finance (2nd ed.). Prentice Hall.

Bernake, B. (1983). Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression. The American Economic Review, 73(3), 257-276.

Berndt, A., & Obreja, I. (2010). Decomposing European CDS Returns. Review of Finance, 14, 189-233.

Bodie, Z., Kane, A., & Marcus, A. (2011). Investment and Portfolio Management (9th ed.). New York: McGraw-Hill.

Brückner, R., Lehmann, P., Schmidt, M., & Stehle, R. (2014, 12 5). Fama/French Factors for

Germany. Retrieved from Humboldt-Universität zu Berlin:

https://www.wiwi.hu- berlin.de/professuren/bwl/bb/data/fama-french-factors-germany/fama-french-factors-for-germany

Brückner, R., Lehmann, P., Schmidt, M., & Stehle, R. (2014, February 3). Fama/French Factors for Germany: Which Set is Best? Retrieved from SSRN: http://dx.doi.org/10.2139/ssrn.2390063 Chodorow-Reich, G. (2014). The employment effects of credit market disruptions: firm-level evidence

from the 2008-9 financial crisis. The Quarterly Journal of Economics, 129(1), 1-59.

Cohen, L., & Frazzini, A. (2008). Economic Links and Predictable Returns. The Journal of Finance,

63(4), 1977-2011.

Committee on the Global Financial System. (2011). The impact of sovereign credit risk on bank

funding conditions. Basel: Bank for International Settlements.

De Bruyckere, V., Gerhardt, M., Schepens, G., & Vander Venner, R. (2013). Bank/sovereign risk spollovers in the European debt crisis. Journal of Banking and Finance, 33(12), 4793-4809.

(29)

Driessen, J., & Van Hemert, O. (2012). Pricing of commercial real estate securities during the 2007– 2009 financial crisis. Journal of Financial Economics, 105(1), 37-61.

Fama, E. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work. The Journal of

Finance, 25(2), 383-417.

Fama, E., & French, K. (1993). Common risk factors in the returns on stocks and bonds. Journal of

Financial Economics, 33, 55-84.

Fama, E., & French, K. (1996). Multifactor explanations of assets pricing anomalies. Journal of

Finance, 51, 55-84.

Fama, E., & French, K. (2004). The Capital Asset Pricing Model: Theory and Evidence. Journal of

Economic Perspectives, 18(3), 25-46.

Fama, E., & French, K. (2014, December 29). Description of Fama/French Factors. Retrieved from K. R. French personal website:

http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/Data_Library/f-f_factors.html Flannery, M., & Protopapadakis, A. (2002). Macroeconomic Factors Do Influence Aggregate Stock

Returns. The Review of Financial Studies, 15(3), 751-782.

Galil, K., Shapir, O., Amiram, D., & Ben-Zion, U. (2014). The determinants of CDS spreads. Journal

of Banking & Finance, 41, 271-282.

Gennaioli, N., Martin, A., & Rossi, S. (2014). Banks, Government Bonds, and Default: What do the Data Say? Retrieved from https://www.imf.org/external/pubs/cat/longres.aspx?sk=41731.0 Greenspan, A. (1999, September 27). Lessons from the Global Crisis. Retrieved from The Federal

Reseve Board: http://www.federalreserve.gov/newsevents/speech/1999speech.htm Hale, G., & Santos, J. (2008). The decision to first enter the public bond market: The role of firm

reputation, funding choices, and bank relationships. Journal of Banking & Finance, 32(9), 1928-1940.

Hardin, W., & Wu, Z. (2010). Banking Relationships and REIT Capital Structure. Real Estate

Economics, 38(2), 257-284.

Hideaki, M., & Yasuhiro, A. (1999). Relationship Banking and Debt Choice: Evidence from the

Liberalization in Japan. Tokyo: Weseda University Institute of Finance.

Hoshi, T., & Kashyap, A. (1992). The Japanese banking crisis: where did it come from and how will it

end? Cambridge: National Bureau of Economic Research.

Koziol, C., Koziol, P., & Schön, T. (2014). Do correlated defaults matter for CDS premia? An

empirical analysis. Frankfurt am Main: Deutsche Bundesbank.

Kutsuna, K., Smith, J., & Smith, R. (2007). Banking relationships and access to equity capital markets: Evidence from Japan’s main bank system. Journal of Banking & Finance, 31(2), 335-360. Lane, P. (2012). The European Sovereign Debt Crisis. Journal of Economic Perspectives, 26(3),

(30)

Lintner, J. (1965). The Valuation of Risk Assets and the Selection of Risky Investments in Stock Portfolios and Capital Budgets. The Review of Economics and Statistics, 47(1), 13-37. Markowitz, H. (1952). Portfolio Selection. The Journal of Finance, 7(1), 77-91.

Mattana, P., Petroni, F., & Sonia, S. (2015). A test for the too-big-to-fail hypothesis for European banks during the financial crisis. Applied Economics, 47(4), 319-332.

Mink, M., & de Haan, J. (2013). Contagion during the Greek sovereign debt crisis. Journal of

International Money and Finance, 34, 102-113.

Mossin, J. (1966). Equilibrium in a Capital Asset Market. Econometrica, 34(4), 768-783.

Newman, P., Milgate, M., & Eatwell, J. (1992). New Palgrave Dictionary of Money and Finance. London: Macmillan.

Ongena, S., Smith, D., & Michalsen, D. (2003). Firms and their distressed banks: lessons from the Norwegian banking crisis. Journal of Financial Economics, 67, 81-112.

Reinganum, M. (1981). Abnormal Returns in Small Firm Portfolios. Financial Analysts Journal,

37(2), 52-56.

Roman, A., & Bilan, I. (2012). The Euro Area Sovereign Debt Crisis and the Role of ECB's Monetary Policy. Procedia Economics and Finance, 3, 763-768.

Sharpe, W. (1964). Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk.

The Journal of Finance, 19(3), 425-442.

Slovin, M., Sushka, M., & Polonchek, J. (1993). The Value of Bank Durability: Borrowers as Bank Stakeholders. The Journal of Finance, 48(1), 247-266.

Stock, J., & Watson, M. (2012). Introduction to Econometrics (3rd ed.). Pearson Education.

Wooldridge, J. (2002). Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.

Yasuda, A. (2005). Do Bank Relationships Affect the Firm's Underwriter Choice in the Corporate-BondUnderwriting Market? The Journal of Finance, 60(3), 1259-1292.

(31)

Appendix

Table 1. Summary statistics of winsorized CDS spread sample and winsorized banks’ stocks returns sample

Obs Mean Std. Dev. Min Max

RiskFree 5510 0.1271034 0.125531 .01 .398 RetMkt 5510 0.0193276 6.130931 -16.929 14.254 SMB 5510 -0.2239655 3.516154 -7.904 8.218 HML 5510 0.8028793 3.423304 -6.317 15.663 RetFirm1 5510 0.2740103 12.33269 -57.143 124.373 RetCDSsprd 5510 4.167655 22.76356 -42.57426 76.21608 ExcMktRet 5510 -0.1077759 6.178139 -17.04 14.17 ExcRetFirm1 5510 0.0875379 11.39146 -32.379 35.934 RetFirm2 6032 0.3628148 12.90746 -57.143 185.641 ReturnOfBank 6032 -1.429477 14.68978 -46.352 48.825 IdioReturnOfBank 6032 -0.0599271 10.76638 -27.44669 41.35911 ExcMktRet 6032 -0.1077759 6.17809 -17.04 14.17 ExcRetFirm2 6032 0.1351638 11.65627 -32.715 36.615 25% 50% 75% RiskFree 0.034 0.0725 0.134 RetMkt -2.919 0.739 3.568 SMB -2.512 0.36 2.537 HML -0.726 0.6915 2.422 RetFirm1 -6.029 0.113 6.491 RetCDSsprd -10.078 0 14.993 ExcMktRet -3.033 0.5925 3.502 ExcRetFirm1 -6.155 -0.0045 6.312 RetFirm2 -6.182 0.018 6.654 ReturnOfBank -9.92 -2.073 6.069 IdioReturnOfBank -6.252 -0.627 5.434 ExcMktRet -3.033 0.5925 3.502 ExcRetFirm2 -6.283 -0.09 6.549

(32)

Table 2. Regression with CDS spreads returns sample (1) CAPM regression

(2) Fama French regression

(3) Main regression with CDS spreads returns of banks

(1) (2) (3)

ExcRetFirm ExcRetFirm ExcRetFirm ExcMktRet 1.063*** 1.213*** 1.223*** (39.24) (39.76) (38.00) Crisis -0.675*** -0.430 -0.337 (-2.59) (-1.62) (-1.24) CrisisMktRet -0.174*** -0.131** -0.195*** (-4.01) (-2.08) (-2.65) SMB 0.582*** 0.581*** (10.34) (10.31) HML 0.230*** 0.221*** (5.04) (4.73) CrisisSMB -0.184* -0.228** (-1.75) (-2.08) CrisisHML -0.0987 -0.105 (-1.12) (-1.17) RetCDSsprd 0.00768 (1.00) CrisisRetCDSsprd -0.0248* (-1.78) Constant 0.607*** 0.395** 0.378* (3.29) (2.05) (1.95) N 5510 5510 5510 adj. R-sq 0.289 0.308 0.309 t statistics in parentheses * p<0.10, ** p<0.05, *** p<0.01

F-test of all Crisis factors – regression (3): F-test of all CDS spreads – regression (3):

F(5, 5500) = 2.12 F(2, 5500) = 1.58

(33)

Table 3. Regression with Stock returns sample (1) CAPM regession

(2) Fama French regression

(3) Main regression with Stock Returns of banks (4) Main regression with Idiosyncratic returns of banks

(1) (2) (3) (4)

ExcRetFirm ExcRetFirm ExcRetFirm ExcRetFirm

ExcMktRet 1.038*** 1.200*** 1.113*** 1.197*** (38.63) (39.66) (29.57) (39.61) Crisis -0.513** -0.269 -0.193 -0.178 (-1.98) (-1.02) (-0.72) (-0.67) CrisisMktRet -0.149*** -0.119* -0.0998 -0.118* (-3.45) (-1.90) (-1.32) (-1.89) SMB 0.627*** 0.624*** 0.653*** (11.22) (11.18) (11.61) HML 0.233*** 0.225*** 0.229*** (5.14) (4.97) (5.05) CrisisSMB -0.230** -0.274** -0.283*** (-2.21) (-2.56) (-2.69) CrisisHML -0.108 -0.0971 -0.0962 (-1.23) (-1.11) (-1.10) ReturnOfBank 0.0514*** (3.86) CrisisReturnOfBank -0.0118 (-0.48) IdioReturnOfBank 0.0533*** (3.70) CrisisIdioReturnOfBank -0.00961 (-0.37) Constant 0.561*** 0.352* 0.356* 0.298 (3.06) (1.84) (1.86) (1.56) N 6032 6032 6032 6032 adj. R-sq 0.268 0.289 0.291 0.291 t statistics in parentheses * p<0.10, ** p<0.05, *** p<0.01

(34)

F-test of all Crisis factors: (3) F(5, 6022) = 1.62 Prob > F = 0.1504 (4) F(5, 6022) = 1.62 Prob > F = 0.1521 F-test of all returns on banks: (3) F(2, 6022) = 9.26 Prob > F = 0.0001 (4) F(2, 6022) = 8.85 Prob > F = 0.0001

(35)

Table 4 – Regressions divided by bond holders

(1) Regression with firms with bonds using CDS returns (2) Regression with firms without bonds using CDS returns

(1) (2) ExcRetFirm ExcRetFirm ExcMktRet 1.275*** 1.148*** (31.66) (21.91) SMB 0.413*** 0.818*** (5.86) (8.95) HML 0.304*** 0.102 (5.21) (1.34) Crisis -0.258 -0.453 (-0.76) (-1.03) CrisisMktRet -0.228** -0.147 (-2.46) (-1.25) CrisisSMB -0.234* -0.215 (-1.70) (-1.21) CrisisHML -0.287** 0.153 (-2.55) (1.05) RetCDSsprd 0.0177* -0.00577 (1.83) (-0.47) CrisisRetCDSsprd -0.0466*** 0.00605 (-2.62) (0.27) Constant 0.0939 0.789** (0.39) (2.51) N 3248 2262 adj. R-sq 0.363 0.246 t statistics in parentheses * p<0.10, ** p<0.05, *** p<0.01

F-test of all Crisis factors: F-test of all CDS spreads (1) F(5, 3238) = 3.21 (1) F(2, 3238) = 3.56 Prob > F = 0.0068 Prob > F = 0.0286 (2) F(5, 2252) = 1.20 (2) F(2, 2252) = 0.11 Prob > F = 0.3088 Prob > F = 0.8950

The effect of banks’ risk on firms