Banks' Equity Stakes and Lending: Evidence from a Tax Reform

K.7 Banks’ Equity Stakes and Lending: Evidence from a Tax Reform von Beschwitz, Bastian, Daniel Foos Please cite paper as: von Beschwitz, Bastian, Daniel Foos (2016). Banks’ Equity Stakes and Lending: Evidence from a Tax Reform. International Finance Discussion Papers 1183. https://doi.org/10.17016/IFDP.2016.1183 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1183 October 2016

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1183 October 2016 Banks’ Equity Stakes and Lending: Evidence from a Tax Reform Bastian von Beschwitz and Daniel Foos NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Banks’ Equity Stakes and Lending: Evidence from a Tax Reform Bastian von Beschwitz* and Daniel Foos** Abstract: Several papers find a positive association between a bank’s equity stake in a borrowing firm and lending to that firm. While such a positive cross-sectional correlation may be due to equity stakes benefiting lending, it may also be driven by endogeneity. To distinguish the two, we study a German tax reform that permitted banks to sell their equity stakes tax-free. After the reform, many banks sold their equity stakes, but did not reduce lending to the firms. Thus, our findings suggest that the prior evidence cannot be interpreted causally and that banks’ equity stakes are immaterial for their lending. Keywords: Relationship banking, Ownership, Monitoring JEL classifications: G21, G32 * Bastian von Beschwitz, Federal Reserve Board, International Finance Division, 20th Street and Constitution Avenue N.W., Washington, D.C. 20551, tel. +1 202 475 6330, e-mail: bastian.vonbeschwitz@insead.edu (corresponding author). ** Daniel Foos, Deutsche Bundesbank, Directorate General Banking and Financial Supervision, Wilhelm-Epstein-Str. 14, 60431 Frankfurt am Main, Germany, tel. +49 (0) 69 9566-2665, e-mail: daniel.foos@bundesbank.de We thank Viral Acharya, Franklin Allen, Tobias Berg, Jean Dermine, Alex Edmans, Lily Fang, Rüdiger Fahlenbrach, Itay Goldstein, Todd Gormley, Denis Gromb, Maria Guadalupe, Wei Jiang, Kose John, Laura Lindsey, Elena Loutskina, Ron Masulis, Pedro Matos, Massimo Massa, Atif Mian, Holger Mueller, Lars Norden, Daniel Paravisini, Joel Peress, Urs Peyer, Sascha Steffen, Philip Strahan, Ingo Walter and seminar participants at the Bundesbank, ECB, INSEAD, Wharton, Tuck, Federal Reserve Board, Rotterdam, Norwegian School of Economics, HEC Paris, NTU, HKU, the Swiss Finance Institute and the 4th INSEAD PhD Workshop for valuable feedback. We thank Rainer Haselmann and Thomas Kick for sharing their expertise in using Bundesbank databases. Bastian von Beschwitz acknowledges financial support from the INSEAD Alumni Fund and the hospitality of the Deutsche Bundesbank. This paper represents the authors’ personal opinions and does not necessarily reflect the views of the Deutsche Bundesbank or the Federal Reserve Board or their staff. All remaining errors are our own.

1. Introduction For most of the twentieth century, the Glass–Steagall Act enforced a division between commercial and investment banking in the United States. When it was repealed in 1999, U.S. commercial banks not only obtained the right to engage in underwriting and advisory activity but also the right to hold large equity stakes in firms they lend to (Kroszner (2000)).1 While this option was new in the United States, banks in other countries such as Germany and Japan frequently held equity in their borrowers (Allen and Gale (1995)). Indeed, in the 1990s, some researchers praised bank equity ownership as an advantage of such bank-based economies (e.g., Prowse 1990), while others saw it as a risk to financial stability (e.g., John, John, and Saunders (1994), Boyd, Chang, and Smith (1998)). Nowadays, this discussion gains new importance as the United States reintroduce some of the Glass-Steagall provisions in form of the Volcker Rule, which limits banks’ ability to engage in private equity investments. We contribute to this debate by studying how a bank’s equity stake in a borrowing firm affects lending to that firm. There are two reasons why equity stakes may provide an advantage in lending to a firm: First, the cash flow rights of an equity stake align the incentives of debt and equity holders and thus reduce the agency cost of debt (John, John, and Saunders (1994), Santos (1999), Mahrt-Smith (2006)). Second, the control rights of the equity stake may allow the lender to better monitor the borrower. Prior papers (e.g. Ferreira and Matos (2012), Prowse (1990)) find a positive association between banks’ equity stakes and lending, which they argue supports that equity stakes are beneficial for lending. However, such a positive cross-sectional correlation can also be driven by endogeneity issues because banks decide on their debt and equity holdings jointly. 1 The Gramm-Leach-Bliley Act of 1999 allows banks to engage in “merchant banking investments”. Banks can obtain any percentage of equity in an industrial company and keep it for up to 15 years. The bank is not allowed to “routinely manage or operate” the company, but can exert some influence on the firm. Under the Glass-Steagall Act, such merchant banking investments were limited to 5% of voting equity (Kroszner, 2000). 3

To address such endogeneity issues, we use a German tax reform which caused banks to sell their equity stakes as identification. While we confirm prior papers in finding a positive cross-sectional correlation between banks’ equity stake and lending, we do not find any causal evidence that equity stakes benefit lending: following the reform, banks sell their equity stakes, but they do not decrease lending to these firms. Thus, our findings suggest that the prior evidence cannot be interpreted causally and that banks’ equity stakes are immaterial for lending. We use the German capital gains tax reform in 2000 for identification. Before the reform, German banks held large minority stakes in German industrial companies. For example, Deutsche Bank held equity stakes in 24 industrial companies ranging from 0.25% to 30% of the firms’ equity and adding up to a total market value of EUR 22.7 billion (45% of Deutsche Bank’s own market capitalization). Many of these stakes had already been established in the 1950s and 1960s and thus had accumulated large unrealized capital gains over time. Consequently, banks maintained the equity stakes mainly to avoid capital gains taxation (Sautner and Villalonga (2010)). The German government decided to abolish the 50% capital gains tax in 2000, thus removing a major obstacle to the divestiture of banks’ equity stakes (Edwards et al. (2004)). Banks’ stock prices increased by 5% upon the announcement of the tax reform (see Figure 1), which suggests that (1) the reform was a surprise to the market, and (2) that the tax acted as a binding constraint on banks. As these equity stakes tied up much capital, one would expect that banks divest their stakes despite potential benefits these stakes might have for the lending relationships. Indeed, 75% of banks’ equity stakes were divested in the six years following the tax reform (see Figure 2). In our main tests, we use loan-level data from the German Large Credit Register provided by the German Central Bank (Deutsche Bundesbank) to study how equity stakes by banks affected lending. We start by conducting the same cross-sectional analysis as prior studies (e.g. Ferreira and Matos (2012), Prowse (1990)) for the time period before the announcement of the tax reform. We find a 4

positive correlation between equity stakes and lending for both the extensive and intensive margin. The existence of an equity link is correlated with a higher probability of a lending relationship (44.1% instead of 18.7%) and conditional on the existence of a lending relationship, the bank lends on average 2.6 times as much money to the firm if it holds its equity. These results are significant at the 1 percent level in all our specifications. While the positive cross-sectional correlation is in line with the prior papers, it does not imply that equity stakes are beneficial for lending. Rather it could be explained by reverse causality or omitted variable bias. Therefore, we go on to study how banks change their lending to firms whose equity they divest following the tax reform. Because divestiture decisions are endogenous, we condition our analysis only on the ownership of equity stakes at the time of the tax reform. This methodology is essentially a reduced form instrumental variable approach, in which the sale of an equity stake is instrumented with the existence of an equity stake prior to the tax reform. Specifically, we run a difference-in-difference regression where the treated group consists of bank-firm pairs with an equity stake in 1999. Since the treatment occurs at the bank-firm-level, we can control for trends at the bank- and firm-level with bankquarter and firm-quarter fixed effects, respectively. Intuitively, bank-quarter fixed effects imply that we examine the change in the bank’s lending to the firm in which it held equity relative to its lending to other firms, while firm-quarter fixed effects imply that we study how the company’s borrowing from the (formerly) equity-linked bank changes relative to its borrowing from other banks. Thus, we control for any bank or firm specific effects, such as the additional capital that banks obtain from selling their equity stakes. If equity stakes were indeed beneficial for lending (as prior papers have argued), we would expect a reduction in lending to firms in which banks held equity before the tax reform. However, we do not find any evidence of this. Rather there seems to be no significant change in lending to previously equity linked firms following the tax reform. If anything, there is an (insignificant) increase in lending. This 5

finding suggests that the evidence in the prior papers may be seriously affected by endogeneity issues and thus misleading. We show several additional analyses that confirm our results and address potential concerns with our methodology. For example, one concern may be that our results are distorted by including equity stakes that were not sold. Thus, in a second test, we rerun our difference-in-difference set-up, but only include equity stakes that were divested until 2005. The result remains unchanged. Next, we directly study the change in lending around divestitures after the tax reform. While this analysis is affected by endogeneity concerns, such endogeneity would most likely bias the analysis towards finding a negative effect of divestitures on lending. For example, if a bank has negative information about the firm, it may choose to divest the equity stake and reduce lending. However, even using this specification, we find that banks (insignificantly) increased lending following the divestiture of an equity stake. Next, we examine whether we can find a negative effect of the tax reform in the subset of bankfirm pairs where equity stakes are particularly likely to benefit lending. Specifically, we study the subset of cases where the bank also sits on the company’s supervisory board and where the company is not publicly listed. We don’t observe a negative effect of the tax reform on lending for either of these two subsets. We conduct several robustness checks. For example, we add companies and banks without equity links to our control group or study different regression set-ups. We show that our sample is representative: Banks that hold equity stakes account for 59% of the assets of the German banking system and firms with banks as equity holders are similar to other German firms along a number of characteristics. Overall, our findings show that banks do not reduce lending to firms in which they divest equity stakes for exogenous reasons. This finding suggests that banks’ equity stakes do not provide any 6

benefits for the lending relationship. Contrary to how the cross-sectional correlation has been interpreted, equity stakes seem to be immaterial for lending. This suggests that the cross-sectional evidence provided in other studies is affected by endogeneity issues such as reverse causality and omitted variable bias. Reverse causality seems likely because several studies document that lenders can use their access to private information for profitable investments in the borrower’s equity (Massa and Rehman (2008), Ivashina and Sun (2011), Massoud et al. (2011)). Also omitted variable bias seems relevant, because good relationships with a company may cause a bank to invest in both, the company’s debt and equity. An alternative interpretation is that an equity stake might facilitate the initiation of a lending relationship, but it may provide no further benefits afterward. Consistent with this view, Hellmann, Lindsey, and Puri (2008) show that banks are more likely to extend credit to companies in which they have previously held a venture capital investment. We can only speculate which mechanism is at work here, but our results suggest that the cross-sectional evidence is misleading for established lending relationships. The finding that equity stakes do not benefit lending also has important policy implications. It suggests regulations preventing banks from holding equity in their borrowers (such as under Glass– Steagall) do not adversely affect lending. Our paper adds to a strand of literature studying the effect of equity stakes on lending. These studies generally document a positive association between banks’ debt and equity investments in a firm, be it in Portugal (Antão, Ferreira, and Lacerda (2011)) or in Japan (Flath (1993), Prowse (1990), Sheard (1989)). Similarly, Ferreira and Matos (2012) find that banks are more likely to act as lead arrangers for syndicated loans to companies in which they hold an equity stake through their mutual fund or asset management divisions. Different to our paper, these studies are based on cross-sectional comparisons. Our difference in difference analysis suggests that their results cannot be interpreted as showing that equity stakes benefit lending. 7

Our research is also related to analyses of the relationship between equity stakes and loan pricing. Santos and Wilson (2008) find that banks request lower interest rates from borrowers whose voting rights they control. Fang, Ivashina, and Lerner (2013) show that banks provide cheaper financing to companies in which their private equity arms invest. Jiang, Li, and Shao (2010) show that non-bank institutional investors charge lower interest rates on syndicated loans if they also hold equity in the borrower. Contrary to these studies, Lim, Minton, and Weisbach (2012) find that leveraged loans of non-bank institutional investors have higher spreads when the lenders hold equity in the borrower. While our data do not allow us to study loan pricing, our findings show the importance of controlling for endogeneity when studying equity investments of banks, which may explain the inconsistent results on loan pricing. Other studies focus on the effect of bank ownership on the borrowing firm. For Japan, bank ownership is associated with better performance in financial distress (Hoshi, Kashyap, and Scharfstein (1990)) and better access to financing (Hoshi, Kashyap, and Scharfstein (1991)), but also with lower growth and profitability (Weinstein and Yafeh (1998)). For Germany, several studies find a positive effect of bank influence on firm performance (Gorton and Schmid (2000), Lehmann and Weigand (2000), Cable (1985)), while Dittmann, Maug, and Schneider (2010) find that bankers on the firm’s supervisory board add little value. Bank equity ownership may be beneficial for two reasons: Banks may provide more financing or they may be efficient shareholders. Our results suggest that the first is not the case. 2. Institutional details 2.1 The German Tax Reduction Act of 2000 Our source of identification is Germany’s abolition of its corporate capital gains tax in 2000. Before the reform, many German banks held equity stakes in German industrial companies. These equity stakes were part of a system of minority stakes and cross-holdings often called “Germany Inc.” (Deutschland 8

AG) (Höpner and Krempel (2006)). Many of these holdings had been acquired in the distant past and thus had book values significantly below their market value (Edwards et al. (2004)). For example, some of the holdings of Deutsche Bank in old industrial companies dated back to the companies’ foundation before World War II. Other equity stakes of banks were acquired in the 1950s and 1960s, potentially to exercise control over industrial companies through board representation. In the 1990s, German banks increasingly moved towards investment banking and wanted to divest their equity holdings to free up capital (Beyer (2003), Vitols (2005)). However, the prevailing 50% corporate capital gains tax rate implied that banks would have been subject to a significant tax cost from divesting their equity holdings in German companies (Sautner and Villalonga (2010)).2 This lock-in was lifted when the government introduced a tax exemption on the sale of equity stakes with the explicit intention to facilitate the sale of equity stakes (Hoepner, 2000). The abolition of the capital gains tax was part of a wider tax reform, which included a reduction in individual and corporate tax rates as well as a change in dividends taxation. However, there is no reason to believe that the general tax changes specifically affected banks’ lending to firms in which they hold equity.3 The tax reform was first announced in December 1999 and the German parliament it in the summer of 2000 to become effective on January 1, 2001. The abolition of the capital gains tax did not enter into force until January 1, 2002. However, there were ways for companies to divest holdings before 2002 and still capture the better tax treatment. For example, Deutsche Bank sold a stake in Allianz on 6 June 2000 and stated in its investor relations release: “The economic disposal has been achieved by an innovative structure which allows Deutsche Bank to obtain the full benefits from the upcoming tax reform in Germany. The transaction will qualify as a 2 Capital gains were taxed at 40% corporate tax and trade tax that varies across regions, but was approximately 10%. For more details see Edwards et al. (2004). 3 For a detailed description of the tax reform, see Keen (2002). 9

disposal for the Deutsche Bank Group in its IAS accounts, giving rise to a capital gain in excess of EUR 2 billion, but without triggering a tax disposal in the current year.”4 Furthermore, anticipating the tax reform, banks may have changed their lending to firms in which they plan to sell equity and thus the tax reform may already have had an effect before January 1st, 2002. Accordingly, we use the announcement of the tax reform in December 1999 as our event date. 2.2 The regulatory and supervisory environment Our sample period ranges from 1998 to 2005. It therefore lies entirely before the implementation of the Basel II Accords in Germany in 20075. Thus, according to Basel I rules, equity stakes and loans were part of risk-weighted assets, which banks had to back with 4% core capital and with 8% total core and additional capital. The risk weight for equity stakes was 100% and thus the same as that of uncollateralized corporate bonds. For tax and regulatory purposes, banks applied German GAAP accounting and thus valuing equity stakes at historical cost. Loans were generally valued at face value. We conclude from this environment that, first, changes in the market value of the equity stakes did generally not affect the bank’s book equity or risk-weighted assets. Second, the tax reform created an incentive for banks to divest their holdings because a sale of an equity stake increased their capital by the amount of the capital gains. Different to the United States, there are no lender liability laws in Germany. Thus, banks do not risk equitable subordination in the case of bankruptcy if they are represented on the firm’s supervisory board or hold the firm’s equity (Dittmann, Maug, and Schneider (2010)). 4 "Deutsche Bank reduces Allianz stake to 4.1%" (https://www.db.com/ir/en/content/ir_releases_2000_3464.htm) and "Deutsche plays clever in Allianz sale to avoid CGT" (http://www.efinancialnews.com/story/2000-06-12/deutsche-playsclever-in-allianz-sale-to-avoid-cgt?ea9c8a2de0ee111045601ab04d673622). 5 The Basel II Accords were published already in June 2004 and discussed before that. Therefore, German banks might have taken the expected change in regulation into account during our sample period. 10

We provide additional information on the German banking system and German corporate governance in Appendix 2 to 4. 3. Data and Variables 3.1 Ownership data Our data on equity holdings of German companies is from Who owns Whom? (Wer gehört zu wem?) provided by Picoware as of July 1999. This database contains ownership data for private and public companies. The data are based on public sources and self-reported information. In addition, we manually add holdings of banks reported in Hoppenstedt Aktienführer 2001 (these holdings are as of December 1999). To focus on equity stakes rather than subsidiaries, we define an equity link as a holding by a bank in an industrial company of less than 50% equity. However, we do include equity holdings of a bank’s subsidiary if the bank holds at least 75% of the subsidiary’s equity.6 We exclude holdings in other banks, private equity companies and vehicles of project finance using the industry information provided in Who owns Whom? as well as manual checks. After these filters, our sample includes 135 equity stakes that 26 banks held in 117 companies. We also determine the year in which an equity stake is divested. This exercise is not trivial given that both Who owns Whom? and Hoppenstedt Aktienführer provide only yearly cross-sections rather than a panel dataset. We choose a conservative measure of divestiture, which only counts an equity stake as divested if a company is still covered in the data and the equity stake is listed as belonging to another owner or has decreased by 50 percent. We also ensure through manual checks that the new owner is indeed a different company and that the equity stake has not just moved into a different holding vehicle of the same bank. If a company disappears from the data before a divestiture, we set the time of divestiture to missing. 6 We also include the equity holdings of lower levels of subsidiaries (i.e. subsidiaries of subsidiaries) as long as any link consists of at least 75%. The subsidiary can be a non-financial company. Results are also robust to using a 50% cut-off instead. 11

3.2 Loan-level data Our loan-level data comes from the German Large Credit Register (Millionenkreditevidenz) provided by the German Central Bank and covers the time period from 1998 to 2005. German banks must report their debt exposures to companies and individuals at the end of each quarter if the exposure exceeds EUR 1.5 million during the quarter. For ease of interpretation, we refer to this quarterly exposure as a loan (following Khwaja and Mian (2008)). The credit exposure is further broken down into on-balance sheet items such as loans and bonds, as well as off-balance sheet exposure through credit derivatives, guarantees and undrawn credit lines. We focus on the total credit risk a bank faces towards a borrower. We use four measures of loan size in the paper: Loan Size (log) is the natural logarithm of the size of the loan, while Loan Size (EUR million) is the unstandardized loan size winsorized at the 1% and 99% threshold. Bank Share is the share of a borrower’s total credit that is provided by a bank. Borrower Percentile measures the importance of a borrower to a bank. It is a percentile based on the size of a borrower’s loan relative to other borrowers of the bank. Detailed definitions of all variables are reported in Appendix 1. 3.3 Bank-level data We conduct our main analysis with the 26 banks that hold equity stakes in industrial companies. These large banks account for 59% of the German banking system by assets. An extended sample used in robustness checks includes other banks that are covered by Who owns Whom? and have more than EUR 1 billion in total loans outstanding from 1998 to 1999 according to the German Large Credit Register. For these 89 banks, we obtain balance sheet data as of December 1999 from the Regulatory Credit Information System (Bankaufsichtliches Kredit-Informationssystem, BAKIS) of the German Central Bank. 12

3.4 Firm-level data We match the firms in the German Large Credit Register to the firms which are covered in Who owns Whom?. For publicly listed firms, balance sheet data of the last fiscal year ending before December 1999 comes from Worldscope and stock market data from Datastream and Compustat Global. For private companies, we obtain balance sheet data from the Jalys/USTAN database, which was constructed for the rediscount business of the German Central Bank. We match Jalys/USTAN to the German Large Credit Register using the links employed in Haselmann, Schoenherr, and Vig (2013). Jalys/USTAN is “the best and most comprehensive firm data set in Germany” for our time period (Stöss (2001)). Still some companies are not covered in every year. If a company is not covered in 1999, we instead use the last data point available in the dataset going back to 1995. We manually collect information on the composition of supervisory boards in 1999 from annual reports. 3.5 Summary statistics We include in our main analyses only banks and firms with equity links in 1999, which we will refer to as the Inside Sample. Table 1 reports summary statistics for the Inside Sample as of 1999. The firms in our sample are fairly large with average assets of EUR 9,129 million (median EUR 304 million) and they receive on average loans from 12.8 (median 7) banks. The high number of banking relationships allows us to use firm-time effects in our difference-in-difference set up. Half of the sample consists of publicly listed companies. Also the banks in our sample are large with mean assets of EUR 107 billion (median EUR 53 billion). On average, they hold equity stakes in 5.2 companies (median 1.5), making up 23% of the bank’s equity on average (median 3.8%). This means that equity stakes are large enough to affect a bank’s risk assessment. Of our 26 banks, 11 banks belong to the commercial sector, 11 banks belong to the public sector and 4 banks belong to the cooperative sector. 13

If there is an equity link, a bank holds 14.6% of the firm’s equity on average (median 10.5%). For 66% of the equity links, the bank also provides a loan to the company. In this case, the market value of equity is typically twice as large as the loan given that the debt constitutes on average 36% (median 32%) of total financing.7 In 75% of the cases where the bank holds equity in the company, it is also represented on the company’s supervisory board. 4. Results 4.1 Announcement of the tax reform: Event study The abolition of the capital gains tax was part of a wider tax reform. While the general tax reform was announced on December 21, 1999, the plan to abolish the corporate capital gains tax was not confirmed until December 23, 1999. This enables us to examine the market reaction to the capital gains tax announcement separately (following Edwards et al. (2004)). In Panel A of Figure 1, we plot the average stock returns for banks with equity stakes around the announcement. Because almost all publicly listed German banks had equity stakes, we use German industrial companies and US banks as control groups to proxy for country and industry specific shocks. During a quiet stock market environment in the last week before Christmas, the stock prices of banks with equity stakes shot up by 5.1%, while the return for both control groups was only 0.3%. This finding is consistent with Edwards et al. (2004) In Panel B, we display the returns on the announcement day for the four largest German banks, which also held the most equity stakes. To better understand these returns, we compare them to the aggregate value of the banks’ equity stakes. Deutsche Bank, whose equity stakes had the highest aggregate value (EUR 18.6 billion), experienced a stock price increase of 13.6%, while Commerzbank, whose equity stakes had the lowest value (EUR 2.2 billion), experience a stock price increase of only 7 If the firm is privately listed, we estimate the market capitalization using a multiple of its book equity, where this multiple is the median of the book to equity ratio of publicly listed companies. 14

4.7%. This pattern suggests that the holding of equity stakes determined the market reaction. The average increase in market capitalization on the announcement day divided by the aggregate value of equity stakes was 33%. Given that the tax rate before the reform was 50%, the capital gains made up at least 66% of the value of the equity stakes and the surprise element of the tax reform was at least 66%.8 In addition to the event study, newspaper articles suggest that the tax reform was a surprise to the market. For example, Handelsblatt, a German business newspaper, headlined “Eichel [the German minister of finance] surprises stock market with a Christmas present”.9 The fact that the reform was a surprise alleviates worries that equity stakes or lending decisions before 1999 were endogenous to the anticipation of the tax reform. 4.2 Divestitures of equity stakes following the tax reform Several prior studies find that most of the bank equity stakes were divested following the tax reform (Kengelbach and Roos (2006), Höpner and Krempel (2006)). In Figure 2, we confirm this result for our dataset. We show that 75% of banks’ equity stakes in industrial firms were divested at least by half until 2005. This result is not surprising given that banks had strong incentives to divest their equity stakes: First, banks realize large accounting gains by selling the holdings, which improves their capital position and leads to potential income and reputation gains for the bank’s management. Second, divesting a holding leads to a cash proceed, which improves the liquidity position of the bank. Third, there was a general election in Germany in September 2002, and the political opposition had announced a plan to reintroduce capital gains taxation. The opposition narrowly lost the election and the corporate capital 8 The value increase of the equity stakes on the announcement day should depend on the amount of capital gains and on the change in probability of the passage of the tax reform. It can be approximately computed as: 𝑉𝑎𝑙𝑢𝑒 𝐼𝑛𝑐𝑟𝑒𝑎𝑠𝑒= 𝐶ℎ𝑎𝑛𝑔𝑒 𝑖𝑛 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑇𝑎𝑥 𝑅𝑒𝑓𝑜𝑟𝑚∗𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝐺𝑎𝑖𝑛𝑠∗0.50= 0.33 where 50% is the capital gains tax rate. This calculation implies that both Change in Probability of Tax Reform and Capital Gains were at least 66%. 9 “Eichel überrascht die Börse mit einem Weihnachtsgeschenk”, Handelsblatt, December 24, 1999. 15

gains tax has not returned since. However, in 2002 the opposition was ahead in the polls potentially causing some banks to sell their holdings to front-run a return of the tax (Pauly and Schäfer (2002)). 4.3 Cross-sectional correlation between equity stakes and lending Several prior papers find a positive correlation between banks’ equity stakes and lending, for example in Japan and Portugal ((Antão, Ferreira and Lacerda (2011), Flath (1993), Prowse (1990), Sheard (1989)). In this paper we study whether the same cross-sectional correlation can be found in our data for Germany. We report our results in Table 2. In Panel A, we use a dataset with all possible bankfirm combinations, following Ferreira and Matos (2012) and Hellmann, Lindsey and Puri (2008). The dependent variable Lending Relationship is a dummy variable equal to one if the company received a loan from the bank in December 1999. The explanatory variable of interest is Equity Link in 1999. We control for geographic proximity using dummy variables equal to one if bank and firm are located in the same city or the same region. We use size, leverage and tangibility of assets as firm-specific controls. In regressions 2 to 4, we replace them with firm fixed effects. Since our dependent variables are binary, we use a logit specification, which we replace with a linear probability model (OLS) in the specifications with firm fixed effects due to the incidental parameter problem.10 We double-cluster standard errors at the bank and firm level. We find that an equity link is correlated with a higher probability of a lending relationship (44.1% instead of 18.7%). This result is significant at the 1 percent level in all our specifications. Next, we want to examine whether an equity link is also associated with larger loans conditional on the existence of a lending relationship. We run OLS regressions where each observation is a bankfirm pair for which average lending in December 1999 is positive. The results are presented in Panel B. 10 Since our sample includes many firms, but only few banks, a simple non-linear probability model cannot be consistently estimated due to an incidental parameters problem (Chamberlain (1980); Puri, Rocholl and Steffen (2011), Greene (2004), Neyman and Scott (1948)). Therefore, following Puri, Rocholl and Steffen (2011) and Khwaja and Mian (2008), we employ a linear probability model estimated with OLS. 16

We find that in the cases where a bank holds a firm’s equity, it lends 2.6 times as much or EUR 21.9 million more.11 A bank that holds an equity stake in the company on average provides 8.9 percentage points more of the company’s debt and the company ranks 11 percentage points higher amongst the banks borrowers. All these effects are significant at least at the 1% level. These results indicate that there is the same positive cross-sectional correlation between equity stakes and lending volume in Germany as has been shown in previous studies for other countries. This suggests that the findings of our study may be extended to other countries as well. However, it is important to note that the positive cross-sectional correlation does not imply that equity stakes are beneficial for lending. Rather the positive cross-sectional correlation could be explained by reverse causality or omitted variable bias. Thus, we now turn to our difference-in-difference set up, which studies how lending reacts to exogenously triggered divestitures of equity stakes. 4.4 Impact of the tax reform on lending: Graphical evidence If equity stakes do indeed benefit lending as has been argued by the prior literature, we would expect that lending to equity-linked firms decreases after the equity stakes are divested following the tax reform. Thus, we compare how lending changes after the tax reform to firms in which the bank held an equity stake in 1999. Since the treatment takes place at the bank-firm level, we do not need other firms or banks as a control group. Rather, we include only banks and firms with at least one equity link in 1999 in our sample (Inside Sample) and use their lending relationships without an equity link as the control group. To focus only on existing lending relationships, we only include bank-firm pairs for which the average lending before the tax reform (1998 to 1999) was positive. We start with a simple graphical analysis where we compute the average log loan size each year for bank-firm pairs with and without an equity link. If equity stakes benefit lending, we would expect 11 A difference in log loans of 0.96 corresponds to a difference in loans of 𝑒0.96−1=160%. 17

the lending to firms with equity links to decrease after the tax reform relative to firms without equity links. We do not find any evidence of this. Rather, the two lines seem to move extremely parallel both before and after the tax reform. 4.5 Impact of the tax reform on lending: Difference-in-difference Next, we conduct a more rigorous difference-in-difference analysis to study how the sale of equity holdings following the tax reform affected lending. We use the following set up that controls for bank and firm specific trends: 𝐿𝑜𝑎𝑛 𝑆𝑖𝑧𝑒 = 𝛼 +𝛼 +𝛼 +𝛼 +𝛽∗𝐸𝑞𝑢𝑖𝑡𝑦 𝑆𝑡𝑎𝑘𝑒 1999 ∗𝑃𝑜𝑠𝑡 𝑅𝑒𝑓𝑜𝑟𝑚 𝑏,𝑓,𝑞 𝑞 𝑏,𝑓 𝑏,𝑦 𝑓,𝑦 𝑏,𝑓 𝑞 + 𝛾∗𝐶𝑡𝑟𝑙 ∗𝑃𝑜𝑠𝑡 𝑅𝑒𝑓𝑜𝑟𝑚 +𝜀 𝑏,𝑓 𝑞 𝑏,𝑓,𝑞 where f denotes the firm, b the bank, q the quarter, y the year and pr the post tax reform period. Equity Stake 1999 is a dummy variable equal to one if the bank holds an equity stake in the firm in 1999 and Post Reform is a dummy variable equal to one from 2000 to 2005 (after the announcement of the tax reform). The explanatory variable of interest is the interaction between Equity Stake 1999 and Post Reform. It measures how the loan size of bank-firm pairs with equity links in 1999 changes relative to bank-firm pairs without links in 1999 following the tax reform. In this test, we only condition our analysis on the existence of an equity stake in 1999 rather than the endogenous decision to divest it. This methodology is essentially a reduced form instrumental variable approach, in which the sale of an equity stake is instrumented with the existence of an equity stake prior to the tax reform. We include quarter fixed effects (denoted 𝛼 ) controlling for changes in the economic environment 𝑞 and bank-firm fixed effects (denoted 𝛼 ) controlling for any time-invariant heterogeneity at the bank- 𝑏,𝑓 firm-level. In addition, there may be time-varying effects on the firm or bank-level such as changes in credit demand or credit supply, which are potentially correlated with having an equity link. For example, banks may have used the additional capital from divesting their equity stakes to increase overall lending. 18

We account for these issues by including bank and firm fixed effects interacted with yearly dummy variables (denoted 𝛼 and 𝛼 ). With this regression set up, the only omitted variables that can distort 𝑏,𝑦 𝑓,𝑦 our results are relationship-specific variables which are correlated with equity links and have a timevarying effect. Thus, we include Same Region and Same City indicators interacted with Post Reform. We display our results in Panel A of Table 3. We use the main specification explained above and a specification in which bank-year and firm-year fixed effects are replaced with firm–Post Reform fixed effects. We use all four of our loan size measures as dependent variables. For none of our dependent variables the tax reform decreases lending for bank-firm pairs with an equity link in 1999. In fact, all coefficients are (insignificantly) positive. This suggests that equity stakes do not benefit lending. Rather it suggests that equity stakes are immaterial to lending and the positive cross-sectional correlation is explained by endogeneity issues. Because the decision to divest an equity stake is endogenous, we condition our analysis only on the existence of an equity stake in 1999. However, given that we would not expect any effect on lending if equity stakes are not sold, this set-up biases our results towards zero. This could potentially explain why we do not observe a negative effect on lending. To address this concern, we run our analysis again, but remove from our sample all equity stakes that have not been divested (or whose divestiture information is missing). The results are presented in Panel B. They are very close to the results in Panel A and all coefficients remain positive. 4.6 Change in lending after divestitures Next, we study how lending changes after actual divestitures. This analysis is affected by endogeneity concerns. For example, if a bank has negative information about the firm, it may choose to divest the equity stake and reduce lending. However, such endogeneity would bias the analysis towards finding a negative effect of divestitures on lending. Thus, if we don’t find a negative effect in this endogeneous analysis, it would suggest that the causal effect is also not negative. 19

We start by a simple graphical analysis. We plot the Bank Share, i.e. the average share of the firm’s total borrowing provided by the bank around the divestiture time. We focus on the Bank Share, because it controls for general demand effects without needing a control group. The results are presented in Figure 4. We do not observe a reduction in lending after the divestiture. If at all, it seems like the opposite: While lending slightly decreased before the divestiture, it seems to increase somewhat after the divestiture. Next we study the effect of divestitures in a regression setting. We use a setting similar to the difference-in-difference analysis in Table 3. However, now the main explanatory variable of interest is Equity Link is Divested, which is a dummy variable equal to one after the equity stake is divested and zero before. For bank-firm pairs without an equity stake, this variable is always zero. Therefore, it captures how lending changes after the divestiture relative to bank-firm pairs without an equity stake. The results are presented in Table 4. They are very similar to the results in Table 3. Once again, all coefficients are very close to zero and if anything, they are positive. This finding suggests that there is no decrease in lending after the divestitures of the equity stakes. Given that endogeneity effects should bias the results to be more negative, these findings suggest that there is no negative causal effect of divestitures on lending. This suggests that equity stakes do not benefit lending and that the crosssectional correlation above is driven by endogeneity issues. 4.7 Sample splits So far, we were not able to find a negative effect of exogenous divestitures of equity stakes on lending when focusing on our whole sample. However, the effect may be present in certain subsamples, in which bank equity ownership is especially beneficial. We consider two such subsamples: First, the proposed monitoring benefits of equity ownership may be higher if the equity stake allows the bank to be represented on the company’s board. Second, bank monitoring through equity stakes may be more important in private firms, because they are more opaque as they are less monitored by the general 20

public. There is a second reason, why bank equity stakes may be more beneficial in private firms: If a bank has a close lending relationship with a firm, it may get inside information that gives it monopoly power. If the firm turns to other banks, they will perceive this as a negative signal (Sharpe (1990), Rajan (1992), von Thadden (1995)). Mahrt-Smith (2006) argues that bank equity ownership in the firm can help prevent such hold-up problem. Such a hold-up problem is much more likely in private firms, which are more opaque, than in public firms that are better able to access public capital markets. Therefore, we now split our sample by board links and whether the firm is publicly listed. We present the results in Table 5. In Panel A, we report difference-in-difference regressions, where we split the treated group by whether a representative of the bank sits on the firm’s supervisory board in 1999 (we will refer to this as a board link). More specifically, we define two dummy variables: Equity Link with Board Link 1999 is equal to one if the bank holds an equity stake in the firm in 1999 and there is a board link, while Equity Link without Board Link 1999 equals one if there is an equity link, but no board link in 1999. If a bank-firm pair does not have an equity link in 1999, both variables are set to zero. By including both of these variables (interacted with Post Reform) in the regression, we estimate the effect of equity stakes with and without board links both compared to bank-firm pairs without equity links. This set up is comparable to a sample split, but has the advantage of using a larger sample to compute coefficients of fixed effects and control variables. Using this set up, we do not find a negative effect of the tax reform on lending in either subgroup. Almost all coefficients are positive and the ones that are negative have t-statistics very close to zero. In fact, for some measures of loan size we even find a positive significant effect. This finding suggests that equity stakes do not benefit lending even when they allow the bank to obtain representation of the firm’s supervisory board. Next, in Panel B, we conduct a similar analysis where we split the treated group by whether the firm is publicly listed. Once again, almost all coefficients remain (insignificantly) positive and those two cases where coefficients are negative, the t-statistics are very close to zero. This finding suggests 21

that equity stakes do not benefit lending even when the firm is not publicly listed. More generally, these findings show that our results seem to be the same in different subsamples of the data. 4.8 External validity While using a natural experiment has the important advantage of addressing endogeneity, it always comes at the cost of being limited to a specific sample and time period. Thus, it is important to determine whether the results can be generalized to other settings. This question cannot be tested, but we examine in this section whether external validity seems likely. The equity stakes in our sample are fairly large with an average size of 14.6 % in terms of the firm’s equity (see Table 1). Given that we do not find any effect of the divestitures of these large equity stakes, it seems unlikely that smaller equity stakes should benefit lending. Thus, it is likely that our findings transfer to other countries where banks hold smaller equity stakes. This is further supported by the fact that we find the same cross-sectional correlation as studies in Portugal and Japan (see section 4.3). Next, to assess how banks owning equity stakes compare to the entire population of German banks, we plot the 50 largest German banks by assets in Figure 5. Of the 26 banks with equity stakes, 22 are amongst the largest 50. In general, the 26 banks with equity stakes in 1999 account for 59% of the total assets of the German banking system. Thus, they are representative of the German banking system as a whole. In Figure 6, we compare firms with equity links to a control group matched by assets and the criterion whether the firm is publicly listed. We find that there is no significant difference between the two groups for several control variables and in terms of industry composition. Thus, our firms are representative of German firms in general. 22

To summarize, the features of our data make it likely that our findings are general and not specific to our sample. Nonetheless, external validity can only be verified using different samples, which calls for further research in this area. 5. Robustness checks 5.1 Less fixed effects Given that the main result of our paper is that there is no significant effect, one can be concerned that we do not find statistically significant results, because our tests to not have enough power. The power of our test will be reduced by the large number of fixed effects we employ to control for supply and demand effects. Thus, in the robustness check in Table 6, we rerun our main analysis, but drop the firmyear and bank-year fixed effects as well as the controls for geographical proximity. This barely changes our results at all. Neither do they change when we add back the controls for geographical proximity and bank-Post Reform fixed effects. This suggests that our results are not driven by the inclusion of these fixed effects. 5.2 Different control groups In our main test, we only include the Inside Sample of banks and firms with equity links. Using this Inside Sample has the advantage that our results cannot be driven by different time trends between treated and control firms or banks. The disadvantage is that bank-firm pairs in the control sample may be indirectly affected by the tax reform because the firm (bank) always has an equity link to another bank (firm). Therefore, we show in this section that our results are robust to conducting the same analysis on a larger sample. In this sample, we include all 89 banks that are covered in Who owns Whom? and have more than EUR 1 billion in total loans outstanding from 1998 to 1999 according to the German Large Credit Register, as well as all firms covered in Who owns whom? with assets above EUR 5 million. As the sample is larger, we report regressions using Difference in Loan Size as the 23

dependent variable (following the methodology of Khwaja and Mian (2008)).12. We report the results of this specification in Table 7. In Panel A, we repeat our analysis on the Inside Sample to show robustness to the alternative methodology. As expected given the similarity of the two approaches, the results are very similar. In Panel B, we report results for the full sample of all banks and firms. Once again, all coefficients remain positive and the positive effect for Borrower Percentile even becomes signficant. Finally, in Panel C, we use only firms that do not have equity links with any bank as the control group. This set up has the advantage that the control firms are not affected by the tax reform. However, this comes at the cost of not being able to control for firm-specific trends because the control group consists of different firms than the treated group. Thus, we report regressions where we replace the firm fixed effects with firm controls. Once again all coefficients are (insignificantly) positive. 5.3 Equity repurchases following the tax reform Potentially negative results could be outweighed by a positive effect coming from banks extending credit to firms in order to enable these firms to buy back the equity stake from the bank. In this section, we study whether our results are driven by this phenomenon. Thus, we examine whether firms with an equity link exhibit an increase in the number of stock repurchases following the tax reform. We report the results in Table 8. The dependent variable in this regression is a dummy variable equal to one if the firm undertakes a buyback in a year measured as an increase in the number of treasury shares held by the firm. In Regression 1, we estimate a logit model only after the tax reform, which shows that there is no effect of equity stakes on repurchases cross-sectionally. In Regressions 2 to 4, we apply a difference-in-difference set up using a linear probability model in the sample from 1998 to 2005. None of the specifications shows an increase of repurchases for firms with equity links. The coefficients are 12 Due to the larger sample, it is computationally infeasible to run a full-fledged panel model as in Table 2 and Table 3. Due to data confidentiality reasons, we have to run all analyses on the computers of the German central bank, which have low computing power. 24

never significant and in two cases they are, in fact, negative. This finding suggests that the banks did not sell their equity stakes to the specific firms. 6. Conclusion This paper studies how a bank’s equity stake in a borrowing firm affects lending to that firm. Several papers have argued that equity stakes may be beneficial for lending because the control rights of the equity stake provide the lender with an advantage in monitoring the borrower and the cash flow rights align the incentives of debt and equity holders. Indeed, the prior literature has documented that equity stakes and lending are positively correlated. However, such a positive cross-sectional correlation does not necessarily mean that equity stakes benefit lending, but may may also be driven by endogeneity issues. In this paper, we address such endogeneity issues by using the German capital gains tax reform in 2000 as a natural experiment. The tax reform abolished the corporate capital gains tax of 50%, enabling banks to divest their equity stakes in industrial companies. We find that banks sell most of their equity stakes in industrial firms after the tax reform. However, we cannot find any evidence that banks reduced lending to these firms following the divestitures. If anything, there is an (insignificant) increase in lending. This suggests that the positive cross-sectional correlation between equity stakes and lending does not imply that equity stakes benefit lending. Rather equity stakes seem to be irrelevant to lending. Thus, the positive cross-sectional correlation between equity stakes and lending, which we also confirm in our sample, seems to be driven by endogeneity issues. For example, banks may use their private information from lending to invest into a firm’s equity (reverse causality) or good relationships with a company may cause a bank to invest in both, the firm’s debt and equity (omitted variable bias). An alternative interpretation is that an equity stakes might facilitate the initiation of a lending relationship, but provide no further benefits afterward. We can only speculate which mechanism is at 25

work here, but our results suggest that the cross-sectional evidence is misleading for established lending relationships. Our findings have important policy implications for bank-based economies such as Germany and Japan as well as for the United States where banks are allowed to hold large equity stakes in their borrowers since the repeal of the Glass–Steagall Act. Our results indicate that the benefits of equity stakes for lending may be overstated in the literature. We find no evidence that regulation preventing banks from owning equity in industrial companies adversely affects lending. More broadly, this finding suggests that conflicts between debt and equity holders are of minor importance in our sample. Our findings are also relevant for institutional investors that participate in lending syndicates because these “shadow banks” often hold both debt and equity of the same company (Jiang, Li and Shao (2010), Lim, Minton, and Weisbach (2012)). 26

References Allen, Franklin, and Douglas Gale, 1995, A welfare comparison of intermediaries and financial markets in Germany and the US, European Economic Review 39, 179-209. Antão, Paula, Miguel A. Ferreira, and Ana Lacerda, 2011, Bank loans and banks' corporate control: evidence for Portugal, Economic Bulletin and Financial Stability Report Articles. von Beschwitz, Bastian, 2013, Cash Windfalls and Acquisitions, Working Paper. Beyer, Jürgen, 1999, Unternehmensverflechtungen und Managerherrschaft in Deutschland, Leviathan 27. Beyer, Jürgen, 2003, Deutschland AG a.D.: Deutsche Bank, Allianz und das Verflechtungszentrum des deutschen Kapitalismus, Wolfgang Streeck und Martin Höpner (Hrsg.) Alle Macht dem Markt? Fallstudien zur Auflösung der Deutschland AG. Boyd, John H., Chun Chang, and Bruce D. Smith, 1998, Moral Hazard under Commercial and Universal Banking, Journal of Money, Credit and Banking 30, 426-468. Cable, John, 1985, Capital Market Information and Industrial Performance: The Role of West German Banks, The Economic Journal 95, 118-132. Chamberlain, Gary, 1980, Analysis of Covariance with Qualitative Data, The Review of Economic Studies 47, pp. 225-238. Dittmann, Ingolf, Ernst Maug, and Christoph Schneider, 2010, Bankers on the Boards of German Firms: What They Do, What They Are Worth, and Why They Are (Still) There, Review of Finance 14, 35-71. Edwards, Courtney H., Mark H. Lang, Edward L. Maydew, and Douglas A. Shackelford, 2004, Germany’s Repeal of the Corporate Capital Gains Tax: The Equity Market Response. The Journal of the American Taxation Association 26(supplement), 73-97. Engelberg, Joseph, Pengjie Gao, and Christopher A. Parsons, 2012, Friends with money, Journal of Financial Economics 103, 169-188. Fang, Lily H., Ivashina, Victoria and Lerner, Josh, 2013, Combining Banking with Private Equity Investing, Review of Financial Studies 26, 2139-2173. Ferreira, Miguel A., and Pedro Matos, 2012, Universal Banks and Corporate Control: Evidence from the Global Syndicated Loan Market, Review of Financial Studies 25, 2703-2744. Flath, David, 1993, Shareholding in the Keiretsu, Japan’s Financial Groups, Review of Economics and Statistics 75, 249-257. Gormley, Todd A., and David A. Matsa, 2011, Growing Out of Trouble? Corporate Responses to Liability Risk, Review of Financial Studies 24, 2781-2821. Gorton, Gary and Frank A. Schmid, 2000, Universal banking and the performance of German firms, Journal of Financial Economics 58, 29-80. Hackethal, Andreas, 2004, German banks and banking structure, Krahnen, J.P., Schmidt, R.H. (Eds.), The German Financial System. Oxford University Press, Oxford. Haselmann, Rainer, David Schoenherr and Vikrant Vig, 2013, Lending in Social Networks, Working Paper. Hellmann, Thomas, Laura Lindsey, and Manju Puri, 2008, Building Relationships Early: Banks in Venture Capital, Review of Financial Studies 21, 513-541. Höpner, Martin, 2000, Unternehmensverflechtung im Zwielicht: Hans Eichels Plan zur Auflösung der Deutschland AG, Working Paper. 27

Höpner, Martin, and Lothar Krempel, 2006, Ein Netzwerk in Auflösung: Wie die Deutschland AG zerällt, Max Planck Institute for the Study of Societies. Hoshi, Takeo, Anil Kashyap, and David Scharfstein. 1990. ‘‘The Role of Banks in Reducing the Costs of Financial Distress in Japan’’. Journal of Financial Economics 27, 67–88. Hoshi, Takeo, Anil Kashyap, and David Scharfstein. 1991. ‘‘Corporate Structure, Liquidity, and Investment: Evidence from Japanese Industrial Groups’’. Quarterly Journal of Economics, 33–60. Ivashina, Victoria, and Zheng Sun, 2011, Institutional stock trading on loan market information, Journal of Financial Economics 100, 284-303. Jiang, Wei, Kai Li, and Pei Shao, 2010, When Shareholders Are Creditors: Effects of the Simultaneous Holding of Equity and Debt by Non-commercial Banking Institutions, Review of Financial Studies 23, 3595-3637. John, Kose, Teresa A. John, and Anthony Saunders, 1994, Universal banking and firm risk-taking, Journal of Banking & Finance 18, 307-323. Keen, Michael, 2002, The German Tax Reform of 2000, International Tax and Public Finance 9, 603-621. Kengelbach, Jens, and Alexander Roos, 2006, Entflechtung der Deutschland AG - Empirische Untersuchung der Reduktion von Kapital- und Personalverflechtungen zwischen deutschen börsennotierten Gesellschaften, M&A Review 1, 12-21. Khwaja, Asim I., and Atif Mian, 2008, Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market, American Economic Review 98, 1413-1442. Kroszner, Randall S., 2000, The Legacy of the Separation of Banking and Commerce Continues in Gramm-Leach-Bliley, The Region, 14, 1-18. Kroszner, Randall S., and Philip E. Strahan, 2001, Bankers on boards: monitoring, conflicts of interest, and lender liability, Journal of Financial Economics 62, 415-45. Laeven, Luc, and Ross Levine, 2007, Is there a diversification discount in financial conglomerates?, Journal of Financial Economics 85, 331-367. Lehmann, Erik, and Jürgen Weigand, 2000, Does the Governed Corporation Perform Better? Governance Structures and Corporate Performance in Germany, European Finance Review 4, 157- 195. Lim, Jongha, Bernadette A. Minton, and Michael S. Weisbach, 2012, Equity-Holding Institutional Lenders: Do they Receive Better Terms? National Bureau of Economic Research Working Paper Series No. 17856. Mahrt-Smith, Jan, 2006, Should banks own equity stakes in their borrowers? A contractual solution to hold-up problems, Journal of Banking & Finance 30, 2911-2929. Massa, Massimo, and Zahid Rehman, 2008, Information flows within financial conglomerates: Evidence from the banks–mutual funds relation, Journal of Financial Economics 89, 288-306. Massoud, Nadia, Debarshi Nandy, Anthony Saunders, and Keke Song, 2011, Do hedge funds trade on private information? Evidence from syndicated lending and short-selling, Journal of Financial Economics 99, 477-499. Neyman, J., and Elizabeth L. Scott, 1948, Consistent Estimates Based on Partially Consistent Observations, Econometrica 16, 1-32. Pauly, Christoph, and Ulrich Schäfer, 2002, Jäger des verlorenen Schatzes, Spiegel. Prowse, Stephen D., 1990, Institutional Investment Patterns and Corporate Financial Behavior in the United States and Japan, Journal of Financial Economics 27, 43–66. 28

Puri, Manju, Jörg Rocholl, and Sascha Steffen, 2011, Global retail lending in the aftermath of the US financial crisis: Distinguishing between supply and demand effects, Journal of Financial Economics 100, 556-578. Santos, João A., and Kristin E. Wilson, 2008, Does Banks' Corporate Control Benefit Firms? Evidence from US Banks' Control Over Firms' Voting Rights, SSRN eLibrary. Santos, João A. C., 1999, Bank capital and equity investment regulations, Journal of Banking & Finance 23, 1095-1120. Sautner, Zacharias, and Belen Villalonga, 2010, Corporate Governance and Internal Capital Markets, SSRN eLibrary. Sheard, Paul., 1989, The Main Bank System and Corporate Monitoring and control in Japan, Journal of Economic Behavior and Organization 11, 399–422. Stöss, Elmar, 2001, Deutsche Bundesbank's Corporate Balance Sheet Statistics and Areas of Application, Schmollers Jahrbuch 121, 131-137. Thompson, Samuel B., 2011, Simple formulas for standard errors that cluster by both firm and time, Journal of Financial Economics 99, 1-10. Vitols, Sigurt, 2005, German corporate governance in transition: Implications of bank exit from monitoring and control, International Journal of Disclosure and Governance 2, 357-367. Weinstein, David E. and Yishay Yafeh. 1998. On the Costs of a Bank-Centered Financial System: Evidence from the Changing Main Bank Relations in Japan. Journal of Finance 53, 635–672. 29

Figure 1: Announcement returns of the tax reform In Panel A, we display the mean returns of German banks with equity stakes around the announcement of the tax reform on December 23rd, 1999. As control groups we report German firms without equity stakes and US banks (SIC codes 6000-6199). An equity stake is defined as a holding of another firm’s equity of less than 50%, which is held either directly or through a chain of subsidiaries at the 75% threshold. In Panel B, we display the returns of the four largest publicly traded German banks. To put those returns into perspective, we also display the aggregated value of the equity stakes, the market capitalization of the bank before the tax reform and the market capitalization increase implied by the return computed as return multiplied by market capitalization. In the last column, we display the market capitalization increase as a function of the value of the equity stakes computed as market capitalization increase divided by the aggregate value of the equity stakes. Panel A: Stock returns around the announcement of the tax reform 5.50% 5.00% 4.50% 4.00% 3.50% German banks with 3.00% equity stakes 2.50% n r u 2.00% German firms without te equity stakes R 1.50% 1.00% US banks 0.50% 0.00% -0.50% 20 Dec 21 Dec 22 Dec 23 Dec 27 Dec 28 Dec 29 Dec 30 Dec (Event) -1.00% -1.50% Panel B: Announcement returns on the banks with the largest holdings Market Market Value of Equity Market Capitalization Return on 23 Capitalization Name Stakes Capitalization Increase in % of December 1999 Increase (EUR million) (EUR million) Equity Stakes (EUR million) Value Deutsche Bank AG 18,614 51,513 13.6% 7,019 38% Dresdner Bank AG 15,213 28,110 10.9% 3,077 20% Bayerische Hypo- und 6,286 28,577 7.6% 2,166 34% Vereinsbank AG Commerzbank AG 2,244 18,716 4.7% 882 39% 30

Figure 2: Divestitures of equity stakes following the tax reform This figure illustrates the divestitures of banks’ equity stakes following the tax reform. An equity stake is defined as a bank holding less than 50% in an industrial company, either directly or through a chain of subsidiaries at the 75% threshold. We display whether an equity stake held by a bank in December 1999 (before the tax reform) has been divested or maintained by December 2002 and December 2005. We define a stake as divested if the firms is still covered in the respective database (“Who owns whom?” or Hoppenstedt) and the stake has been decreased by at least 50%. We also supplement our data with regulatory data from the German Central Bank on equity stakes of banks. For 14 of 135 equity stakes the coverage ends before the bank divests the equity stake. These are set to missing and are excluded from computing these percentages. Equity stakes by divestiture (%) 100% 90% 80% 70% 60% 50% Divested 40% Maintained 30% 20% 10% 0% 1999 Tax Reform 2002 2005 31

Figure 3: Effect of the tax reform on lending This figure reports the average log loan size for bank-firm pairs with and without equity stakes. The analysis includes only banks and firms with equity stakes. The vertical line indicates the announcement of the tax reform in December 1999. Loan Size (log) 10 9.5 9 Without Equity Link With Equity Link 8.5 8 7.5 1998 1999 2000 2001 2002 2003 2004 2005 32

Figure 4: Lending around divestitures This figure reports the average Bank Share for bank-firm pairs with equity links around the time of the divestiture of the equity link. Bank Share is defined as the percentage of the firms’ borrowing that the linked bank provides. The years are measured relative to the time of divestiture. Year zero is the year in which the equity stake was divested. Bank Share for bank-firm pairs with equity link 30% 28% 26% 24% e 22% r a h 20% S k 18% n a 16% B 14% 12% 10% -3 -2 -1 0 1 2 Years relative to divestiture 33

Figure 5: Size of banks with equity stakes This figure illustrates the size of the banks that own equity stakes (Inside Sample) compared to other German banks (excluding special banks such as mortgage banks). Bank Size is defined as total assets in 1999 taken from the Bankscope database. We plot the 50 largest banks according to this measure as well as the four banks that own equity stakes and are not amongst the largest 50 banks. Top 50 banks by size (assets) 600 Aggregated Market Share 500 Banks with equity stakes: 59% Banks without equity stakes: 41% 400 ) n o illib 300 With Equity Stakes ( s t No Equity Stake e s s A 200 100 0 123456789012345678901234567890123456789012345678901171 111111111122222222223333333333444444444457123 135 Size Rank 1 34

Figure 6: Comparison of firms with and without equity links This figure compares firms with banks as equity holders to a control sample matched by assets and whether the company is publicly traded. We match each company having a bank as equity holder to the next smaller and next larger company by assets within the group of public or private companies respectively (due to two firms with equity links being next to each other, the control sample is not exactly twice as large). In Panel A, we compare the firms based on 12 Fama-French industry groups. In Panel B, we report the means and medians for firm-specific control variables in 1999. Firm Leverage is defined as book value of debt divided by total assets. Firm Tangibility of Assets is defined as cash and equivalents plus net PPE divided by total assets. Number of Bank Relationships is the number of banks from which the company receives a loan. Return on Assets is EBIT divided by total assets. Profit Margin is net income divided by sales. Asset Turnover is sales divided by assets. We also report a Student t-test for the difference in means and a Wilcoxon rank-sum test for the difference in medians. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Panel A: 12 Fama-French industry groups 35.0% 30.0% 25.0% 20.0% 15.0% With bank as equity holder 10.0% Without bank as equity holder 5.0% 0.0% Panel B: Summary Statistics Mean Median No bank as Bank as T-test of No bank as Bank as Z-test of equity holder equity holder difference equity holder equity holder difference Firm Leverage 19.9 22.1 -0.96 15.2 21.0 -1.52 Firm Tangibility of Assets (%) 37.3 33.6 1.41 36.8 35.1 1.25 Number of Bank Relationships 14.1 12.8 0.20 5 7 0.15 Return on Assets (%) 5.71 5.53 0.13 5.37 5.85 -0.14 Profit Margin (%) 7.08 6.85 0.093 5.30 4.72 0.29 Asset Turnover (per year) 1.35 1.31 0.27 1.11 1.17 -0.50 Observations 207 117 207 117 35

Table 1: Summary statistics This table displays summary statistics as of December 1999. Panel A covers variables on the firm level for all firms in which a bank holds an equity stake. Panel B reports the summary statistics on the bank level for the 26 banks which hold equity in an industrial company. Number of Equity stakes is the number of industrial companies in which the bank holds equity. Value of Equity Stakes is the estimated market value of these stakes, while Equity Stakes / Equity is this value divided by the bank’s equity. Panel C reports the summary statistics on the equity stake level. The size of the stake is given in EUR million and as a percentage of the company’s equity. Existence of a Loan is a dummy variable indicating whether the bank holding the equity gives a loan to the firm. Debt Share of Bank Funding gives the loan value from the bank to the company divided by the total funding (debt + equity) taking into account only the 89 equity stakes with a loan. Board Link 1999 is an indicator variable equal to one if a representative of the bank sits on the firm’s supervisory board. Panel D reports the summary statistics on the loan level. Following Khwaja and Mian (2008), we refer to a bank-firm pair as a loan, i.e. multiple loans are aggregated. All other variables are defined in Appendix 1. Panel A: Firms 25th 75th Standard Variable Mean Median Percentile Percentile Deviation Assets (EUR million) 9128.8 41.3 303.7 1697.7 41375.0 Firm Leverage (%) 22.1 6.68 21.0 35.2 16.1 Firm Tangibility of Assets (%) 33.6 21.9 35.1 43.4 15.9 Number of Bank Relationships 12.8 2 7 14 16.9 Publicly Listed 0.50 0 0 1 0.50 Return on Assets (%) 5.53 3.05 5.85 7.91 6.99 Observations 117 Panel B: Banks 25th 75th Standard Variable Mean Median Percentile Percentile Deviation Assets (EUR billion) 107.0 12.4 53.3 207.6 128.1 Equity (EUR million) 3336.8 550.4 1523.7 4820.5 4443.2 Number of Loans 5986.3 1271 3135 7187 7110.5 Number of Equity Stakes 5.19 1 1.50 9 6.25 Value of Equity Stakes (EUR million) 2382.3 5.52 36.4 386.8 6393.1 Equity Stakes / Equity (%) 22.8 0.76 3.75 18.3 45.6 Lending Focus of Bank (%) 43.4 33.7 43.4 54.4 15.5 Return on Equity (%) 5.44 3.39 4.75 7.60 3.65 Observations 26 Panel C: Equity stakes 25th 75th Standard Variable Mean Median Percentile Percentile Deviation Size of Equity Stake (%) 14.6 5.52 10.5 21.6 12.0 Size of Equity Stake (EUR million) 458.8 3.31 17.3 97.6 1549.2 Existence of a Loan 0.66 0 1 1 0.48 Debt Share of Bank Funding (in case of loan, %) 35.9 6.06 32.2 58.1 29.9 Board Link 1999 0.75 0.5 1 1 0.435 Observations 135 Panel D: Loans 25th 75th Standard Variable Mean Median Percentile Percentile Deviation Loan Amount (EUR million) 35.3 2.86 10.1 30.5 83.5 Bank Share (%) 14.0 1.79 6.36 15.8 21.1 Observations 492 36

Table 2: Comparison to other studies: Cross-sectional correlation This table displays cross-sectional regressions as of December 1999, examining whether equity links are correlated with the existence and the size of a lending relationship. The analysis includes only banks and firms with equity stakes. In Panel A, the observational unit is a bank-firm pair. The dependent variable is a dummy variable equal to one if the bank provides a loan to the firm in December 1999. In Regressions 1, we estimate a logit model. In Regressions 2 to 4 we estimate a linear probability model, i.e. OLS. In Panel B, the observational unit is a loan (i.e. a bank-firm pair in which the bank provides credit to the firm). The dependent variables are Loan Size (log), Loan Size (EUR million), Bank Share, and Borrower Percentile. The explanatory variable of interest is a dummy variable equal to one if the bank holds an equity stake in the company. All standard errors are double-clustered at the bank and the firm level. We report t-statistics below the coefficients in parenthesis. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Panel A: Lending Relationship Lending Relationship (1) (2) (3) (4) Equity Link 1999 1.944*** 0.543*** 0.352*** 0.248*** (6.01) (11.11) (8.27) (3.54) Same City 1.534** 0.125* 0.181*** 0.134*** (2.26) (1.71) (2.95) (2.77) Same Region 0.516 0.104 0.068 0.104 (0.70) (1.35) (1.02) (1.56) Board Link 1999 0.113*** (2.78) Firm Size (log) 0.431*** (6.57) Firm Leverage 2.820*** (3.46) Firm Tangibility of Assets 0.421 (0.47) Observations 2250 2756 2756 1976 Adjusted R2 0.38 0.25 0.40 0.43 Regression Method Logit OLS OLS OLS Firm Fixed Effects No Yes Yes Yes Bank Fixed Effects Yes No Yes Yes Economic effect based on regression 1: 18.7% (no hold) 44.1% (hold) at means: 7.6% (no hold) 38.1% (hold) Panel B: Size of Lending Relationship Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link 1999 1.156*** 0.961*** 28.733*** 21.876*** 0.093*** 0.089*** 0.149*** 0.110*** (3.66) (3.23) (2.88) (2.86) (2.83) (2.85) (3.74) (2.70) Same City 0.056 0.531 -12.158 0.050 -0.015 -0.007 0.023 0.078 (0.12) (0.82) (-0.92) (0.00) (-0.30) (-0.12) (0.36) (1.05) Same Region 0.078 -0.419 14.129 0.155 0.003 -0.001 -0.038 -0.075 (0.09) (-0.41) (0.56) (0.01) (0.04) (-0.01) (-0.36) (-0.64) Observations 468 468 468 468 468 468 468 468 Adjusted R2 0.24 0.32 0.21 0.25 0.71 0.72 0.19 0.24 Firm Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Bank Fixed Effects No Yes No Yes No Yes No Yes 37

Table 3: Impact of the tax reform on lending: Difference-in-difference This table presents difference-in-difference regressions examining the effect of the tax reform on lending. The observational unit in this regression is the bank-firm-quarter and the time period spans from 1998 to 2005. A bank-firm pair is only included if there is a lending relationship before the tax reform (average lending from 1998 to 1999 is positive). The dependent variables are Loan Size (log), Loan Size (EUR million), Bank Share, and Borrower Percentile. Variables are defined in Appendix 1. In all regressions, we include fixed effects for the bank-firm pair and quarter. In Regressions 1, 3, 5 and 7 we add firm–post reform fixed effects. In regressions 2, 4, 6 and 8 we add bank-year and firm-year fixed effects. In Panel A, the main explanatory variable of interest is the interaction between Equity Link 1999 and Post Reform. In Panel B, we exclude all bank-firm relationships with equity stakes that were not sold or for which divestiture information is missing. All standard errors are double-clustered at the bank and the firm level. We report t-statistics below the coefficients in parenthesis. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Panel A: All Equity Stakes Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link 1999 * Post Reform 0.128 0.123 14.858*** 15.803*** 0.020 0.020 0.002 0.008 (0.93) (0.87) (3.61) (3.90) (1.10) (1.04) (0.07) (0.27) Same City * Post Reform -0.375 -0.382 -13.347 -10.312 -0.050 -0.057 -0.063 -0.058 (-0.48) (-0.41) (-1.13) (-0.65) (-1.19) (-1.30) (-1.22) (-0.82) Same Region * Post Reform -0.232 -0.237 8.430 2.052 0.038 0.045 0.015 0.000 (-0.30) (-0.26) (0.51) (0.11) (0.89) (1.01) (0.29) (0.00) Observations 13327 13327 18554 18554 18554 18554 18554 18554 Adjusted R2 0.66 0.71 0.65 0.68 0.70 0.74 0.66 0.72 Bank-Firm and Quarter F.E. Yes Yes Yes Yes Yes Yes Yes Yes Firm–Post Reform Fixed Effects Yes No Yes No Yes No Yes No Firm-year and bank-year F.E. No Yes No Yes No Yes No Yes Panel B: Only Divested Equity Stakes Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Divested Equity Link * Post Reform 0.136 0.210 14.733*** 15.434*** 0.016 0.025 0.000 0.018 (0.81) (1.34) (2.87) (2.88) (0.77) (1.12) (0.01) (0.62) Same City * Post Reform -0.288 -0.384 -14.087 -11.015 -0.053 -0.064 -0.066 -0.063 (-0.36) (-0.39) (-1.15) (-0.68) (-1.22) (-1.39) (-1.23) (-0.84) Same Region * Post Reform -0.303 -0.221 4.400 -2.868 0.030 0.037 0.025 0.006 (-0.37) (-0.24) (0.24) (-0.14) (0.70) (0.83) (0.48) (0.08) Observations 12532 12532 17640 17640 17640 17640 17640 17640 Adjusted R2 0.66 0.70 0.64 0.67 0.67 0.72 0.66 0.71 Bank-Firm and Quarter F.E. Yes Yes Yes Yes Yes Yes Yes Yes Firm–Post Reform Fixed Effects Yes No Yes No Yes No Yes No Firm-year and bank-year F.E. No Yes No Yes No Yes No Yes 38

Table 4: Impact of divestitures on lending This table presents difference-in-difference regressions examining the effect of divestitures on lending. The observational unit in this regression is the bank-firm-quarter and the time period spans from 1998 to 2005. A bank-firm pair is only included if there is a lending relationship before the tax reform (average lending from 1998 to 1999 is positive). The dependent variables are Loan Size (log), Loan Size (EUR million), Bank Share, and Borrower Percentile. Variables are defined in Appendix 1. In all regressions, we include fixed effects for the bank-firm pair and quarter. In Regressions 1, 3, 5 and 7 we add firm–post reform fixed effects. In regressions 2, 4, 6 and 8 we add bank-year and firm-year fixed effects. In Panel A, the main explanatory variable of interest is the interaction between Equity Link 1999 and Post Reform. The main explanatory variable is Equity Link is Divested, which is equal to 1 after an equity has been divested and 0 before it is divested or if there is no equity link. All standard errors are double-clustered at the bank and the firm level. We report t-statistics below the coefficients in parenthesis. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link is Divested -0.002 0.064 4.238 3.405 0.004 0.015 -0.001 0.014 (-0.01) (0.45) (0.59) (0.47) (0.17) (0.82) (-0.03) (0.45) Same City * Post Reform -0.282 -0.362 -12.994 -8.734 -0.049 -0.055 -0.060 -0.057 (-0.37) (-0.39) (-1.12) (-0.54) (-1.15) (-1.23) (-1.15) (-0.79) Same Region * Post Reform -0.311 -0.241 11.258 4.454 0.040 0.048 0.012 -0.001 (-0.40) (-0.27) (0.73) (0.24) (0.90) (1.04) (0.24) (-0.01) Observations 13119 13119 18316 18316 18316 18316 18316 18316 Adjusted R2 0.67 0.71 0.66 0.68 0.69 0.74 0.67 0.71 Bank-Firm and Quarter F.E. Yes Yes Yes Yes Yes Yes Yes Yes Firm–Post Reform Fixed Effects Yes No Yes No Yes No Yes No Firm-year and bank-year F.E. No Yes No Yes No Yes No Yes 39

Table 5: Sample splits This table presents difference-in-difference regressions examining the effect of the tax reform on lending. The observational unit in this regression is the bank-firm-quarter and the time period spans from 1998 to 2005. A bank-firm pair is only included if there is a lending relationship before the tax reform (average lending from 1998 to 1999 is positive). The dependent variables are Loan Size (log), Loan Size (EUR million), Bank Share, and Borrower Percentile. Variables are defined in Appendix 1. In all regressions, we include fixed effects for the bank-firm pair and quarter. In Regressions 1, 3, 5 and 7 we add firm–post reform fixed effects. In regressions 2, 4, 6 and 8 we add bank-year and firm-year fixed effects. In Panel A, we split firms with equity links in 1999 into those that also had a board link to the bank and those that did not. In Panel B, we split firms with equity links in 1999 into two groups depending on whether the firm is publicly listed. All standard errors are double-clustered at the bank and the firm level. We report t-statistics below the coefficients in parenthesis. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Panel A: Split by board link Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link with Board Link 1999 * Post Reform 0.114 0.109 14.651** 14.946** -0.006 -0.002 0.001 0.006 (0.60) (0.55) (2.31) (2.04) (-0.26) (-0.06) (0.04) (0.16) Equity Link without Board Link 1999 * Post Reform 0.158 0.154 15.262* 17.474** 0.070** 0.063*** 0.002 0.011 (0.97) (0.85) (1.69) (1.97) (2.53) (2.67) (0.06) (0.30) Same City * Post Reform -0.378 -0.385 -13.370 -10.380 -0.053 -0.058 -0.063 -0.058 (-0.49) (-0.41) (-1.13) (-0.66) (-1.28) (-1.35) (-1.21) (-0.82) Same Region * Post Reform -0.229 -0.234 8.451 2.096 0.040 0.046 0.015 0.000 (-0.29) (-0.26) (0.52) (0.11) (0.97) (1.05) (0.29) (0.00) Observations 13327 13327 18554 18554 18554 18554 18554 18554 Adjusted R2 0.67 0.71 0.66 0.68 0.70 0.74 0.67 0.72 Bank-Firm and Quarter F.E. Yes Yes Yes Yes Yes Yes Yes Yes Firm–Post Reform Fixed Effects Yes No Yes No Yes No Yes No Firm-year and bank-year F.E. No Yes No Yes No Yes No Yes Panel B: Split into public and private firms Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Public Equity Link 1999 * Post Reform 0.096 0.101 17.219*** 18.423*** 0.019* 0.028** 0.007 0.022 (0.66) (0.65) (2.87) (2.90) (1.70) (2.09) (0.28) (0.73) Private Equity Link 1999 * Post 0.233 0.199 8.928 9.569 0.021 0.002 -0.012 -0.025 Reform (0.56) (0.48) (1.11) (1.07) (0.37) (0.04) (-0.19) (-0.39) Same City * Post Reform -0.370 -0.379 -13.671 -10.676 -0.050 -0.058 -0.064 -0.060 (-0.48) (-0.41) (-1.16) (-0.67) (-1.20) (-1.32) (-1.23) (-0.84) Same Region * Post Reform -0.234 -0.236 8.499 2.014 0.038 0.045 0.015 0.000 (-0.30) (-0.26) (0.52) (0.11) (0.89) (1.00) (0.30) (0.00) Observations 13327 13327 18554 18554 18554 18554 18554 18554 Adjusted R2 0.67 0.71 0.66 0.68 0.70 0.74 0.67 0.72 Bank-Firm and Quarter F.E. Yes Yes Yes Yes Yes Yes Yes Yes Firm–Post Reform Fixed Effects Yes No Yes No Yes No Yes No Firm-year and bank-year F.E. No Yes No Yes No Yes No Yes 40

Table 6: Robustness check: Less Fixed Effects This table presents a robustness check to Table 3 with less fixed effects. The observational unit in this regression is the bankfirm-quarter and the time period spans from 1998 to 2005. A bank-firm pair is only included if there is a lending relationship before the tax reform (average lending from 1998 to 1999 is positive). The dependent variables are Loan Size (log), Loan Size (EUR million), Bank Share, and Borrower Percentile. Variables are defined in Appendix 1. In all regressions, we include fixed effects for the bank-firm pair and quarter. In regressions 2, 4, 6 and 8 we add bank-Post Reform fixed effects. In Panel A, the main explanatory variable of interest is the interaction between Equity Link 1999 and Post Reform. In Panel B, we exclude all bank-firm relationships with equity stakes that were not sold or for which divestiture information is missing. All standard errors are double-clustered at the bank and the firm level. We report t-statistics below the coefficients in parenthesis. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Panel A: All Equity Stakes Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link 1999 * Post Reform 0.158 0.230 11.173*** 12.727*** 0.021 0.023 -0.000 0.004 (0.90) (1.26) (3.83) (3.33) (1.09) (1.10) (-0.01) (0.12) Same City * Post Reform -0.388 -22.328 -0.039 -0.189* (-0.40) (-1.61) (-1.54) (-1.65) Same Region * Post Reform 0.310 18.692 0.033** 0.206* (0.30) (1.27) (2.03) (1.68) Observations 13327 13327 18554 18554 18554 18554 18554 18554 Adjusted R2 0.63 0.63 0.63 0.64 0.68 0.68 0.64 0.64 Bank-Firm and Quarter F.E. Yes Yes Yes Yes Yes Yes Yes Yes Bank–Post Reform Fixed Effects No Yes No Yes No Yes No Yes Panel B: Only Divested Equity Stakes Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Divested Equity Link * Post Reform 0.186 0.230 11.445** 12.919** 0.010 0.018 0.019 0.023 (0.96) (1.06) (2.20) (2.42) (0.48) (0.77) (0.75) (0.79) Same City * Post Reform -1.027 -22.840 -0.052** -0.219* (-1.17) (-1.51) (-2.27) (-1.82) Same Region * Post Reform 1.034 15.268 0.030* 0.250** (1.21) (0.85) (1.81) (1.99) Observations 12532 12532 17640 17640 17640 17640 17640 17640 Adjusted R2 0.63 0.63 0.62 0.63 0.65 0.65 0.63 0.63 Bank-Firm and Quarter F.E. Yes Yes Yes Yes Yes Yes Yes Yes Bank–Post Reform Fixed Effects No Yes No Yes No Yes No Yes 41

Table 7: Robustness check: Different control groups and regression set up This table presents regressions examining the effect of the tax reform on lending using different control groups. The dependent variables are differences averages between the pre and post-tax reform period (following Khwaja and Mian (2008)). The prereform period contains the average over the quarters in 1998 and 1999, while the post-reform period contains the average over all quarters from 2000 to 2005. The regression includes only bank-firm pairs for which pre-tax reform lending is positive. In Panel A, the sample consists only of firms and banks that have an equity link (as in Tables 3). In Panel B, the sample consists of all banks and companies in our wider sample, i.e. banks that are covered in Who owns whom? and have aggregate average lending of more than EUR 1 billion before the tax reform (1998 to 1999) as well as firms that are covered in Who owns whom? and have assets above 5 million EUR. In Panel C, we exclude non-linked loans of companies with equity links. In Panel A and Panel B, we report specifications with bank fixed effects as well as specifications with bank and firm fixed effects. In Panel C, we replace firm fixed effects with firm controls (because firm fixed effects cannot be estimated). In all regressions, we include three sets of loan size quartile fixed effects based on Loan Size, Bank Share and Borrower Percentile before the tax reform (average from 1998 to 1999). All standard errors are double-clustered at the bank and the firm level. We report tstatistics below the coefficients in parenthesis. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Panel A: Banks and firms with equity stakes Difference in Difference in Difference in Difference in Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link 1999 0.213 0.153 23.683*** 23.797*** 0.027 0.027 0.020 0.030 (1.17) (0.91) (3.24) (3.74) (1.41) (1.30) (1.29) (1.32) Same City -0.550 -0.883 -6.596 -4.115 -0.072 -0.078 -0.071 -0.070 (-0.84) (-1.29) (-0.77) (-0.33) (-1.33) (-1.35) (-1.37) (-0.94) Same Region -0.311 0.006 8.853 3.867 0.061 0.068 0.019 0.014 (-0.39) (0.01) (0.64) (0.25) (1.18) (1.20) (0.40) (0.20) Observations 541 541 608 608 608 608 608 608 Adjusted R2 0.12 0.13 0.16 0.16 0.06 0.08 0.18 0.22 Firm Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Bank Fixed Effects No Yes No Yes No Yes No Yes Panel B: All banks and firms Difference in Difference in Difference in Difference in Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link 1999 0.135 0.151 6.608*** 6.365*** 0.028 0.022 0.042** 0.035* (0.78) (0.88) (4.06) (4.02) (1.43) (1.08) (2.34) (1.89) Same City -0.192** -0.238** -0.023 -0.365 0.010 0.002 -0.003 -0.005 (-1.99) (-1.98) (-0.05) (-0.65) (0.84) (0.22) (-0.46) (-0.68) Same Region 0.195* 0.209* 0.926 1.495** 0.001 0.010 -0.002 0.006 (1.87) (1.70) (1.58) (2.05) (0.07) (0.72) (-0.34) (0.56) Observations 9027 9027 11024 11024 11024 11024 11024 11024 Adjusted R2 0.11 0.12 -0.02 0.02 0.17 0.20 0.02 0.05 Firm Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Bank Fixed Effects No Yes No Yes No Yes No Yes 42

Panel C: Only different firms Difference in Difference in Difference in Difference in Loan Size Loan Size (log) Bank Share Borrower Percentile (EUR million) (1) (2) (3) (4) (5) (6) (7) (8) Equity Link 1999 0.456*** 0.482*** 7.455*** 7.442*** 0.014 0.012 0.025 0.022 (3.10) (3.17) (4.74) (4.79) (0.83) (0.66) (1.49) (1.30) Same City -0.048 -0.031 -0.362 -0.533 0.013* 0.012* -0.007 -0.010 (-0.73) (-0.36) (-0.84) (-1.17) (1.82) (1.65) (-1.22) (-1.56) Same Region 0.156** 0.051 0.690 0.817 0.003 0.003 0.003 0.015 (2.34) (0.62) (1.31) (1.49) (0.24) (0.23) (0.64) (1.28) Firm Size (log) 0.007 0.012 -0.005 0.040 0.009*** 0.010*** -0.005** -0.005** (0.57) (1.01) (-0.04) (0.34) (4.22) (4.96) (-2.45) (-2.43) Firm Leverage -0.060 -0.081 -0.590 -0.772 0.033*** 0.030*** -0.001 0.000 (-0.57) (-0.79) (-0.95) (-1.31) (3.66) (4.07) (-0.11) (0.01) Firm Tangibility of Assets 0.263*** 0.232*** 1.334** 1.076* -0.017* -0.021** -0.004 -0.005 (3.45) (2.90) (2.07) (1.96) (-1.79) (-2.33) (-0.53) (-0.77) Observations 9380 9380 11101 11101 11101 11101 11101 11101 Adjusted R2 0.00 0.01 0.01 0.04 0.01 0.04 0.02 0.04 Bank Fixed Effects No Yes No Yes No Yes No Yes 43

Table 8: Equity repurchases following the tax reform This table presents regressions examining the effect of the tax reform on the number of equity repurchases depending on whether the firm has a bank as an equity holder. The dependent variable in this regression is a dummy variable equal to one if there was an increase in the number of treasury shares in a year and equal to zero if the change was zero (and neither value was missing). We base this variable first on the number of treasury shares from USTAN/Jalys. If it cannot be determined (because either value is missing), we use data from Compustat Global instead. In Regressions 1, we estimate a logit model only after the tax reform. In Regressions 2 to 4, we estimate a difference-in-difference set up using a linear probability model, i.e. OLS. In regression 4, we add 12 Fama-French Industry fixed effects interacted with Post Reform. All standard errors are clustered at the firm level. We report t-statistics below the coefficients in parenthesis. *, **, *** indicate significance at the 10%, 5%, and 1% level, respectively. Repurchase Dummy (1) (2) (3) (4) Equity Link 1999 0.019 (0.05) Equity Link 1999 * Post Reform 0.001 -0.008 -0.012 (0.03) (-0.25) (-0.39) Firm Size (log) 0.121* (1.85) Firm Leverage -0.151 (-0.16) Firm Tangibility of Assets -1.722* (-1.91) Firm Size (log) * Post Reform 0.000 0.000 (0.05) (0.02) Firm leverage * Post Reform 0.009 0.033 (0.14) (0.47) Firm Tangibility of Assets * Post Reform -0.172** -0.110 (-2.37) (-1.41) Observations 1047 1486 1473 1473 Adjusted R2 0.03 0.20 0.20 0.20 Sample Period After Tax Reform Full Sample Full Sample Full Sample Regression Method Logit OLS OLS OLS Firm Fixed Effects No Yes Yes Yes Industry–Post Reform Fixed Effects No No No Yes Probability change based on regression 1: 8.25% to 8.4% 7.65% to 7.78% (at means) 44

Appendix 1: Variable definitions This table displays the variable definitions for all variables used in the regressions. Panel A: Bank-firm-quarter-level variables Variable Name Definition Loan Size The aggregate exposure of a bank to a firm at the end of each quarter from the German Large Credit Register (Millionenkreditevidenz). It includes on-balance sheet lending such as loans and bonds as well as off-balance sheet exposures through guarantees, derivatives and undrawn credit lines. Loan Size is set to missing if either the firm or the bank are not in the database for other loans in that quarter. Loan Size (log) Log (𝐿𝑜𝑎𝑛 𝑠𝑖𝑧𝑒) Where log is the natural logarithm and the size of the loan is measured in EUR. Loan Size (EUR million) Loan size in EUR million. This variable is winsorized at the 1% and 99% threshold. Bank Share 𝐿𝑜𝑎𝑛 𝑠𝑖𝑧𝑒 𝑆𝑢𝑚 𝑜𝑓 𝑎𝑙𝑙 𝑙𝑜𝑎𝑛𝑠 𝑜𝑓 𝑎 𝑓𝑖𝑟𝑚 All loan data for this variable comes from the German Large Credit Register. In the sum of all loans we include also loans of banks that are not in our sample. Bank Share is set to missing if either the firm or the bank are not in the database for other loans in that quarter. Borrower Percentile We sort all borrowers of a bank by the average size of their loans in the quarter and assign percentiles with the largest borrower receiving 100% and the smallest borrower 0%. We include only borrowers from the respective sample. If a firm does not borrow from this bank, but both the bank and the firm are in the database, we assign a zero. Equity Link is Divested Dummy variable equal to 1 in the years after an equity stake was divested. More specifically: If Equity Link 1999 is equal to 1 and we know the divestiture time, this variable is 0 before the divestiture and 1 starting in the year when it was divested. If Equity Link 1999 is equal to 0, this variable is equal to 0. If Equity Link 1999 is equal to 1 and the divestiture time is missing, this variable is missing. Panel B: Bank-firm-level variables Variable Name Definition Lending Relationship Dummy variable equal to one if the company receives a loan from the bank in December 1999. Set to missing if either the firm or the bank is not covered in the German Large Credit Register in December 1999. Equity Link 1999 Dummy variable equal to one if the bank holds an industrial firm’s equity of less than 50%, which is held either directly or through a chain of subsidiaries at the 75% threshold. Unless stated otherwise, we take these holdings as of December 1999. We use data from the “Who owns Whom?” database and manually add holdings of banks reported in Hoppenstedt Aktienführer 2001 (these holdings are as of December 1999). We exclude holdings in other banks, private equity companies and vehicles of project finance. Divested Equity Link This variable is the same as Equity Link 1999, but is set to missing if the equity stake was not divested by 2005 or the divestiture information is missing. Same City Dummy variable equal to one if the bank is located in the same city as the firm (based on the address in “Who owns Whom?”). Same Region Dummy variable equal to one if the bank is located in the same region as the firm. Regions are based on the first two digits of the German postal code (based on the address in “Who owns Whom?”). Board Link 1999 Board Link 1999 is a dummy variable equal to one if a representative of the bank sits on the firm’s supervisory board. As bank representative we count: any employee of the bank (in almost all cases they are members of the bank’s management board), any former management board member of the bank that does not have a new job and the chairman of the supervisory board of the bank if he does not have a fulltime job at another company. This variable is missing if we do not have information on the firm’s supervisory board composition. Equity Link with Board Link Dummy variable equal to 1 if Equity Link 1999 and Board Link 1999 are equal to 1. It is zero for all 1999 other bank-firm pairs. Equity Link without Board Dummy variable equal to 1 if Equity Link 1999 is equal to 1 and Board Link 1999 is not equal to 1. It Link 1999 is zero for all other bank-firm pairs. Public Equity Link 1999 Dummy variable equal to 1 if Equity Link 1999 is equal to 1 and the firm is publicly listed. It is zero for all other bank-firm pairs. Private Equity Link 1999 Dummy variable equal to 1 if Equity Link 1999 is equal to 1 and the firm is not publicly listed. It is zero for all other bank-firm pairs. Difference in Loan Size (log) 𝑀𝑒𝑎𝑛 (log (𝐿𝑜𝑎𝑛 𝑆𝑖𝑧𝑒))−𝑚𝑒𝑎𝑛 (log (𝐿𝑜𝑎𝑛 𝑆𝑖𝑧𝑒)) 2000−2005 1998−1999 Difference in Loan Size (EUR 𝑀𝑒𝑎𝑛 (𝐿𝑜𝑎𝑛 𝑆𝑖𝑧𝑒)−𝑚𝑒𝑎𝑛 (𝐿𝑜𝑎𝑛 𝑆𝑖𝑧𝑒). Then this variable is winsorized at the 1% 2000−2005 1998−1999 million) and 99% thresholds. Difference in Bank Share 𝑀𝑒𝑎𝑛 (𝐵𝑎𝑛𝑘 𝑆ℎ𝑎𝑟𝑒)−𝑚𝑒𝑎𝑛 (𝐵𝑎𝑛𝑘 𝑆ℎ𝑎𝑟𝑒) 2000−2005 1998−1999 Difference in Borrower 𝑀𝑒𝑎𝑛 (𝐵𝑜𝑟𝑟𝑜𝑤𝑒𝑟 𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑖𝑙𝑒)−𝑚𝑒𝑎𝑛 (𝐵𝑜𝑟𝑟𝑜𝑤𝑒𝑟 𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑖𝑙𝑒) 2000−2005 1998−1999 Percentile 45

Panel C: Firm-level variables and firm-quarter level variables Variable Name Definition Firm Size (log) Log (total firm assets in 1999) Log is the natural logarithm and firm assets are taken from Worldscope for publicly listed companies and from JALYS/USTAN for privately listed companies. Firm Leverage The firm’s book value of debt divided by the firm’s total assets in (both in 1999). Firm Tangibility of Assets The firm’s cash and equivalents plus net PPE divided by the firm’s total assets (both in 1999). Repurchase Dummy Dummy variable equal to one if there was an increase in the number of treasury shares in a year and equal to zero if the change was zero (and neither value was missing). We base this variable first on number of treasury shares from USTAN/Jalys. If it cannot be determined (because either value is missing), we use data from Compustat Global instead. Panel D: Other variables and definitions Variable Name Definition Post Reform Dummy variable equal to one for quarters from 2000 to 2005 and equal to zero in 1998 and 1999. Value of Equity Stake We determine the value of an equity stake by multiplying the firm’s market capitalization with the share that the bank holds. If the firm is privately listed, we estimate the market capitalization using a multiple of its book equity, where this multiple is the median of the book to equity ratio of publicly listed companies. 46

Appendix 2: The German banking system The German banking system consists mostly of universal banks offering a broad range of financial services, including deposits, loans, payment services, and securities transactions. They can be classified into commercial sector banks, public sector banks and cooperative sector banks (Hackethal (2004)).13 Specialized banks (such as mortgage banks, building societies, securities brokers etc) only account for a market share of around 20% in terms of total business volume and are not covered in our data. Commercial sector banks are organized on the basis of private-sector principles with a clear forprofit orientation. This sector includes large universal banks, which are usually listed on the German stock exchange, regional banks and smaller “private bankers”. Compared to others, commercial sector banks have a stronger market position in the securities trading and underwriting business, and they generally provide all domestic and international banking services. Public sector banks include Landesbanken and savings banks (“Sparkassen”), which are entities under public law with ultimately public ownership. Savings banks are smaller institutions with a regionally specified mandate and a less sophisticated business model. On the other hand, Landesbanken are larger and usually operate in one or more of Germany’s 16 states and offer more complex financial products. Furthermore, it is important to note that many savings banks are not allowed to take equity stakes in industrial firms due to their respective articles of association. Finally, the cooperative sector consists of small credit cooperatives with a regional orientation as well as their head institutions. Compared to commercial sector banks, institutions in the public or cooperative sector rely more strongly on retail than on wholesale funding and on retail and small business lending rather than on lending to large businesses. While the commercial banking sector has the largest market share (around 40% of total business volume according to statistics of the German Central Bank), it is relatively concentrated. In contrast, more than three-quarters of German banks are in the public sector (with a market share of 35%) or cooperative sector (with a market share of 12%). Appendix 3: German corporate governance 13 Please note that “public” refers to “state- owned” and not to “publicly listed”. 47

Germany has a two-tier board system. The management board (Vorstand) is responsible for operative decisions and the supervisory board (Aufsichtsrat) is responsible for representing the interests of shareholders and employees. All stock companies (Aktiengesellschaften) must have a supervisory board, even if they are not publicly listed. A limited company (Gesellschaft mit beschränkter Haftung (GmbH)) must have a supervisory board if it has more than 500 employees and can always have a supervisory board voluntarily. The supervisory board consists of shareholder representatives and representatives of employees. If the company has above 2,000 employees, there are an equal number of shareholder and employee representatives. If there are between 500 and 2000 employees, employee representatives make up one-third of the supervisory board. Below 500 employees, there are no employee representatives. In either case, the chairman of supervisory board (Aufsichtsratsvorsitzender) decides in the case of a tied vote. Thus theoretically, shareholder representatives can always overrule employee representatives if they vote in unison. Appendix 4: Proxy voting In Germany, banks are allowed to exercise voting rights on behalf of their customers. Banks indicate to the customer how they plan to vote and the customer can instruct the bank to vote differently. Since most small customers do not use this option, the bank controls the voting right de facto.This proxy-voting is often seen as an additional channel of bank influence (e.g. Gorton and Schmid (2000)). We examine the extent of proxy voting using data from the German Central Bank on the shares that banks hold in custody (Depotstatistik). The data is only available from 2005. Since private investors stock market participation did not dramatically change in this time period, the 2005 values should be a good approximation for 1999. On average, all private customers of a bank combined hold only 0.66% of shares in a sample company (median is 0.05%, 90th percentile is 1.5%). Thus, even if none of the private customers voted themselves, the importance of proxy-voting was small compared to the equity stakes, which had an average size of 14.6% (median 10.5%). 48