Are Universal Banks Better Intermediaries? ∗ Daniel Neuhann Farzad Saidi

advertisement

Are Universal Banks Better Intermediaries?

Daniel Neuhann

University of Pennsylvania

Farzad Saidi

University of Cambridge

October 30, 2014

Abstract

Are banks of wide scope better intermediaries? Using the variation in bank scope generated by the stepwise repeal of the Glass-Steagall Act in the U.S. and the subsequent rise of universal banking, we provide evidence that economies of scope in concurrent lending and underwriting improve the access to finance for risky ventures of publicly traded companies.

Exploiting a bank-level deregulatory shock, as well as detailed data on bank-firm interactions, we identify increases in sales-growth, stock-return, and option-implied volatilities for universal-bank-financed firms. These firms also exhibit lasting increases in total factor productivity of 3 to 4%, echoed by similar findings for increases of 6 to 7% in capital expenditure and 5 to 9% in market capitalization. Our findings suggest that the facilitation of cross-selling of loans and non-loan products through bank-scope deregulation may have led to an increase in the supply of credit for firms making risky, productivity-increasing investments.

JEL classification: E20, G20, G21

Keywords: universal banking, financial deregulation, bank scope, firewalls, cross-selling

We thank Hal Cole, Alexandra Effenberger, Xavier Gabaix, Itay Goldstein (discussant), Dirk Krueger,

Alessandro Lizzeri, Ulrike Malmendier, Hamid Mehran, Anthony Saunders, Philipp Schnabl, Sophie Shive

(discussant), Per Strömberg, Vikrant Vig, Alexander Wagner, Jeffrey Wurgler and Alminas Zaldokas, as well as seminar participants at NYU Stern, NYU (Department of Economics), Federal Reserve Bank of Boston,

HEC Paris, London Business School, Cambridge Judge Business School, University of Cambridge (Faculty of Economics), Stockholm School of Economics, Brown University, Federal Reserve Bank of New York,

University of Illinois at Urbana-Champaign, Federal Reserve Board of Governors, Federal Reserve Bank of Philadelphia, EIEF Rome, Collegio Carlo Alberto, University of Amsterdam, University of Wisconsin-

Madison, Brandeis University, the 14 th FDIC/JFSR Annual Bank Research Conference, the 10 th Annual

Cambridge-Princeton Conference, and the 6 th

Bocconi-CAREFIN International Banking Conference for their comments and suggestions.

University of Pennsylvania, Department of Economics, 160 McNeil Building, 3718 Locust Walk, Philadelphia, PA 19104. E-mail: neuhann@sas.upenn.edu

University of Cambridge, Judge Business School, Trumpington Street, Cambridge CB2 1AG, United

Kingdom. E-mail: f.saidi@jbs.cam.ac.uk

1 Introduction

In this paper, we exploit the stepwise repeal of the Glass-Steagall Act in the U.S. to empirically evaluate the effects of bank-scope deregulation on the performance of bank-dependent firms. In doing so, we take a step towards measuring the value added of large universal banks as suppliers of financing to the real economy.

The Glass-Steagall Act of 1933 imposed a strict separation between commercial banking, such as borrowing and lending, and investment banking, such as securities underwriting. The repeal of the Glass-Steagall Act proceeded in a sequence of distinct deregulatory events that successively relaxed constraints on bank size and bank scope. A first step, in 1989, allowed commercial banks to increase in size and become universal banks that were allowed to engage in lending and underwriting, with strict firewalls in place separating the two activities.

While banks were, thus, free to diversify their business interests, they remained limited in their ability to bring increased bank scope to bear on bank-firm relationships. These restrictions were lifted in 1996, when regulators removed the firewalls, which previously limited information and financial flows, between securities-underwriting and commercialbank divisions within universal banks. We argue that this move to full-fledged universal banking transformed the nature of bank-firm relationships, and had important consequences for real outcomes at the firm level for externally financed firms.

Why might this be the case? Many firms require a large number of differentiated financial services over their life cycle, ranging from pure loan contracts and initial public offerings to more complex transactions involving third-party investors. If banks and firms benefit from repeated interaction, for example through reduced information asymmetries or broader contracting opportunities, economies of scope across financial products are a salient feature of financial intermediation. The advent of universal banking – and the associated opportunities to concurrently offer multiple financial services – allows financial intermediaries and their clients to fully realize these economies of scope. As a consequence, extant financial constraints in the provision of external finance for firms and investment projects subject to strong moralhazard or asymmetric-information concerns may no longer bind under universal banking.

Since risky projects tend to be particularly sensitive to these concerns (see, e.g., Stiglitz and

Weiss (1981) and Greenwood, Sanchez, and Wang (2010)), we map this channel to the data by asking whether the deregulation of universal banks led to an increase in the supply of credit for firms making risky investments.

To delineate the effect of bank scope on firm-level outcomes, we focus on the 1996 deregulatory event, and use data on lending and underwriting relationships between banks and publicly listed firms. This allows us to directly identify repeated interactions between banks and firms after the expansion of bank scope and the associated opportunities for the crossselling of financial products in 1996. By cross-selling, or cross-marketing, we understand the offer and concurrent provision of loans and non-loan products, most notably corporatesecurities underwriting, to firms by universal banks. We treat these incidences as a measure

1

Figure 1: Loan-weighted Average Six-year [t,t+5] Sales-growth Volatility associated with Loans granted to Public Firms by Commercial and Universal Banks

(1987-2005).

Post-1996 loans by universal banks are split into cross-sold and non-cross-sold loans, where cross-sold loans are defined as loans whose debtor firms also received an underwriting product from the same universal bank anytime within the last five years. Source: own analysis based on CRSP/Compustat, DealScan loan data, and SDC underwriting data.

of the extent to which banks and firms were able to realize economies of scope across financial products through closer relationships. We compare a firm’s riskiness before and after receiving a universal-bank loan after 1996 in order to estimate a potential treatment effect of universal-bank cross-selling.

In Figure 1, we plot the loan-weighted average six-year sales-growth volatility of pub-

lic firms in the U.S. that received loans from commercial and universal banks.

1

Among universal-bank loans, we differentiate between cross-sold and non-cross-sold ones starting in

1996, where we label loans as cross-sold when the respective debtor firms also received an underwriting product from the same universal bank. Until 1996, commercial- and universalbank loans are associated with similar levels of firm risk, but after 1996 the firm-level risk associated with cross-sold universal-bank loans exceeds that of all other kinds of loans.

To solidify this finding, in the empirical analysis, our identification strategy is to compare firms with cross-sold universal-bank loans in or after 1996 to firms with universal-bank loans that were not cross-sold. That is, the treatment group consists of firms that received at least

1

In Figure A.1 of the Appendix, we also provide this graph for a second measure of volatility, namely

stock-return volatility.

2

one loan and underwriting service concurrently from the same universal bank in or after

1996. The control group consists of firms that in or after 1996 received at least one loan from a universal bank but trusted another bank with an underwriting mandate. To pick a control group of firms with universal-bank loans in or after 1996 that were not cross-sold for plausibly exogenous reasons, we limit the control group to firms that were locked into an underwriting relationship with an investment bank just before the 1996 deregulation.

In this manner, we provide empirical evidence that the increased scope of universal banking boosted lending to riskier firms, as measured by higher sales-growth, stock-return, and option-implied volatilities of the treatment vis-à-vis the control group. The estimated treatment effects are sizable, and their order of magnitude corresponds to within-firm increases of at least 5% across all outcome variables (and up to 14% for sales-growth volatility). We also show that these increases in firm-level risk were not associated with higher default risk.

We then turn to the question as to whether the realization of economies of scope through cross-selling has not just enabled universal banks to finance riskier projects, but whether the risk-increasing developments were accompanied by higher productivity and investment by universal-bank-financed firms. In doing so, we aim to provide tentative evidence on whether universal banking may be efficiency-enhancing at the firm level. Using the same identification strategy as before, we find that treatment effects for cross-sold vs. a control group of non-cross-sold universal-bank loans led to long-lasting within-firm increases of 3 to

4% in total factor productivity (TFP), 6 to 7% in capital expenditure, and 5 to 9% in market capitalization. Our findings attest to a potentially efficiency-increasing effect of deregulating bank scope: when universal banks receive the ability to realize economies of scope across loans and underwriting services through cross-selling, then this leads to an increase in the supply of credit for firms making risky, productivity-increasing investments.

Last, we complement our analysis based on loans issued by mature, public firms with evidence on firms early in their life cycle. Namely, we examine whether universal banks extended their risk-taking behavior to their role as underwriters by serving as bookrunners for IPOs of younger and, thus, potentially riskier firms. In this setting, the focus of analysis shifts to the bank level by comparing the age of firms in IPOs run by universal banks compared to investment banks, whose scope of banking activities was unaffected by the deregulation, before and after 1996. We find that, as a response to the deregulation, universal banks took firms public that were at least 5 years younger than those serviced by investment banks. Our evidence on IPO age supports the idea that the increase in bank scope relaxed universal banks’ constraints that previously kept them from contracting with riskier firms.

In summary, our results establish that the financial deregulation of universal banks has led to the financing of riskier projects with higher-return prospects. From this we infer that increasing bank scope from pure commercial banking (i.e., lending) to combined lending and corporate-securities underwriting has not just changed the landscape of U.S. banks, but – through a transformation of the nature of bank-firm relationships – also left its mark on publicly listed firms that interacted with, and borrowed from, universal banks.

3

Related Literature

Our paper is related to the literature that attempts to link credit-supply shocks to real effects, such as investment and employment, of financially constrained firms (see, for example,

Campello, Graham, and Harvey (2010) and Chodorow-Reich (2014)), most notably in the aftermath of financial deregulation (e.g., Bertrand, Schoar, and Thesmar (2007) using the

French Banking Act of 1985). Since the 1970s, multiple episodes of financial deregulation have significantly changed the architecture of the U.S. banking sector. Two deregulations have received notable attention in this literature, namely the staggered passage of intra- and interstate-banking (i.e., branching) deregulations in the U.S. between 1970 and 1994.

A number of papers from this literature have considered the impact of branch deregulation on some of the real effects discussed in this paper, most notably risk. However, it appears that this episode of deregulation is not associated with an increase in risk: Morgan,

Rime, and Strahan (2004) find a stabilizing effect on state-level growth, which is confirmed

by Correa and Suarez (2009) at the firm level.

2

Other than risk as a relevant outcome variable in the real economy, Amore, Schneider, and Zaldokas (2013), Chava, Oettl, Subramanian, and Subramanian (2013), and Cornaggia, Mao, Tian, and Wolfe (2013) analyze the impact of branch deregulation and, thus, banking competition on different dimensions of innovation. Correa and Suarez (2009) is one of the few papers scrutinizing the impact of financial deregulation on the volatility of large, publicly listed firms in the U.S. Unlike us, they find a stabilizing effect on firm volatility.

Furthermore, our identification strategy differs from those in the literature on branch deregulation in important ways. To establish the causal effects of branch deregulation on firm-level outcomes, such as innovation, identification strategies in the literature generally exploit the staggered timing of branch deregulation across states, and then distinguish between bank-dependent and non-bank-dependent firms in treated states (see Amore, Schneider, and

Zaldokas (2013) and, using Italian data, Benfratello, Schiantarelli, and Sembenelli (2008)).

In contrast, we use data on firms’ lending relationships with universal banks in an at-

tempt to directly identify the impact of financial deregulation on firm-level outcomes.

3

Our treatment is defined at the bank-firm level, and affects the scope of activities engaged in by universal banks, rather than an expansion of the geographical scope, as is the case under branch deregulation. On the other hand, intrastate- and interstate-banking deregulations impact firms’ financing decisions through an increase in credit supply, while leaving the nature of financial intermediation unaltered (see, among others, Jayaratne and Strahan (1996)).

2

3

The finding in Kerr and Nanda (2009) as well as Kerr and Nanda (2010) that there has been large-scale entry at the extensive margin following branch deregulation appears to be instrumental in explaining the negative effects on the median firm volatility of publicly listed firms (unlike small entering firms) and on state-level volatility (namely, through diversification).

While this idea is similar in spirit to that pursued by Herrera and Minetti (2007), the authors, using data from Italy, do not make use of any regulatory quasi-experiment to identify the impact of informed lending on firm outcomes.

4

This renders it difficult to use loan data to assign any firm-level effects to increased credit supply due to branching. Furthermore, by presenting evidence on higher risk taking by universal banks not just in the loan market but also in the IPO market, we vary control groups to include investment banks. In this way, we minimize the possibility that our results may be confounded by any long-run consequences of branch deregulation in the late 1990s, as investment banks were not affected by the latter.

Besides the literature on the real effects of financial deregulation, our paper naturally connects with a large literature on universal and relationship banking, which is surveyed by Drucker and Puri (2007). One strand of this literature looks at the bank-level, rather than the firm-level, effects, such as risk indicators (Saunders, Strock, and Travlos (1990) and

Cornett, Ors, and Tehranian (2002)), of the repeal of the Glass-Steagall Act. More than that,

Ang and Richardson (1994), Kroszner and Rajan (1994), and Puri (1996) provide evidence that, in the pre-Glass-Steagall era, investors were willing to pay higher prices for securities underwritten by universal rather than investment banks. Also, the price differential between universal-bank and investment-bank underwritings, both pre-Glass-Steagall and post-1989, is found to be greater for securities with high information costs, such as non-investment-grade securities (see, for instance, Gande, Puri, Saunders, and Walter (1997)). Our approach differs from these papers in that we focus on the firm-level real effects of universal banking.

Very few empirical papers use the stepwise nature of the repeal of the Glass-Steagall

Act. Among those that do, Bhargava and Fraser (1998) use an event-study approach to analyze wealth and risk effects at the bank-holding-company level around different stages

of the repeal.

4

Interestingly, while the authors do find increases in risk, alongside positive abnormal stock returns, for the early stages of the repeal, they find no such effects for the

August 1, 1996 event that we use in this paper. Furthermore, the bank-level risk measures employed may also capture diversification effects across commercial- and investment-bank divisions.

Another strand of the literature on universal banking focuses on concurrent lending and

underwriting.

5

With the notable exception of Ljungqvist, Marston, and Wilhelm (2006), who scrutinize the role of prior relationships on banks’ likelihood of winning underwriting mandates, most studies document pricing effects for firms contracting with universal banks.

For example, Drucker and Puri (2005) present evidence that issuers derive benefits from concurrent lending and underwriting. Their view is that both universal banks and investment banks compete for such deals, but through different channels: while universal banks are more likely to offer discounted yield spreads on concurrent loans, which is also confirmed by Calomiris and Pornrojnangkool (2009), investment banks are more likely to discount the underwriter spread for seasoned equity offerings. Regarding the former, the authors

4

5

A similar exercise, albeit not an event study, with different time frames is conducted by Geyfman and

Yeager (2009).

Note that our paper does not focus on universal banks’ holding equity stakes in companies and their representation on the latter’s boards (see Ferreira and Matos (2012)), as is the case under the classical model of universal banking in Germany.

5

report that, unlike investment-grade borrowers, non-investment-grade borrowers receive significantly lower yield spreads on concurrent loans relative to matched non-concurrent loans.

Furthermore, Schenone (2004) finds significantly less IPO underpricing for firms that have pre-IPO lending relationships with prospective underwriters (i.e., universal banks).

These papers have in common the notion that realized economies of scope, as reflected by pre-existing lending relationships, affect universal banks’ underwriting performance (see also Kanatas and Qi (1998) and Kanatas and Qi (2003)). We deviate from this literature in two ways. First, we characterize economies of scope by means of firm-level real effects rather than effects on the pricing of financial products. Our findings attest to the idea that the realization of economies of scope across financial products through cross-selling enabled universal banks to finance riskier projects, and that this kind of risk taking is rewarded by higher productivity and returns. Second, using the 1996 deregulatory shock, we are among the first to combine the two-staged structure of the repeal of the Glass-Steagall Act with the incidence of cross-selling. This is central to the causal interpretation of the effects of increased bank scope and cross-selling on firm outcomes. To the best of our knowledge, we provide the most comprehensive effort to relate variation in bank scope to firm-level risk and productivity.

Lastly, insofar as we can address policymakers’ challenges in regulating the scope of banking, our paper is also generally related to recent work on the regulation of the financial sector (see Opp, Opp, and Harris (2013), Harris, Opp, and Opp (2014), and Hoffmann,

Inderst, and Opp (2014)).

2 Empirical Methodology and Identification

2.1

Institutional Background

In this paper, we consider the gradual dismantling of the Glass-Steagall Act, which separated commercial and investment banking, and the rise of universal banking, culminating in the

Gramm-Leach-Bliley Act in November 1999. Under Section 20 of the Glass-Steagall Act, commercial banks were prohibited from engaging in any kind of underwriting or securities business, which was subsequently entirely in the hands of investment banks and other investment houses. The Glass-Steagall Act characterized the financial-architectural landscape in the U.S. until 1987. Starting April 30, 1987, commercial banks were allowed to open socalled Section 20 subsidiaries and generate up to 5% of gross revenues from underwriting and dealing in certain securities, namely municipal revenue bonds, mortgage-related securities, consumer-receivable-related securities, and commercial paper. Two years later – on January

18, 1989 – banks were allowed to engage in veritable investment-banking activities, most notably corporate debt and equity underwriting, and on September 13, 1989, the revenue limit was raised to 10%. This gave rise to another possibility for commercial banks to become

6

universal banks, other than by opening Section 20 subsidiaries, namely by purchasing or merging with investment banks. These measures summarize what we understand as the first stage of the repeal of the Glass-Steagall Act, followed by seven years of no further activity.

A major expansion of universal-banking deregulation took place on August 1, 1996, when the Federal Reserve Board eliminated firewalls within bank-holding companies, while relaxing the revenue limit on underwriting securities from 10 to 25%. These measures enhanced the ability of universal banks to engage in cross-selling, which was previously prohibited, or at least severely restricted, under the Federal Reserve Act (Sections 23A and B). We argue that a major driver of universal banks’ capacity to finance riskier firms comes from their cross-selling and thereby realizing economies of scope across financial products. Loans are granted upon approval by a credit committee, often on the basis of high expected depth of cross-selling. This phenomenon has also been discussed in the academic literature: Bharath,

Dahiya, Saunders, and Srinivasan (2007) provide ample evidence of cross-selling of loans and

non-loan products (fee-generating services) such as debt and equity underwriting.

6

We use transaction-level data to determine whether a bank was a universal bank at the time of a given loan transaction. We delineate this by comparing the completion date of a bank-scope-expanding (from commercial to investment banking) acquisition or the opening date of the respective bank’s first Section 20 subsidiary to the transaction date.

As an example, consider the historical anatomy of J.P. Morgan. Before acquiring Bank

One on July 1, 2004, J.P. Morgan already became a universal bank by opening a Section 20 subsidiary on April 1, 1987, followed by a merger with Chase Manhattan, which had a Section

20 subsidiary since December 30, 1988 (and later merged with Chemical Bank). Similarly,

Bank One, J.P. Morgan’s acquisition target in 2004, maintained a Section 20 subsidiary which it had opened on February 2, 1989. Thus, despite a series of mergers, J.P. Morgan became a universal bank through opening a Section 20 subsidiary in 1987, and any loan granted by J.P. Morgan before April 1, 1987 is labeled as a loan provided by a commercial bank that eventually became a universal bank, but was not a universal bank at the time of the transaction. Similarly, any loan after this date is labeled as a loan granted by a universal

bank.

7

In Table 1, we provide an overview of all universal banks in our loan data.

We next face the challenge of identifying cross-sold universal-bank loans. The 1996 deregulation enabled universal banks to coordinate their offerings of loans and non-loan products.

Whereas it is still possible to observe concurrent lending and underwriting before 1996, when

6

7

Furthermore, Drucker and Puri (2005) and Yasuda (2005) examine the relationship between past lending relationships and seasoned equity offerings and debt underwriting, respectively.

Note that we also have U.S. banks of international origin in our sample. These banks are special cases in that, before the International Banking Act of 1978, they were not subject to the Glass-Steagall Act. As a consequence, international banks that were active in the U.S. before 1978 and established as universal banks outside the U.S. were allowed to continue their business model in the U.S. (as long as they would not expand their activities further). None of the banks in our sample were subject to the International

Banking Act. For instance, Deutsche Bank became a universal bank only after acquiring Morgan Grenfall, a London-based investment bank, in 1990. Similarly, Crédit Suisse acquired a controlling stake in the

American investment bank First Boston Corporation in December 1988.

7

Figure 2: Proportion of Cross-sold Loans granted to Public Firms by Universal

Banks (1987-2010).

Source: own analysis based on DealScan loan data as well as SDC underwriting and M&A data.

the firewalls were still in place, because firms could interact separately with a universal bank’s commercial-bank and securities divisions, the coordinated marketing of loans and non-loan products across the two divisions took off only in 1996. Besides increased informational exchange between commercial-bank and securities divisions, this should naturally contribute to more frequent occurrences of concurrent lending and underwriting due to cross-selling, and generally tighter bank-firm relationships. The role of information – the free flow of which was brought about by the elimination of firewalls in 1996 – is paramount: the exchange of information supports the cross-selling efforts of commercial-bank and securities divisions within universal banks, and the very process of cross-selling generates further information about the client through closer intermediation relationships.

While we do not observe these cross-marketing efforts directly, we can observe the increased incidence of concurrent lending and corporate-securities underwriting by universal

banks in or after 1996. In Figure 2, we plot separately two time series for the proportions

of universal-bank loans that are associated with (i) cross-sold underwriting services and (ii)

cross-sold M&A-advisory services.

8

The figure shows that while firms did occasionally receive loans and underwriting services together from universal banks before 1996, these incidences

8 We define a pair of loans and non-loan products to be cross-sold if they are issued to a firm by the same universal bank within five years (from year t − 4 to t ). To avoid double-counting, for each cross-selling incidence, we use only the first year t in which it holds that a firm received a loan and a non-loan product from the same universal bank anytime from t − 4 to t .

8

have become more prominent starting in 1996, with the respective proportion increasing sharply from 38% in 1995 to 54% in 1996. From this we conclude that the 1996 deregulation indeed boosted the cross-selling of loans and corporate-securities underwriting.

We use the cross-selling of loans and M&A-advisory services as a benchmark to demonstrate that the 1996 deregulation affected solely the cross-marketing of underwriting, rather than any other, services.

In particular, unlike corporate-securities underwriting, M&Aadvisory services were not forbidden under Glass-Steagall. What we can see from the figure is that the cross-selling of loans and M&A-advisory services hardly responded to the 1996 deregulation. Instead, the corresponding time series displays a surge only after 1999, potentially stemming from the Gramm-Leach-Bliley Act and the subsequent merger waves among universal banks and their consolidation.

2.2

Hypothesis Development

The goal of this paper is to empirically evaluate the impact of increases in bank scope on real outcomes of borrower firms. As in previous theoretical work, such as Kanatas and Qi

(1998) and Kanatas and Qi (2003), we start from the premise that cross-selling represents a positive shock to the quality of banks’ information about borrower firms. Increased scope allows banks and firms to interact more frequently and across a wider array of financial products through cross-selling. Our basic hypothesis is that this deepening of relationships improved universal banks’ ability to efficiently provide external finance to firms through reduced intermediation frictions. To derive testable implications from this channel, we use economic theory to map increased lender informedness to observable firm outcomes. We do not, however, take a stand on the precise nature of lender information. Indeed, we view reductions in ex-ante asymmetric information, improvements in ex-post monitoring efficiency, or a closer understanding of firm-level moral hazard as a-priori equally plausible determinants of increased efficiency in financial contracting. To account for this richness, we look for theoretical predictions that hold true across a wide variety of frameworks used to study financial intermediation. We find that that a robust conclusion is that lender informedness is particularly valuable for risky firms .

In a costly-state-verification framework, one of the canonical environments in which to study financial intermediation, Greenwood, Sanchez, and Wang (2010) show that risky firms are particularly difficult to monitor. This is the case because large spreads in potential outcomes make concealing good outcomes very attractive to borrowers. As such, only lenders for whom it is easy to detect borrower malfeasance, such as those that are well-informed about their borrowers, are able to deter borrowers from misrepresenting returns. Greenwood,

Sanchez, and Wang (2010) thus show that lender informedness is a crucial determinant of the access to external finance for risky enterprises.

In a second canonical environment, ex-ante asymmetric information, Stiglitz and Weiss

9

(1981) show that risky borrowers may be credit rationed due to two channels: the adverseselection channel and the incentive channel. In the adverse-selection channel, banks may refuse funding to risky firms in order to screen firms that are unlikely to repay loans, even if the risky firm is highly productive. Due to the incentive channel, this problem cannot be solved by raising interest rates on loans. In particular, firms may react to higher interest rates by taking on projects with lower chances of success, but higher payoffs if they succeed.

Therefore, again, increased lender informedness proves particularly valuable for risky firms.

In Neuhann and Saidi (2014), we complement this analysis by examining a third canonical environment, namely borrower moral hazard. To delineate the role of lender informedness in this context, we assume that lenders are ex-ante uninformed about the precise nature of the borrower’s moral-hazard problem, but can acquire information about this problem at a cost. We formally represent this uncertainty by assuming that the lender is uninformed about which actions the borrower can take in a given state of the world. If the lender acquires information, he learns which actions the borrower can take, which enables him to tailor intermediation contracts to the moral-hazard problem at hand. In the absence of this information, the lender must design a contract that reckons with a multitude of potential actions. Lender informedness therefore allows for more efficient financial contracting, and reduces frictions in financial intermediation. We find that the benefits of lender informedness are especially large for risky investment projects, because imprecisely tailored financial contracts are particularly costly when potential actions and outcomes vary substantially.

Hence, in all three frameworks, lender informedness is particularly valuable for risky projects. We therefore hypothesize universal-bank-financed firms to exhibit higher risk after an increase in the scope of the respective universal banks’ activities. We will show that this holds across various risk measures, not just limited to corporate lending, but also in terms of risk associated with younger firms taken public by universal banks.

2.3

Identifying Economies of Scope from Concurrent Lending and

Underwriting

Our identification strategy is based on a deregulatory shock in 1996 that affected the scope of banking activities engaged in by universal banks. Most notably, the 1996 deregulatory shock boosted cross-selling by universal banks by enabling them to freely exchange information between their commercial-bank and securities divisions, and thereby coordinate their cross-marketing efforts. We focus on the 1996 deregulation because it represents a direct shock to bank-firm relationships, and allowed for unconstrained, or at least less constrained, contracting across multiple financial products.

In our analysis, we use the distinction between loans that were cross-sold and those that were not to identify the role of economies of scope across financial products for the capacity of universal banks to finance riskier firms. That is, we compare two groups of loans granted

10

by universal banks in or after 1996: loans to firms that received a loan and underwriting services from the same universal bank (i.e., cross-sold loans) vs. loans to firms that received a loan from a universal bank, but issued a corporate security through a separate bank (i.e., non-cross-sold loans). For the latter group to be a legitimate control group, we require the non-incidence of cross-selling to be for exogenous reasons. We construct this control group by looking at loans that could not be cross-sold to firms after 1996 because these firms had already received underwriting services from an investment bank just before the

1996 deregulation and were, thus, already locked into an underwriting relationship with a non-universal bank when the 1996 deregulation took place.

In constructing this control group, our identification argument is as follows. Prior to the

1996 deregulation, universal banks and investment banks were functionally equivalent in their ability to offer underwriting services to firms and, therefore, competed on equal footing in the underwriting market. In particular, given that no bank was able to offer cross-sold financial products prior to 1996, firms did not sort into universal banking or investment banking according to the potential value of cross-selling to the firm. Once the 1996 deregulation occurred, however, firms that had previously built an underwriting relationship with an investment bank were at a disadvantage in realizing economies of scope across multiple financial products relative to firms that had initially built a relationship with a universal bank.

The validity of our identification argument rests on two key assumptions. First, in order for firms not to sort into universal vs. investment banking based on the value of cross-selling to the firm, the timing of the 1996 deregulation must have been unexpected. This assumption is affirmed by the fact that the banking industry had already proposed the elimination of firewalls in 1991, but had been rejected by the United States House Committee on Financial

Services. Hence, it is unlikely that banks and firms were anticipating the deregulatory policy before 1996. Second, in order for pre-existing underwriting relationships to have an impact on future cross-selling opportunities, there must be substantial switching costs when moving from one (type of) underwriter to another. This assumption is verified in the literature on lock-in in underwriting relationships (see, for example, James (1992) and Ljungqvist,

Marston, and Wilhelm (2006)).

We verify the presence of lock-in effects in our sample by showing that the incidence of pre-1996 underwriting relationships indeed had a lasting (negative) impact on the incidence

of cross-selling after 1996, giving us the required variation. In Figure 3, we plot the proportion

of cross-sold universal-bank loans anytime in or after 1996 , conditional on the borrower firm also receiving an underwriting product in or after 1996, for two groups of firms: those that in a given pre-deregulation year received an underwriting product from an investment bank and those that received an underwriting product from a universal bank, which can be, but need not be, the same as the one granting the loan in or after 1996. We vary the years prior to the deregulation during which the latter firms could be locked into an underwriting relationship with an investment bank. The post-1996 cross-selling probabilities are similar

11

Figure 3: Determination of the Control Group.

The graph plots the proportion of cross-sold universal-bank loans in or after 1996, conditional on the borrower also receiving an underwriting product from any bank in or after 1996, and compares firms that received an underwriting product from an investment bank to firms that received an underwriting product from a universal bank in year t before 1996 (varying t from 1991 to 1995). Source: own analysis based on DealScan loan data and SDC underwriting data.

for the two groups that received an underwriting product either from an investment bank

(our potential control group) or from a universal bank in the pre-deregulation period from

1991 to 1993. This changes abruptly in 1994 and 1995, and the cross-selling probabilities for our (actual) control group diverge sharply from those for firms that received an underwriting product from a universal bank, implying a lasting difference in the post-1996 cross-selling probability of 10.5 to 16.2 percentage points.

We lend further support to our identification strategy by showing that firms in the treatment and control group are comparable along numerous dimensions. Both groups received universal-bank loans after 1996, and both also received underwriting products around the

time of their universal-bank loans. In addition, in Table 2, we provide evidence that firms

in the treatment and control group are similar along observable characteristics.

Because we define our treatment and control groups at the loan level, and some firms received multiple universal-bank loans in or after 1996, we can compare firms in these two groups only if we first move from the loan level to the firm level. As such, we must account

12

for the fact that some firms are part of both the treatment and the control group.

9

To this end, we classify as control firms those firms with at least one year in which they received a non-cross-sold universal-bank loan, in line with our definition of the control group,

but no cross-sold universal-bank loan.

10

In reverse, this implies that all treatment firms with cross-sold universal-bank loans either never received a non-cross-sold universal-bank loan, in line with our definition of the control group, or, if they did, only when they also received a cross-sold universal-bank loan in the same year.

In the first panel of Table 2, we focus on differences in our outcome variables for firm-

level risk, productivity, and investment in 1993. Treatment firms exhibit lower stock-return volatility, but the difference, albeit statistically significant, is rather small. We also note a higher market capitalization among treatment firms, although the difference is not sta-

tistically significant.

11

Otherwise, treatment and control firms are very similar in terms of sales-growth volatility, TFP, and capital expenditure. In the next panel, we see that the difference in market capitalization holds qualitatively for alternative measures of firm size, namely sales and the number of employees, but the respective differences are again insignificant. In the last panel, we also fail to find any differences in the total number of loan and underwriting transactions until 1993.

Overall, our treatment and control firms appear to be very similar before 1994. To show

that our insights from Table 2 are not specific to the year 1993, we plot the ratio between

the mean of the respective variable for treatment firms and the mean for control firms – i.e., a ratio of one indicates that the two means are identical – for all variables from 1989

to 1993 in Figure A.2. Except for market capitalization (and, to some degree, sales and the

number of employees), all ratios are close to unity. The dashed lines indicate the respective ratios for market capitalization and stock-return volatility, the only significant difference in

Table 2. For stock-return volatility we note that the difference in means, which is always

statistically significant, remains quantitatively small throughout the entire period. While the difference in market capitalization is highest in the years 1989 to 1991, it is never statistically significant, neither then nor thereafter. Instead, the difference in TFP becomes significant at the 6% level in 1989 and 1990, but remains quantitatively small.

9

10

11

To see why this might be the case, consider the following example. A firm received an underwriting product from an investment bank in 1995. The firm then received another underwriting product from the same investment bank in, say, 1998. If the firm also received a non-cross-sold loan from a universal bank in

1998, then the 1998 loan transaction becomes part of the control group. Assume next that the investment bank was later acquired by said universal bank, from which the firm received a cross-sold universal-bank loan in 2000. Thus, the firm’s 2000 loan transaction is part of our treatment group.

For these firms, 65% of all loans in either the treatment or the control group belong to the control group.

The difference in market capitalization would be a concern for the validity of our identification strategy if it represented a difference in the amount of equity raised through cross-sold vis-à-vis non-cross-sold equity underwriting. To control for this possibility, in regressions unreported in this paper, we include the borrowing firm’s market capitalization in all specifications without market capitalization as dependent variable, and find that our treatment effects are robust throughout. Finally, note that our cross-selling variable includes both equity and debt underwriting, further buttressing the robustness of our results to the concerns above.

13

These findings attest to the idea that a key factor in accounting for post-1996 differences in cross-selling between treatment and control are to be found in the lock-in to pre-1996 underwriting relationships. Our bank-firm-level identification has key advantages over a bank-level identification, where one would estimate the average risk associated with loans granted by universal banks compared to pure commercial banks before and after 1996. In particular, we do not rely on the establishment dates of universal banks – i.e., the conversion of commercial into universal banks – as the main variation in bank scope of the lender, as

commercial banks endogenously chose to become universal banks.

12

Conversely, our identification strategy based on the 1996 deregulation focuses exclusively on universal banks, so that we do not use commercial banks that did not become universal banks prior to 1996 and, thus, did not experience a shift in the scope of their activities in 1996 as a control group. Our identification strategy also tackles the concern that post-1996 risk taking by universal banks may be due to the sorting of new firms with different risk profiles seeking financing from universal banks after the deregulation was implemented. This would render it problematic to compare universal-bank loans before and after 1996. Instead, we focus on loans granted by universal banks only after the deregulation.

Empirical Implementation

To test our claim that universal banks financed riskier firms, we use data on loans issued by publicly listed firms in the DealScan database. For our analysis, we collapse our loans sample to the firm-loan-year level, i.e., we summarize all loans of a firm in a given year.

A loan for which a universal bank was a lead arranger is labeled as cross-sold if the same universal bank also served as a bookrunner in at least one underwriting mandate anytime from two years before to two years after the respective loan issue (implying a five-year

circle).

13

Per firm-loan year (summarizing all loans in a given year, including those given out by banks that never become universal banks, as will be described in greater detail in

Section 2.4) we record two observations, namely one pre-loan(s)-year and one post-loan(s)-

year observation. This enables us to compare a firm’s riskiness before and after receiving a

(cross-sold or non-cross-sold) universal-bank loan.

For each loan granted to firm i at time t , we determine whether bank j is a universal bank.

If so, we set U B ijt equal to one. Starting in 1996, conditional on U B ijt

× Af ter (1996) t

= 1, there are three possibilities: the firm did not receive any underwriting product around the time of the loan issue, it received an underwriting product from the same universal bank as the one granting the loan (which is when we set Cross − selling ijt

= 1), or it received an

12

13

As noted by, among others, Bhargava and Fraser (1998), the initiation of universal-banking deregulation from 1987 to 1989 was based on the Federal Reserve’s responses to specific requests from large banks

(Bankers Trust, Citicorp, and J.P. Morgan).

Note that for all estimations using universal-bank loans in or after 1996, our cross-selling definition is always censored at the year 1996. Our results are robust to variations of this circle (in addition to those presented in the paper), and are available upon request.

14

underwriting product from another bank (as indicated by N o CS ijt

= 1). Firms with crosssold universal-bank loans in or after 1996 constitute our treatment group, so we have for those firms U B ijt

× Af ter (1996) t

× Cross − selling ijt

= 1. For our control group, we pick firms that entered into an underwriting relationship with an investment bank in 1994 or 1995; we indicate these firms using the dummy variable U nderwriting (1994 / 95) ijt

, so for our control group we have U B ijt

× Af ter (1996) t

× N o CS ijt

× U nderwriting (1994 / 95) ijt

= 1. Finally, note that we define U nderwriting (1994 / 95) ijt only for firms without cross-sold universalbank loans – i.e., Cross − selling ijt

= 0 – for which N o CS ijt is not necessarily one.

As we have one pre- and one post-observation per firm-loan year, we always have multiple

observations per firm.

14

This enables us to include firm fixed effects and estimate withinfirm effects of universal-bank loans on firm-level risk by estimating the following regression specification: outcome ijt

= β

1

U B ijt

+ β

2

U B ijt

× Af ter (1996) t

+ β

3

U B ijt

× Af ter (1996) t

× Cross − selling ijt

+ β

4

U B ijt

× Af ter (1996) t

× N o CS ijt

× U nderwriting (1994 / 95) ijt

+ β

5

U B ijt

× Af ter (1996) t

× N o CS ijt

+ β

6

U B ijt

× Af ter (1996) t

× U nderwriting (1994 / 95) ijt

+ β

7

X ijt

+ ψ i

+ µ t

+ λ j

+ ijt

, (1) where outcome ijt is the natural log of firm i ’s outcome variable (e.g., sales-growth volatility) associated with the pre-loan(s) year and post-loan(s) year in which firm i received one or multiple loans from one or multiple banks j , U B ijt is an indicator variable for whether, given any loan transactions in a year, at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section

20 subsidiary, Af ter (1996) t is an indicator for whether the firm’s loan year in question was in 1996 or later, and Cross − selling ijt is an indicator for whether any loan in year t was associated with a cross-sold underwriting product by the same bank from t − 2 to t + 2. Conversely, N o CS ijt indicates whether a firm that received a loan in year t also received an underwriting product from t − 2 to t + 2 which was not issued by the same bank.

U nderwriting (1994 / 95) ijt is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in

1994 or 1995.

X ijt denotes time-varying characteristics of the borrowing entity i and of the originated loans j , ψ i and µ t are firm and year fixed effects, respectively, and λ j denotes bank fixed effects, which we include for all lead arrangers of all loans j in year t that are or eventually become universal banks (whereas all remaining commercial banks are grouped together). For the dependent variable and all firm-year and loan(s)-year controls, the first, pre-loan(s)-year observation uses information from the last trading day of year t − 3, and

14 Even if a firm received only one loan throughout our sample period, through our method of generating one pre- and one post-observation per firm-loan year, we would yield two observations for that firm.

15

the second, post-loan(s)-year observation uses information from the last trading day of year t + 2. In case of multiple loans, loan-year(s)-level control variables reflect the respective year average over all loans in year t . All loan(s)-year-level (indexed by j ) variables and fixed effects are equal to zero in the pre-loan(s) year. Standard errors are clustered at the lead-arranger level for both observations of each firm’s loan year, treating each (eventual) universal bank individually and pooling all pure commercial banks.

Note that as the omitted category consists of firm-loan years with only commercialbank (and no universal-bank) loans, and commercial banks cannot cross-sell by definition

(because they offer only loans), we have that whenever Cross − selling ijt

= 1, it must be that

U B ijt

× Af ter (1996) t

= 1, as universal banks could actively cross-sell starting in 1996. This is why the stand-alone variable Cross − selling ijt

drops out from (1). Similarly,

N o CS ijt can only be meaningfully defined if bank j could cross-sell in principle, requiring again that

U B ijt

× Af ter (1996) t

= 1. Conversely, all commercial-bank loans are by definition not crosssold and, as the omitted category, captured by our bank fixed effects. It is worth noting, though, that we do not distinguish whether firms with commercial-bank loans received an underwriting product from another bank. This is because our comparison of interest concerns solely universal-bank loans in our treatment vs. control group.

The treatment effect – i.e., the difference between treatment and control group – is given by the difference between β

3 and the sum of β

4

, β

5

, and β

6

.

2.4

Data Description

The focus of our analysis will be on estimating the impact of universal banking on different firm-level outcomes, most notably risk and productivity. To this end, we use as our main

data sources Compustat accounting data, CRSP stock prices, DealScan loan data,

15

and

SDC debt- and equity-underwriting data. As is customary, we drop public-service, energy, and financial-services firms from our analysis. On the transaction level, we focus on loans granted to public firms in the U.S. in the DealScan database since 1984, as well as on U.S.

IPOs listed in the SDC database since 1976. While for IPOs we consider the bookrunners

(i.e., lead underwriters), the loans in the DealScan database typically constitute syndicated loans. Furthermore, we focus on the package level, comprising multiple facilities, for the definition of loan-level variables, except for identifying lead arrangers.

When we consider cross-selling of loans and corporate-securities-underwriting services by universal banks, we first collapse the loan level to the firm-loan-year level, i.e., we record all loans of a firm in a given year in a single row. By merging the firm-loan-year data with the SDC underwriting data, we can determine whether one of the loans in a given year was accompanied by debt or equity issued through the same universal bank that provided the loan. In general, we fix the relevant time window to five years.

15 We match DealScan with Compustat data using the link provided by Chava and Roberts (2008).

16

Outcome Variables

The most important outcome variables considered in this paper are firm-level risk measures.

We focus primarily on the six-year volatility of sales-growth rates γ it of firm i in year t

.

16

For sales-growth volatility, we follow Davis, Haltiwanger, Jarmin, and Miranda (2007) in constructing annual growth rates that accommodate entry and exit:

γ it

=

1

2 x it

− x i,t − 1

( x it

+ x i,t − 1

)

, where x it denotes sales from Compustat.

(2)

As alternative measures of firm-level risk associated with loans, we also consider (sixyear) stock-return volatilities, which are calculated using monthly CRSP data, as well as five-year implied volatilities calculated using the volatility surface from option prices. The latter data are obtained from Option Metrics, and are available starting in 1996.

t − 7 t − 2 t t + 2

(pre-loan 6-year volatility) t + 7

(post-loan 6-year volatility)

Special care needs to be taken with respect to the time horizon of the borrower firm’s sixyear sales-growth and stock-return volatilities to avoid overlapping observations for the preand post-universal-bank-loan(s)-year periods. The above figure summarizes our procedure.

Given any loan in year t , for the first, pre-loan(s)-year observation we use the six-year volatility from t − 7 to t − 2, where t − 2 indicates the very beginning of the year t − 2 or the last trading day of the year t − 3 (similarly, t − 7 denotes the last trading day of the year t − 7). Accordingly, for the second, post-loan(s)-year observation we use the six-year volatility from year t + 2 to t + 7.

Given that public firms in DealScan are typically mature, we use another outcome measure to capture firm risk earlier in the firm’s life cycle: the firm’s age at the time of its IPO.

To calculate the latter, we use the founding dates of firms with IPOs recorded in SDC until

2006, collected by Loughran and Ritter (2004).

Besides the above-mentioned risk measures, we also analyze effects on firm-level TFP, for which we use data estimated by Imrohoroglu and Tuzel (2014), who employ the semipara-

metric estimation procedure by Olley and Pakes (1996), for the panel of Compustat firms.

17

As alternative outcome variables, we will also use capital expenditure (from Compustat) to show that changes in TFP translate into actual investment, as well as market capitalization

(i.e., market value of equity) from CRSP.

16

17

We use six-year volatilities to limit the number of firms dropping out of our sample due to survival reasons.

We thank the authors for sharing their data with us.

17

Summary Statistics

In Table 3, we present summary statistics of firm-specific and transaction-level variables for

all major regression samples used in the paper. In doing so, we roughly follow the chronology

of the tables: our loans sample is the foundation for Table 7 as well as for generating the

firm-loan-years sample used in Tables 4 to 6. We add observations from Compustat, not

just in terms of outcome variables but also in terms of firms that never received loans in

DealScan, in Tables 8 to 10, Finally, we use SDC IPO data for Table 11. For all samples,

we use the actual regression sample, i.e., the sample that comprises all variables used in any of the specifications of the respective tables.

Our loans sample is based on DealScan data from 1984 to 2010. In Table 7 only, the

sample is limited to transactions with at most one universal-bank lead arranger in order to tie

the universal-bank treatment closer to individual firms.

18

The respective regression sample comprises 15,650 loans in general, and 9,090 and 8,808 observations when using six-year sales-growth volatilities σ ( \ i

) t,t +5 and six-year stock-return volatilities σ ( return i

) t,t +5 , respectively. The sample drops to 5,147 observations (because of availability starting in

1996) when using five-year implied volatilities calculated using the volatility surface from option prices, σ ( implied i

)

5 y t

. Independently from the availability of these risk measures, the

“total” regression sample, including deals with more than one universal-bank lead arranger, has 17,147 observations.

Moving to the firm-loan-years sample, we consider all loans in 1996 or later, including those with more than one universal-bank lead arranger. The regression sample is conditioned on the availability of the dependent variable before and after loan issuance. Here we limit the sample to loans for which we have firm i ’s six-year sales-growth volatility from t − 7 to t − 2 for the first, pre-loan(s)-year observation, and from t + 2 to t + 7 for the second, postloan(s)-year observation. Furthermore, in the case of multiple loans per firm in consecutive years, the pre-loan(s) year is chosen to be the last year without any loans for the respective firm prior to the sequence of years with loans, and the post-loan(s) year is chosen to be the last year in the sequence. This sample corresponds to the first, second, and fourth

columns of Table 4. That is, we have 2,528 observations divided by two, because we use two

observations per firm-loan-year, and summarize here only variables that are non-zero for the second observation.

U B is an indicator variable for whether, given any loan transactions in year t , at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Cross − selling is an indicator for whether any loan in year t was associated with a cross-sold underwriting product by the same bank from t − 2 to t + 2, which corresponds to our treatment group.

Note that because we use underwriting data until 2012, we can – depending on the time horizon of the dependent variable in question – detect cross-sold loans for our entire loans sample, which

18

All results in that table are robust to including the remaining loans.

18

runs until 2010. Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product from t − 2 to t + 2 which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in

1994 or 1995. The interaction between the two variables, N o CS × U nderwriting (1994 / 95), marks our control group of firms that received at least one loan from a universal bank in year t that was not cross-sold due to the firm entering into an underwriting relationship with an investment bank just before the 1996 deregulation.

Judging from the second panel in Table 3, our sample of post-1996 firm-loan years com-

prises 72% universal-bank loans roughly half of which (38%) we identify as cross-sold.

19

We thus have a large group of treated firms. Regarding our control, we also have a sufficiently high proportion of non-cross-sold universal-bank loans issued by firms that had entered into underwriting relationships with investment banks prior to the 1996 deregulation, namely a bit less than one-sixth of the universal-bank loans (or 11% of the total sample). In absolute numbers, these figures correspond to 474 firm-loan years in the treatment and 113 firm-loan years in the control group. The drop in numbers is due to the data availability for the dependent variable, which requires six years of data starting two years before and after the loan-issue year.

Then, in Tables 8, 9, and 10, we merge our DealScan data with Compustat data starting

in 1984, including firms that never received loans recorded in DealScan. The smaller sample size for our TFP measure is due to data availability in our TFP-data source (Imrohoroglu and Tuzel (2014)), which covers the period from 1984 to 2009.

Finally, our SDC IPO sample is limited to IPOs with no more than one bookrunner, leaving us with a regression sample of 3,827 initial public offerings. This sample is conditional on the availability of IPO age (based on Loughran and Ritter (2004)), which we use as a measure of firm-level risk early in the life cycle.

In Table 11, we explicitly distinguish by whether a universal bank was established through

M&A or through opening a Section 20 subsidiary. As we will use the 1996 deregulation as a shock to the scope of pre-existing universal banks, we report the number of universal banks

established prior to that date.

20

Besides our SDC IPO data set, we also present the general breakdown for DealScan. In DealScan, 6 out of 8 universal banks established through M&A existed before August 1, 1996. In the SDC IPO data, 4 out of 7 universal banks established through M&A existed before that date. Among Section 20 subsidiaries, 28 out of 37 and 10 out of 14 were founded before August 1, 1996 in DealScan and the IPO data, respectively.

19

20

Note that while 72% of the investigated loans (firm-loan years) in or after 1996 came from universal banks, this proportion is not entirely made up of universal-bank loans for which Cross − selling or N o CS is equal to one, indicating that 72% − 38% − 28% = 6% of the universal-bank loans were issued by firms that did not receive any underwriting services from t − 2 to t + 2.

A detailed overview is provided in Table 1.

19

3 Results

We now turn to the estimation results using the loan data, and investigate whether universal banks financed riskier firms. In doing so, we explicitly account for cross-selling and economies of scope potentially realized by universal banks across loans and underwriting services. We compare universal-bank loans issued after the 1996 deregulation, and build treatment and control groups corresponding to whether universal-bank loans were cross-sold (treatment), or whether they could not be cross-sold for plausibly exogenous reasons (control). In this manner, we show that cross-sold loans led to substantial within-firm increases in firm-level risk, and this finding is robust to using a wide range of risk measures (based on book and market data alike). We then investigate whether these risk-increasing developments were accompanied by within-firm increases in total factor productivity. Indeed, we find that firms that received cross-sold universal-bank loans exhibit increases in TFP, lasting up to six years.

This is echoed by similar effects for the firms’ capital expenditure and market capitalization.

3.1

Impact of Universal Banking on Firm Risk

In this section, we test the impact of changes in bank scope on firm-level risk. For this

purpose, we move to our firm-loan-years sample. In Table 4, we estimate specification (1)

for the sample of post-1996 loans (so that U B ijt

= U B ijt

× Af ter (1996) t

) without and with firm-year and loan(s)-year controls, and use as dependent variable the natural logarithm of the firm’s six-year sales-growth volatility σ ( \ i

) 6 y . As we include firm fixed effects, we can interpret our universal-bank coefficients as percentage changes. Note that the omitted category corresponds to pure-commercial-bank-loan years.

Here and in the following tables, in the first panel we focus on the difference between our two groups of universal-bank loans depending on their cross-selling status, and highlight the estimates for our treatment and control groups. In doing so, we provide p-values from two (two-sided) test statistics, most importantly the difference between the treatment group

(captured by U B ijt

× Af ter (1996) t

× Cross − selling ijt

) and the control group (as captured by the sum of the coefficients on U B ijt

× Af ter (1996) t

× N o CS ijt

× U nderwriting (1994 / 95) ijt

,

U B ijt

× Af ter (1996) t

× N o CS ijt

, and U B ijt

× Af ter (1996) t

× U nderwriting (1994 / 95) ijt

).

For completeness, we include all estimated coefficients in the second panel.

U B

We also test whether the sum of the coefficients on U B ijt

, U B ijt ijt

× Af ter (1996) t

× Cross − selling ijt

× Af ter (1996) t

, and is different from zero, i.e., whether firms with cross-sold universal-bank loans became riskier than those with commercial-bank loans. This is, however, not our comparison of interest, as commercial banks are systematically different from universal banks, for example, in that they have chosen not to become universal banks in the first place. In similar spirit, our statistical test focuses on the difference between treatment and control, rather than on the significance of the coefficient on

U B ijt

× Af ter (1996) t

× Cross − selling ijt

. The latter coefficient captures the difference be-

20

tween cross-sold universal-bank loans and any other kind of universal-bank loan, including those granted to firms that did not receive any underwriting product in the period around the universal-bank loan. The specification of our treatment and control groups safeguards that the respective firms received both at least one universal-bank loan and one underwriting product, either from the same or another bank.

We find that cross-sold universal-bank loans (treatment) led to greater within-firm increases in sales-growth volatility, namely 13.6% and even 20.0% after including controls, than plausibly exogenously non-cross-sold ones (control), namely 1.9% and 6.6% after including controls. The differences are significant at the 6% and 3% level, respectively. As mentioned above, we also test whether the sum of the coefficients on U B ijt and U B ijt

× Cross − selling ijt is different from zero. This is generally the case (p-values are 0.08 and 0.01 after including firm-year and loan(s)-year controls). In light of the fact that we include firm fixed effects and, thus, estimate within-firm effects, our results stem from universal-bank-financed firms embarking on riskier projects, rather than a reallocation of funds towards generally riskier firms. That is, firms actually became riskier after universal-bank financing, regardless of how risky they were at the outset, and especially so when the respective loans were cross-sold.

While comparing universal-bank loans issued only after the 1996 deregulation makes our treatment and control groups more comparable, we may still face potential endogeneity issues if (i) the respective firms existed before 1996 and were unable to attain loans from universal banks before 1996, and (ii) this selection affected cross-sold and non-cross-sold loans after

1996 differentially. To alleviate this concern, in the third column of Table 4, we re-run the

regression from the second column, and limit our sample to firms that did not enter into

loan agreements with universal banks only in or after 1996.

21

The resulting increase in salesgrowth volatility due to cross-sold universal-bank loans amounts to 12.9%, which is 13.7% higher than the corresponding increase experienced by the control group. This difference is significant at the 4% level.

Finally, we consider the possibility that cross-selling universal banks are different from non-cross-selling universal banks along time-varying characteristics that are correlated with borrower risk. One such explanation would be that cross-selling universal banks could finance riskier projects not because of the economies of scope in information acquisition, but simply because of the higher revenues generated from the bank-firm relationship. Since our treatment and control groups both comprise universal-bank loans, bank size and related too-big-to-fail considerations are unlikely to drive our estimates. In order to account for competing explanations above and beyond bank size, such as bank-level revenue fluctuations, including those that are potentially unrelated to the bank-firm relationship under considera-

tion, we include bank-year fixed effects in the last column of Table 4. The estimates suggest

that the within-firm increase in risk amounts to 19.1% for treated firms and 4.9% for the control group, with the difference being significant at the 4% level. The estimated treat-

21 To make the subsample of commercial-bank clients comparable, we also drop all those firms that entered into loan agreements with commercial banks only in or after 1996.

21

ment effects, i.e., the difference between treatment and control, across the second to fourth columns are remarkably stable around 14%. This implies that our results for cross-sold universal-bank loans are not driven simply by universal banks’ generating higher revenues from cross-selling.

To further buttress these results, in Table 5 we repeat the same estimations as in Table

4, and consider a firm-level risk measure that is based on market, rather than book, values,

namely the firm’s six-year stock-return volatility. From an a-priori perspective, it is unclear whether increases in real measures of volatility should be accompanied by increases in marketbased volatility: to the extent that market participants efficiently incorporate news into prices, stock-return volatility should move in lockstep with sales-growth volatility only insofar as the latter measure of real volatility is associated with news-releasing events.

The results suggest that treated firms experienced significantly greater within-firm increases in their stock-return volatility, ranging from 8.1% (fourth column) to 12.1% (second column), than the control group, namely 2.4% (third column) to 5.0% (second column). The respective p-values generally imply significance at the 10% level after including firm-year and loan(s)-year controls, except in the last column after including bank-year fixed effects. The difference between treatment and control is positive throughout all four columns.

We consider one more alternative risk measure, namely option-implied volatility. Thus far, we have used six-year sales-growth and stock-return volatilities to demonstrate the longlasting nature of the impact of universal banking on firm risk. However, this came at the cost of requiring firms to be publicly listed for a sufficiently long time (in our previous analysis, up to 6 + 5 + 6 = 17 years). Using option-implied volatilities relaxes this data requirement (but, instead, imposes an option-trading requirement). In doing so, we rely on the empirical options literature, most notably the finding that option-implied volatility does not just subsume information from past-realized volatility, but is also forward looking in the sense that it helps forecast future volatility (Christensen and Prabhala (1998)).

In Table 6, we use as dependent variable the natural logarithm of the five-year implied

volatility calculated using the volatility surface from option prices, σ ( implied i

) 5 y . We use for the first, pre-loan(s) year the implied volatility at the end of year t − 3 (or the very beginning of year t − 2) and for the second, post-loan(s) year the implied volatility measured at the end of year t

+ 2. As one can infer from the second column of Table 6, after including firm-year

and loan(s)-year controls, treated firms exhibit significantly higher within-firm increases in their implied volatility (of 2.9%) upon receiving cross-sold universal-bank loans than our

control group (-4.0%).

22

The difference between treatment and control groups is significant throughout all four columns, suggesting treatment effects of 5.1 to 7.9%. However, we do acknowledge that overall, the within-firm increases in option-implied volatility due to crosssold universal-bank loans, although they are significantly different from those for our control group, are not significantly higher, and sometimes even negative, than those associated with

22 We also notice that the inclusion of firm-year and loan(s)-year controls appears to be especially crucial for our estimates of the coefficient on U B ijt

in the first compared to the second column of Table 6.

22

commercial-bank loans.

Altogether, in this section, we have used the 1996 deregulation to build a control group of firms with universal-bank loans that were not cross-sold. We find that firms that received cross-sold universal-bank loans experienced significantly higher within-firm increases in risk.

Our confidence in these estimates is affirmed primarily by the robustness of our findings across three different risk measures based on sales, stock prices, and option prices.

3.2

Impact of Universal Banking on Default Risk and Loan Spreads

In this section, we scrutinize two loan-level outcome variables that are related to the riskiness of the borrower. First, we assess whether cross-sold universal-bank loans were associated with higher credit risk. This is to examine whether universal banks financed riskier firms, or excessively risky ones that were on the verge of defaulting. In doing so, we also exclude the possibility that our analysis of universal-bank loans may be systematically excluding (or prematurely dropping) firms that did not survive 6 + 5 + 6 = 17 years into the future, which was necessary for constructing our outcome variable in the previous section, because they were riskier.

For this purpose, in the first two columns of Table 7, we estimate specification (1) without

and with firm-level and loan-level controls, and use as dependent variable an indicator for whether the borrowing company went bankrupt in the ten years (our results are robust

to variations in the horizon) following the (potential) cross-selling period.

23

Note that the respective loans sample is not conditional on the availability of six-year-volatility data before and after the cross-selling period.

As before, our focal point is the test between treatment and control groups among

universal-bank loans. As can be seen in the first two columns of Table 7, cross-sold universal-

bank loans did not lead to greater default risk among borrower firms, i.e., the coefficient on

U B ijt

× Af ter (1996) t

× Cross − selling ijt is not significantly different from – and, if anything, is more negative than – the sum of the coefficients on U B ijt

× Af ter (1996) t

× N o CS ijt

×

U nderwriting (1994 / 95) ijt

, U B ijt

× Af ter (1996) t

× N o CS ijt

, and U B ijt

× Af ter (1996) t

×

U nderwriting (1994 / 95) ijt

. Additionally, in the second column, after including controls, we find that cross-sold universal-bank loans were no more or less likely to be associated with bankruptcy than commercial-bank loans.

Next, we consider the possibility that cross-selling universal banks extended loans at favorable terms, as measured by the so-called all-in-drawn spread, which is the sum of the

23 We use the following CRSP delisting codes to identify bankruptcy: any type of liquidation (400-490); price fell below acceptable level; insufficient capital, surplus, and/or equity; insufficient (or non-compliance with rules of) float or assets; company request, liquidation; bankruptcy, declared insolvent; delinquent in filing; non-payment of fees; does not meet exchange’s financial guidelines for continued listing; protection of investors and the public interest; corporate governance violation; and delist required by Securities

Exchange Commission (SEC).

23

spread over LIBOR and any annual fees paid to the lender syndicate. We test this in the

last two columns of Table 7 with the natural logarithm of the all-in-drawn spread (in bps)

as dependent variable. In the third and fourth column, respectively, we find that cross-sold universal-bank loans were associated with 7 .

8 − (6 .

2 + 0 .

6 − 5 .

8) = 6 .

8 (fourth column) to

32 .

2 − (12 .

9 + 13 .

3 − 6 .

8) = 12 .

8 (third column) percent lower spreads (significant at the

7% and 1% level, respectively) than the control group after the 1996 deregulation. This shows that universal banks’ realization of economies of scope across financial products led to benefits that were (partially) passed on to their clients. Our estimates confirm the (abovementioned) findings reported by Drucker and Puri (2005) in an empirical setting that uses variation in the incidence of cross-selling generated by the 1996 deregulation.

3.3

Impact of Universal Banking on Productivity and Investment

Thus far, we have considered only measures related to firm-level risk as outcomes. We now turn to the question as to whether the additional risk of universal-bank-financed firms was rewarded by higher productivity. We find that cross-selling has enabled universal banks not just to finance riskier projects, but that cross-sold universal-bank loans also led to lasting increases in the borrowing firms’ TFP, capital expenditure, and market capitalization.

Our analysis proceeds much like that in the previous section. In addition, we generalize the definition of our cross-selling variable with respect to the time lag between loans and underwriting services. In our previous analysis, we were constrained by the data requirements of our sales-growth-volatility and stock-return-volatility measures, in that both were calculated based on six years of data. To safeguard that the two pre- and post-loan volatility measures do not overlap, we had to preserve equal distances between the time window for calculating these risk measures and the actual cross-selling window. For this purpose, we built our cross-selling variable around the loan-issue year, comprising two years before and after. In this section, we consider outcome variables that are based on a single annual observation. As such, we do not face any trade-offs regarding the distance of the time horizon of our outcome variables from the cross-selling window, allowing us to cover any distance between loans and underwriting services from 0 to 5 years.

We now define a pair of loans and non-loan products to be cross-sold if they are issued to the same firm by the same universal bank within five years (from year t − 4 to t ) and,

most importantly, in any order.

24

We modify our loans-related variables along the same lines, in that U B ijt now indicates whether, given any loan transactions from (and including) year t − 4 to (and including) year t , at the time of any loan transaction any one of the lead

24 For example, imagine a firm i interacted only with a universal bank j , and the firm received a loan in year t − 1 and an equity-underwriting service in t , then Cross − selling ijt one from years t to t + 3. Thereafter, Cross − selling ijt is zero in year t − is zero again. Next, imagine firm

1, but equal to i still received a loan in t − 1, but the equity-underwriting service only in year t + 2. Then, Cross − selling ijt from years t − 1 to t + 1, but becomes equal to one from t + 2 to t + 3, and is zero thereafter.

is zero

24

arrangers was a universal bank.

25

As the time window of our cross-selling variable is no longer defined solely by the firmloan year, we can also include all firm-years without any loan transactions or security issues.

The resulting sample comprises all publicly listed firms for which all our non-banking-related variables are available. This corresponds to what we label as our “Compustat sample” in

the third panel of Table 3. We then run regression specification (1) on this sample, including

all firm-year observations from 1984, and cluster the standard errors at the firm-year level.

Note that we now also include firm-year observations for which all loans-related variables are zero, so that firms with no loan in a given year become the omitted category.

In Table 8, we use the natural logarithm of firm-level total factor productivity (TFP) in

year t + 1 as dependent variable. We use TFP in t + 1 because our TFP measure is the result of an estimation, conducted by Imrohoroglu and Tuzel (2014), that uses as input variables capital and labor in t , which are potentially correlated with our right-hand-side variables.

In the first column, we find that universal-bank loans are, in general, not associated with higher TFP, but the universal-bank-loan coefficient becomes significantly more positive after

1996. In the remaining columns, we adopt our tests for the treatment and control group,

as in the first three columns of Tables 4 to 6. Note that as we include data from before

1996, we always include U B ijt

× Af ter (1996) t in our regressions, the coefficient on which is, however, extremely close to zero, suggesting that incidences of post-1996 cross-selling explain that coefficient away. As before, we also provide p-values from two (two-sided) test statistics. Besides testing for the difference between the treatment and control group, which is our primary interest, we also test whether the sum of the coefficients on U B ijt

,

U B ijt

× Af ter (1996) t

, and U B ijt

× Af ter (1996) t

× Cross − selling ijt is different from zero, i.e., whether cross-sold universal-bank loans led to higher within-firm increases in TFP compared to years in which firms received no loans. The respective difference is significant at the 1% level across all outcome variables and specifications discussed hereafter.

In the second column and the third column after including firm-year and loan(s)-year controls, we see that firms with cross-sold universal-bank loans experienced TFP increases compared to our control group of firms with non-cross-sold universal-bank loans after 1996.

While the effect for the control group is -0.6% and -0.8%, the effect for cross-sold universalbank loans amounts to 2.3% and 2.4% in the second and third column, respectively, and the differences are both significant at the 1% level. Note that we can preclude the possibility that cross-selling to the treated group of borrowers leads to higher TFP simply because firms that demand multiple financial products are more productive overall. This is due to two reasons. First, if firms making use of multiple financial products are more productive in a time-invariant sense, then this should be absorbed by the firm fixed effects. Second, and more importantly, firms in both the treatment and the control group received at least one loan and one underwriting service within five years. Hence, they do not differ in the

25 The bank fixed effects for all (eventual) universal banks, as well as the commercial-bank fixed effect (for all commercial banks that never become universal banks pooled together), are defined accordingly.

25

number of financial products used, just in whether they received both products from the same source.

After limiting our sample in the last column of Table 8, as we have already done for

our risk measures, to firms that had universal-bank lending relationships before and after

1996, the treatment effect – i.e., the difference between treatment and control – becomes starker, amounting to 3 .

9% (significant at the 1% level). To sum up, we have presented evidence that the risk-increasing impact of cross-sold universal-bank loans was accompanied by within-firm increases in TFP. The difference between treatment and control group ranges from 2.9% to 3.9%. What is more, our estimated treatment effects are relatively long-lived, up to six years, due to the definition of the five-year cross-selling window and an additional lag due to measurement of TFP in year t + 1.

To show that these increases in productivity also translate into increases in actual invest-

ment and higher market capitalization, we re-run the regressions from Table 8, and use as

dependent variable the natural logarithm of the firm’s capital expenditure in year t as well as the natural logarithm of the firm’s market value of equity in year t . The results are in

Tables 9 and 10, respectively, and demonstrate that our previous findings for TFP are also

valid for these measures. After including firm-year and loan(s)-year controls, our treatment group yields a within-firm increase in capital expenditure of 10.1%, which is 6.9% more – and significantly so (at the 1% level) – than the increase for the control group (cf. third

column of Table 9). In the last column, the treatment effect barely changes for our limited

firm sample: the treatment group’s effect amounts to 12.4%, which is 5.7% higher than for the control group, and this difference is again significant at the 1% level.

We yield a very similar picture with regard to changes in firms’ market capitalization, i.e.,

their market values of equity, in Table 10. Across all three specifications, we find that the

treatment group yields a significantly higher within-firm increase in market capitalization upon receiving a cross-sold universal-bank loan than the control group, implying treatment effects ranging from 4.6 to 9.0%. The respective differences between treatment and control are significant at least at the 4% level. Note that these treatment effects are unlikely to be due to equity-raising activities, which would be associated with equity-underwriting products, as firms in both the treatment and the control group received underwriting services concurrently with their universal-bank loans; the only difference lies in whether these products were crosssold by the same universal bank.

Finally, as we used as outcome variables TFP in year t + 1 for the above-mentioned reasons, but capital expenditure and market capitalization in year t , we also show in Tables

B.2 and B.3 that our results for the latter two dependent variables are robust to using their

realizations in year t + 1.

To conclude, we have found that after the 1996 deregulation, cross-sold universal-bank loans were associated with significantly higher TFP, capital expenditure, and market capitalization, as compared to a control group of firms with universal-bank loans that were not

26

cross-sold for plausibly exogenous reasons. The treatment effects for capital expenditure and market capitalization are larger in size than those we have found for TFP, and measure up with the previously discussed treatment effects for our risk measures. This set of results complements our findings for firm-level risk in a meaningful way, and guides the economic interpretation. Our evidence is consistent with universal-bank relationships resulting in firms making risky, productivity-increasing investments, which implies that there is a real component to the increase in risk that we document in this paper. Still, this leaves open the question of whether the increases in productivity and capital expenditure are commensurate with the increase in risk, which would be necessary to assess efficiency gains. To the extent that we find risk and productivity to move in the same direction, our evidence at the very least does not contradict the possibility of firm-level efficiency gains from universal banking.

3.4

Impact of Universal Banking on IPO Age

The evidence from the loan data suggests that universal-bank-financed firms were riskier, but the analysis is confined to publicly listed and, therefore, mature firms. We now complement our loans-based analysis with evidence on firms earlier in their life cycle , and scrutinize the impact of universal banking on the age of firms when they go public. In our previous analysis, by looking at firms that were already public, we were able to explore the depth of bank-firm relationships in the form of cross-selling by universal banks. For the IPO-level analysis, we are confined to a bank-level identification, where we compare the average age of

IPOs with universal banks as bookrunners to the average age of IPOs with investment-bank bookrunners before and after 1996.

In this section, we consider whether the risk-taking behavior of universal banks documented in the loan data extends to their role as underwriters, and we use as an alternative risk measure the age of firms at the time of their IPOs, following the logic that younger firms are typically riskier (Pastor and Veronesi (2003)). We find that after the 1996 deregulation, universal banks took significantly younger firms public than investment banks. Looking at the effect of universal banking on IPO age may also be a fruitful exercise in the sense that previous research by Brown and Kapadia (2007) and Fink, Fink, Grullon, and Weston

(2010) has found that higher idiosyncratic risk in the U.S. stock market could be driven by increased entry into the stock market of younger and riskier firms. Brown and Kapadia

(2007) hypothesize that increasing financial-market development may have been a decisive catalyst. In order to evaluate this possibility, we use SDC IPO data in conjunction with additional data on firm age at the time of the IPO to test whether universal banks took younger firms public than other underwriters, most notably investment banks.

In Figure 4, we plot the market-value-weighted average age of firms at the time of their

IPOs and the proportion of IPOs accompanied by universal banks. We observe a negative correlation that is fairly strong after 1996. Note, also, that the IPO market share of universal banks soars around 1996 as well.

27

Figure 4: Market-value-weighted Average Age of Firm at IPO vs. Fraction of

IPOs run by Universal Banks (1976-2006).

Source: own analysis based on SDC IPOs and data based on Loughran and Ritter (2004).

Given that commercial banks that are not yet universal banks cannot be bookrunners, our sample is limited to universal banks and investment banks. That is, investment banks are the control group, a subset of which was eventually acquired by commercial or already existing universal banks. In a difference-in-differences setup, we test whether following the

1996 deregulation, universal banks took younger firms public than investment banks whose scope of banking activities was unaffected by the deregulation. For a universal bank to be treated under the 1996 deregulation, it needs to be established before the deregulation.

Against this background, we run the following regression specification:

IP O age ijt

= β

1

U B j

+ β

2

U B j

× Est.

(1996) j

× Af ter (1996) t

+ β

3

U B j

× Est.

(1996) j

+ β

4

Af ter (1996) t

+ β

5

Eventually U B through M & A j

+ β

6

X ijt

+ β

7 industry i

+ µ t

+ ijt

, (3) where IP O age ijt is firm i ’s age in years at the time of the IPO, U B j is an indicator variable for whether the bookrunner was a universal bank (formed through a merger or through opening a Section 20 subsidiary), Af ter (1996) t is an indicator variable for whether the IPO was on or after August 1, 1996, Est.

(1996) j is an indicator variable for whether a universal bank (through M&A or Section 20) was established prior to August 1, 1996, Eventually U B through M & A j is an indicator variable for whether the bookrunner, which was still an invest-

28

ment bank, eventually becomes a universal bank through M&A, X ijt denotes firm and IPO characteristics, and industry i and µ t are industry and IPO-year fixed effects, respectively.

Standard errors are clustered at the bookrunner level.

We use the 1996 deregulation as a shock to the scope of banking activities engaged in by universal banks to examine whether universal banks established before that date took

younger firms public following the deregulation. In the first column of Table 11, we estimate

(3) without any firm or IPO-specific controls.

The difference-in-differences estimate for treated universal banks compared to the control group of pure investment banks, which is captured by the coefficient on U B j

× Est.

(1996) j

× Af ter (1996) t

, is significantly negative

(at the 1% level), reflecting 8.4 years younger and, thus, riskier IPOs.

This difference-in-differences estimate drops to 5.3 and 5.2 years, but remains statistically significant, after including firm-level and IPO-level controls in the second and third column, respectively. Focusing on the estimates in the third column, we make two observations.

First, universal banks established before and after August 1, 1996 took 2.8 (= the sum of all three universal-bank coefficients) and 6.9 years younger firms public, respectively (both effects are significant at the 1% level). Second, the difference-in-differences estimate for treated universal banks implies 5.2 years younger firms, which corresponds to one-quarter of

a standard deviation of IPO age (cf. summary statistics in the last panel of Table 3). The

economic significance of these effects renders it likely that the deregulation of bank scope constitutes an important channel by which financial development led to riskier IPOs.

The 1996 deregulation carries particular significance for the underwriting activities of universal banks. Besides the increased scope for cross-selling, interaffiliate loans could be

used to cross-finance riskier investment-banking operations.

26

An alternative explanation may be that commercial banks inherited the risk-taking properties of the smaller investment banks that they acquired or merged with. To control for this possibility, we include an indicator for whether the bookrunner in question was an investment bank that eventually merged with a commercial or an already existing universal bank, Eventually U B M & A j

, on the right-hand side. However, the respective coefficient is significantly larger (at the 1% and

10% level, respectively, implying that these investment banks took older firms public) than the coefficient on U B j and the sum of all three coefficients for universal banks established before 1996. Therefore, this alternative explanation seems unlikely.

In the fourth column, we delineate our treatment effects by the universal banks’ mode of establishment, namely whether the universal bank in question was established through

M&A or through opening a Section 20 subsidiary. The difference-in-differences estimates are both negative, but only significantly so for universal banks established through M&A. This chimes with our findings in the loan data insofar as given that universal banks established through M&A engage in a wider range of banking activities, they also have more possibilities for realizing economies of scope. This, in turn, enables them to take on more risk, here in

26 The Federal Reserve Act limits such loans to any single securities affiliate to 10% of a bank’s capital.

29

the form of taking younger firms public.

In order to evaluate whether these results may be driven by any other characteristics that differ between universal banks established through M&A and Section 20 subsidiaries, we collected key summary statistics for the bank-holding companies in our sample a year before to a year after becoming universal banks.

As Table B.1 shows, universal banks

established through M&A are typically larger than Section 20 subsidiaries. Such mergers constitute one-time increases in total assets, net income, cash flow (approximated by EBIT), and the number of employees, whereas Section 20 subsidiaries grow more continuously over

time.

27

Most importantly, both types of universal banks are strikingly similar in their equityto-assets and cash-to-assets ratios. That is, higher risk taking by universal banks established through M&A cannot be readily explained by a different leverage position or excess cash.

Loan-to-assets ratios are somewhat higher for universal banks formed through Section 20 subsidiaries, as investment-banking operations are a smaller portion of their business model.

Finally, we consider another, market-structure-based competing explanation for the younger age of firms that were taken public by universal banks. Commercial banks entering the underwriting business as newly formed universal banks naturally lack a track record for IPOs.

This may, in turn, force them to take younger firms public. That is, in an effort to build a track record, universal banks potentially took young and particularly risky firms public – something that incumbent investment banks would not be likely to do.

To test for this possibility, we include interactions of U B through M & A j and U B through

Section 20 j with IP O count jt

, which is equal to the number of IPOs accompanied by the respective universal banks, up to and including the IPO in question (of firm i with bookrunner j at time t ). If inexperience and lack of a track record were responsible for our findings, then one would expect the respective interaction effects to be positive, indicating that universal banks with an established track record of IPOs took older firms public. While the interaction effect for Section 20 subsidiaries is positive and significant at the 8% level in the last column of

Table 11, the interaction effect for universal banks established through M&A is insignificant

and fairly close to zero. At a coefficient of 0.119 for Section 20 subsidiaries, however, it would take at least twice as many IPOs run by Section 20 subsidiaries than the average IPO

count for that group (see summary statistics in the last panel of Table 3) to eliminate the

IPO-age effect compared to regular investment banks. Still, it remains plausible that the age of firms taken public by universal banks is, in general, governed by the degree to which universal banks have less experience in the IPO market than incumbent investment banks.

However, as our remaining estimates are close to those in the fourth column of Table 11, the

explanatory power of this alternative explanation for the effects of increased bank scope on

IPO age seems limited.

27

Note that we could not include universal banks for which the data do not cover all three time periods; i.e., we had to drop universal banks that were established right when the data became available (1987) or that were eventually acquired by other banks.

30

4 Conclusion

In this paper, we focus on a narrowly defined set of deregulatory events that expanded the scope of banking in the U.S. to evaluate bank scope as a determinant of firm-level real outcomes. Our empirical strategy exploits a deregulatory shock to the scope of banking activities in 1996. We use detailed data on bank-firm interactions to identify plausibly exogenous incidences of cross-selling following the 1996 deregulation. In this manner, we provide evidence that the advent of universal banking improved the access to finance for risky enterprises through economies of scope in the provision of concurrent lending and underwriting.

Our results indicate that universal-bank-financed firms in cross-selling relationships exhibit significantly higher volatility than a control group of firms contracting with non-crossselling universal banks, but are not any more likely to default in the long run. We also find that cross-sold universal-bank loans are associated with long-lasting increases in TFP, capital expenditure, and market capitalization. We then investigate the implications of universal banking for the cross-section of publicly traded firms with respect to their riskiness early in their life cycle. In particular, we show that following the 1996 deregulation, universal banks have contributed to increased entry into the stock market of risky firms by taking significantly younger firms public than their investment-bank competitors. Universal banks are, thus, better intermediaries in the sense that they relax financial constraints for volatile but productive firms.

Our paper highlights two avenues for future research. In light of recent proposals to limit the scope of banking, we have taken a first step towards providing an empirical backdrop against which to evaluate the set of activities banks should be allowed to engage in. Namely, we provide evidence that there may be firm-level efficiency gains from allowing banks to engage in concurrent lending and underwriting of corporate securities, which runs counter to recent proposals of re-establishing the Glass-Steagall Act. These benefits would have to be balanced against costs such as risks associated with banks becoming too big to fail and other concerns of macroeconomic fragility.

Besides our potentially policy-relevant contribution, our findings are in accordance with previous research on the evolution of firm-level volatility in the U.S. Based on earlier observations by Campbell, Lettau, Malkiel, and Xu (2001), Comin and Philippon (2006) document empirically that idiosyncratic firm risk has been rising over the past thirty years. Our results suggest that bank-scope deregulation may have contributed to this phenomenon. Therefore, another direction for future research could be to quantify the explanatory power of increased bank scope for the observed run-up in firm-level fluctuations.

The channels through which universal banking impacts firm risk in our paper chime with previous evidence in the literature. First, Brown and Kapadia (2007) and Fink, Fink,

Grullon, and Weston (2010) find that higher idiosyncratic risk in the U.S. stock market is

31

associated with younger firms that went public. This is in line with our documented effect of increased bank scope allowing universal banks to act as bookrunners for IPOs of younger firms. Second, any attempt to explain the observed increases in firm risk must contend with the finding of Davis, Haltiwanger, Jarmin, and Miranda (2007) that volatility has been increasing for publicly listed but not for private firms. Our proposed explanation can accommodate this dichotomy, because equity underwriting is a major cross-selling product, so universal banking affects primarily firms that eventually go, or already are, public.

Placing our findings into a broader macroeconomic context, we end by emphasizing that the strongest surge in firm risk is well known to have taken place in the 1990s (Brandt, Brav,

Graham, and Kumar (2010)), and was accompanied by a simultaneous boom in measured

TFP of public firms. While our findings suggest that bank-scope deregulation could have been an important driver of these effects, this period also saw many other innovations in financial markets and in the real economy. We therefore view our paper as potentially motivating further studies evaluating the importance of universal banking as a contributing factor in the comovement of risk and productivity among publicly listed firms in the U.S.

32

References

Amore, M. D., C. Schneider, and A. Zaldokas

(2013): “Credit Supply and Corporate

Innovation,” Journal of Financial Economics , 109(3), 835–855.

Ang, J. S., and T. Richardson (1994): “The Underwriting Experience of Commercial

Bank Affiliates prior to the Glass-Steagall Act: A Reexamination of Evidence for Passage of the Act,” Journal of Banking & Finance , 18(2), 351–395.

Benfratello, L., F. Schiantarelli, and A. Sembenelli (2008): “Banks and Innovation: Microeconometric Evidence on Italian Firms,” Journal of Financial Economics ,

90(2), 197–217.

Bertrand, M., A. Schoar, and D. Thesmar (2007): “Banking Deregulation and Industry Structure: Evidence from the French Banking Reforms of 1985,” Journal of Finance ,

62(2), 597–628.

Bharath, S., S. Dahiya, A. Saunders, and A. Srinivasan (2007): “So What Do I

Get? The Bank’s View of Lending Relationships,” Journal of Financial Economics , 85(2),

368–419.

Bhargava, R., and D. R. Fraser (1998): “On the Wealth and Risk Effects of Commercial

Bank Expansion into Securities Underwriting: An Analysis of Section 20 Subsidiaries,”

Journal of Banking & Finance , 22(4), 447–465.

Brandt, M. W., A. Brav, J. R. Graham, and A. Kumar (2010): “The Idiosyncratic

Volatility Puzzle: Time Trend or Speculative Episodes?,” Review of Financial Studies ,

23(2), 863–899.

Brown, G., and N. Kapadia

(2007): “Firm-specific Risk and Equity Market Development,” Journal of Financial Economics , 84(2), 358–388.

Calomiris, C. W., and T. Pornrojnangkool (2009): “Relationship Banking and the

Pricing of Financial Services,” Journal of Financial Services Research , 35(3), 189–224.

Campbell, J. Y., M. Lettau, B. G. Malkiel, and Y. Xu (2001): “Have Individual

Stocks Become More Volatile? An Empirical Exploration of Idiosyncratic Risk,” Journal of Finance , 56(1), 1–43.

Campello, M., J. R. Graham, and C. R. Harvey (2010): “The Real Effects of Financial Constraints: Evidence from a Financial Crisis,” Journal of Financial Economics ,

97(3), 470–487.

Chava, S., A. Oettl, A. Subramanian, and K. Subramanian (2013): “Banking

Deregulation and Innovation,” Journal of Financial Economics , 109(3), 759–774.

33

Chava, S., and M. R. Roberts (2008): “How Does Financing Impact Investment? The

Role of Debt Covenants,” Journal of Finance , 63(5), 2085–2121.

Chodorow-Reich, G.

(2014): “The Employment Effects of Credit Market Disruptions:

Firm-level Evidence from the 2008-9 Financial Crisis,” Quarterly Journal of Economics ,

129(1), 1–59.

Christensen, B., and N. Prabhala (1998): “The Relation between Implied and Realized

Volatility,” Journal of Financial Economics , 50(2), 125–150.

Comin, D. A., and T. Philippon (2006): “The Rise in Firm-Level Volatility: Causes and Consequences,” in NBER Macroeconomics Annual 2005, Volume 20 , pp. 167–228.

National Bureau of Economic Research, Inc.

Cornaggia, J., Y. Mao, X. Tian, and B. Wolfe (2013): “Does Banking Competition

Affect Innovation?,” Journal of Financial Economics .

Cornett, M. M., E. Ors, and H. Tehranian (2002): “Bank Performance around the

Introduction of a Section 20 Subsidiary,” Journal of Finance , 57(1), 501–521.

Correa, R., and G. A. Suarez (2009): “Firm Volatility and Banks: Evidence from U.S.

Banking Deregulation,” Board of Governors of the Federal Reserve System, Finance and

Economics Discussion Series Working Paper No. 2009-46 .

Davis, S. J., J. Haltiwanger, R. Jarmin, and J. Miranda (2007): “Volatility and

Dispersion in Business Growth Rates: Publicly Traded versus Privately Held Firms,” in NBER Macroeconomics Annual 2006, Volume 21 , pp. 107–180. National Bureau of

Economic Research, Inc.

Drucker, S., and M. Puri (2005): “On the Benefits of Concurrent Lending and Underwriting,” Journal of Finance , 60(6), 2763–2799.

(2007): “Banks in Capital Markets,” in Empirical Corporate Finance , ed. by B. E.

Eckbo, vol. 1 of Handbook of Corporate Finance , chap. 5, pp. 189–232. Elsevier/North-

Holland.

Ferreira, M. A., and P. Matos (2012): “Universal Banks and Corporate Control:

Evidence from the Global Syndicated Loan Market,” Review of Financial Studies , 25(9),

2703–2744.

Fink, J., K. E. Fink, G. Grullon, and J. P. Weston (2010): “What Drove the

Increase in Idiosyncratic Volatility during the Internet Boom?,” Journal of Financial and

Quantitative Analysis , 45(5), 1253–1278.

Gande, A., M. Puri, A. Saunders, and I. Walter (1997): “Bank Underwriting of

Debt Securities: Modern Evidence,” Review of Financial Studies , 10(4), 1175–1202.

34

Geyfman, V., and T. J. Yeager (2009): “On the Riskiness of Universal Banking: Evidence from Banks in the Investment Banking Business Pre- and Post-GLBA,” Journal of

Money, Credit and Banking , 41(8), 1649–1669.

Greenwood, J., J. M. Sanchez, and C. Wang (2010): “Financing Development: The

Role of Information Costs,” American Economic Review , 100(4), 1875–1891.

Harris, M., C. Opp, and M. M. Opp (2014): “Macroprudential Bank Capital Regulation in a Competitive Financial System,” Unpublished working paper, University of Chicago,

University of Pennsylvania, and UC Berkeley.

Herrera, A. M., and R. Minetti

(2007): “Informed Finance and Technological Change:

Evidence from Credit Relationships,” Journal of Financial Economics , 83(1), 223–269.

Hoffmann, F., R. Inderst, and M. M. Opp (2014): “Regulating Deferred Incentive

Pay,” Unpublished working paper, University of Frankfurt and UC Berkeley.

Imrohoroglu, A., and S. Tuzel (2014): “Firm Level Productivity, Risk, and Return,”

Management Science .

James, C.

(1992): “Relationship-Specific Assets and the Pricing of Underwriter Services,”

Journal of Finance , 47(5), 1865–1885.

Jayaratne, J., and P. E. Strahan (1996): “The Finance-Growth Nexus: Evidence from

Bank Branch Deregulation,” Quarterly Journal of Economics , 111(3), 639–670.

Kanatas, G., and J. Qi (1998): “Underwriting by Commercial Banks: Incentive Conflicts,

Scope Economies, and Project Quality,” Journal of Money, Credit and Banking , 30(1),

119–133.

(2003): “Integration of Lending and Underwriting: Implications of Scope

Economies,” Journal of Finance , 58(3), 1167–1191.

Kerr, W. R., and R. Nanda (2009): “Democratizing Entry: Banking Deregulations,

Financing Constraints, and Entrepreneurship,” Journal of Financial Economics , 94(1),

124–149.

(2010): “Banking Deregulations, Financing Constraints, and Firm Entry Size,”

Journal of the European Economic Association , 8(2-3), 582–593.

Kroszner, R. S., and R. G. Rajan (1994): “Is the Glass-Steagall Act Justified? A Study of the U.S. Experience with Universal Banking before 1933,” American Economic Review ,

84(4), 810–832.

Ljungqvist, A., F. Marston, and W. J. Wilhelm (2006): “Competing for Securities

Underwriting Mandates: Banking Relationships and Analyst Recommendations,” Journal of Finance , 61(1), 301–340.

35

Loughran, T., and J. Ritter (2004): “Why Has IPO Underpricing Changed Over

Time?,” Financial Management , 33(3), 5–37.

Morgan, D., B. Rime, and P. E. Strahan (2004): “Bank Integration and State Business

Cycles,” Quarterly Journal of Economics , 119(4), 1555–1584.

Neuhann, D., and F. Saidi (2014): “Information Sensitivity and the Scope of Financial

Intermediation,” Unpublished working paper, University of Pennsylvania and University of Cambridge.

Olley, G. S., and A. Pakes (1996): “The Dynamics of Productivity in the Telecommunications Equipment Industry,” Econometrica , 64(6), 1263–1297.

Opp, C. C., M. M. Opp, and M. Harris (2013): “Rating Agencies in the Face of

Regulation,” Journal of Financial Economics , 108(1), 46–61.

Pastor, L., and P. Veronesi (2003): “Stock Valuation and Learning about Profitability,”

Journal of Finance , 58(5), 1749–1790.

Puri, M.

(1996): “Commercial Banks in Investment Banking: Conflict of Interest or Certification Role?,” Journal of Financial Economics , 40(3), 373–401.

Saunders, A., E. Strock, and N. G. Travlos (1990): “Ownership Structure, Deregulation, and Bank Risk Taking,” Journal of Finance , 45(2), 643–654.

Schenone, C.

(2004): “The Effect of Banking Relationships on the Firm’s IPO Underpricing,” Journal of Finance , 59(6), 2903–2958.

Stiglitz, J. E., and A. Weiss (1981): “Credit Rationing in Markets with Imperfect

Information,” American Economic Review , 71(3), 393–410.

Yasuda, A.

(2005): “Do Bank Relationships Affect the Firm’s Underwriter Choice in the

Corporate-Bond Underwriting Market?,” Journal of Finance , 60(3), 1259–1292.

36

5 Tables

Table 1: Timeline of Universal Banks

Section 20

Established before August 1, 1996

M&A

BankBoston (later acquired by Fleet)

Bankers Trust (later acquired by Bank of America)

Bank of America

Bank of New England (defunct since 1991)

Bank One (later acquired by J.P. Morgan)

BankSouth

Barnett Bank (later acquired by NationsBank)

Chase Manhattan (later acquired by J.P. Morgan)

Chemical Bank (later acquired by Chase Manhattan)

Citicorp

Dauphin Deposit Corp.

First Chicago NBD

First Union

Fleet (later acquired by Bank of America)

Huntington Bancshares

J.P. Morgan

Liberty National Bank

Marine Midland Bank (later acquired by HSBC Bank USA)

Mellon (later acquired by BNY)

Credit Suisse (First Boston)

Deutsche Bank USA

Equitable (later acquired by SunTrust)

National City (later acquired by PNC)

National Westminster Bank USA (later acquired by Fleet)

NationsBank (later acquired by Bank of America)

Norstar (later acquired by Fleet)

Norwest (later acquired by Wells Fargo)

PNC

Security Pacific Bank (later acquired by Bank of America)

SouthTrust (later acquired by Wachovia/First Union)

SunTrust

Established on or after August 1, 1996

HSBC Bank USA

Sovran Bank (later acquired by NationsBank)

Travelers Group

Citigroup

Wells Fargo

BB&T

BNY

Commerce Bancshares

CoreStates/Philadelphia National Bank

(later acquired by First Union)

Crestar Bank

First Tennessee

KeyBank

U.S. Bancorp

Wachovia (first acquired by First Union and later by Wells Fargo

Citigroup emerged as a result of the merger of Travelers Group and Citicorp on October 8, 1998. Before,

Travelers Group became a universal bank by our definition through a series of mergers, most notably with investment banks Smith Barney and Salomon Brothers, and Citicorp had registered a Section 20 subsidiary.

Given the size of this merger of equals, we do not treat either one as the surviving entity and, instead, label

Citigroup as a separate universal bank established through M&A in 1998.

37

Table 2: Summary Statistics for Treatment and Control Group in 1993

Variable

σ

σ

(

(

\ i

) return

1988 , 1993 i

) 1988 , 1993

TFP in 1993

Capital expenditure in 1993 $bn

Market capitalization in 1993 $bn

Sales in 1993 $bn

No. of employees in thousands in 1993

No. of loans until 1993

No. of underwriting mandates until 1993

Treatment

Mean

(Std. dev.)

N

0.15

(0.14)

0.38

(0.19)

0.66

(0.26)

0.14

(0.45)

2.09

(5.84)

2.16

(6.23)

13.39

(32.59)

1.12

(1.44)

3.12

(3.55)

666

659

693

944

955

960

954

977

977

Control

Mean

(Std. dev.)

0.17

(0.16)

0.41

(0.20)

0.70

(0.36)

0.12

(0.51)

1.63

(5.71)

1.77

(7.69)

10.24

(36.44)

1.16

(1.38)

3.15

(3.48)

N

153

150

177

270

271

272

268

282

282 p-value

0.12

0.05

0.16

0.61

0.25

0.38

0.17

0.68

0.89

Notes: σ ( \ i

) 1988 , 1993 and σ ( return i

) 1988 , 1993 are the six-year standard deviations of firm i ’s sales growth and stock return, respectively, from 1988 to 1993. In the last panel, the number of loans and underwriting mandates is calculated as a given firm’s total number of all transactions from 1984 to 1993.

38

Table 3: Summary Statistics

Loans sample (starting in 1984)

σ ( \ i

) t,t +5

σ ( return i

) t,t +5

σ ( implied i

)

5 y t

UB through M&A ∈ { 0 , 1 }

UB through Section 20 ∈ { 0 , 1 }

Sales at close in 2010 $bn

No. of employees in thousands

Deal size/Assets in %

Refinancing ∈ { 0 , 1 }

No. of lead arrangers

Bankruptcy ∈ { 0 , 1 }

All-in-drawn spread in bps

No. of UBs M&A

No. of UBs M&A with Est.(1996) = 1

No. of UBs Section 20

No. of UBs Section 20 with Est.(1996) = 1

Firm-loan-years sample (starting in 1996)

UB ∈ { 0 , 1 }

Cross-selling (CS) ∈ { 0 , 1 }

No CS ∈ { 0 , 1 }

Underwriting(1994/95) ∈ { 0 , 1 }

No CS × Underwriting(1994/95) ∈ { 0 , 1 }

Compustat sample (starting in 1984)

T F P i t +1

CapEx t i

(in 2010 $bn)

M arketCap t i

(in 2010 $bn)

IPO sample (starting in 1976)

IPO age in years

UB through M&A ∈ { 0 , 1 }

UB through Section 20 ∈ { 0 , 1 }

Eventually UB M&A ∈ { 0 , 1 }

Sales in 2010 $bn

No. of employees in thousands

Book-value leverage

Gross spread in %

IPO count if UB through M&A = 1

IPO count if UB through Section 20 = 1

No. of UBs M&A

No. of UBs M&A with Est.(1996) = 1

No. of UBs Section 20

No. of UBs Section 20 with Est.(1996) = 1

Mean Std. dev.

Min Max N

0.19

0.48

0.50

0.04

0.56

3.34

14.25

27.52

0.48

1.13

0.18

189.85

0.15

0.23

0.22

0.20

0.50

13.52

56.61

49.03

0.50

0.35

0.39

138.71

0.01

0.12

Mean Std. dev.

Min

0.72

0.45

0

0.38

0.28

0.48

0.45

0

0

0.11

0.09

0.32

0.29

0

0

1.90

3.04

9,090

8,808

0.12

0

1.88

1

0 1

0.00

485.15

5,147

15,650

15,650

15,650

0.00

2,100.00

15,650

0.01

3,960.40

15,650

0

1

1

6

15,650

17,147

0 1 8,627

0.70

1,490.02

13,859

8

6

37

28

Max

1

1

1

1

1

N

1,264

1,264

1,264

1,264

1,264

Mean Std. dev.

Min Max N

0.66

0.24

0.35

1.35

0.01

0.00

15.00

59.28

66,986

121,377

2.65

13.67

0.00

780.50

122,760

Mean Std. dev.

Min Max N

14.43

0.11

0.06

0.26

0.31

1.47

0.19

7.48

85.96

23.87

20.28

0.31

0.24

0.44

1.40

6.21

0.21

1.33

59.37

18.96

0.00

165.00

0.00

1.00

0.00

0.00

0.00

1.00

1.00

41.70

0.00

203.00

0.00

0.89

0.70

1

1

20.25

229

74

3,827

3,827

3,827

3,827

3,827

3,827

3,827

3,827

405

231

7

4

14

10

Notes: σ ( \ i

) t,t +5 and σ ( return stock return, respectively, from t to t i

) t,t +5

+5.

σ are the six-year standard deviations of firm i ’s sales growth and

( implied i

)

5 y t is firm i ’s five-year implied volatility calculated using the volatility surface from option prices (source: Option Metrics) in t . Est.(1996) is an indicator variable for whether the universal bank in question was established before August 1, 1996.

39

Table 4: Impact of Universal-bank Financing on Sales-growth Volatility – Firmloan-years Sample, Within-firm Effects

Treatment ([1]+[2])

Control ([1]+[3]+[4]+[5])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

Controls

Bank FE

Bank-year FE

Firm FE

Year FE

Sample

N

× Cross-selling

[3] UB × No CS × Underwriting(1994/95)

[4] UB

[5] UB

×

×

No CS

Underwriting(1994/95)

0.092

(0.17)

-0.059

(0.08)

-0.086

(0.13)

N

Y

N

Y

Y

13.6%

1.9%

0.06

0.08

0.072

(0.07)

0.064

(0.06) (0.06)

0.095

(0.17)

-0.058

(0.07)

-0.095

(0.12)

Y

Y

N

Y

Y ln( σ ( \ i

) 6 y )

20.0% 12.9%

6.6%

0.03

-0.8%

0.04

0.01

0.124*

(0.07)

0.076

0.14

0.089

(0.09)

0.040

(0.07)

0.101

(0.20)

-0.084

(0.10)

-0.114

(0.15)

Y

Y

N

Y

Y

Post 1996 Post 1996 Post 1996, Post 1996 rel. w. UB

2,528 2,528 before 1996

1,936 2,528

0.202

(0.19)

-0.075

(0.08)

-0.194

(0.13)

Y

N

Y

Y

N

19.1%

4.9%

0.04

0.04

0.116

(0.09)

0.075

(0.07)

Notes: All regressions include firm fixed effects. In general, the sample consists of two observations per year during which a firm received at least one loan, where the loans sample consists of all completed syndicated loans (package level) of publicly listed firms, subject to availability of the dependent variable. Furthermore, we limit the sample to the years including and after 1996.

σ ( \ i

) 6 y is the six-year standard deviation of firm i ’s sales growth from t − 7 to t − 2 for the first, pre-loan(s)-year observation, and from t + 2 to t + 7 for the second, post-loan(s)-year observation.

U B is an indicator variable for whether, given any loan transactions in a year, at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the firm’s loan year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan in year t was associated with a cross-sold underwriting product by the same bank from t − 2 to t + 2.

Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product from t − 2 to t + 2 which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. The first, pre-loan(s)-year observation uses information from the last trading day of year t − 3, and the second, post-loan(s)-year observation uses information from the last trading day of year t + 2. Control variables include the log of the firm’s sales, the log of its number of employees, the log of the ratio of the average deal size across all loans in a given year over the firm’s assets, and the average value of the refinancing indicator (the latter two loans-related variables are always zero for the first, pre-loan(s)-year observation). The sample in the third column comprises only firms that did not enter into loan agreements with universal banks only in or after 1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together (omitted category). Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the lead-arranger level for both observations of each firm’s loan year, treating each (eventual) universal bank individually and pooling all pure commercial banks) are in parentheses.

40

Table 5: Impact of Universal-bank Financing on Stock-return Volatility – Firmloan-years Sample, Within-firm Effects

Treatment ([1]+[2])

Control ([1]+[3]+[4]+[5])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

Controls

Bank FE

Bank-year FE

Firm FE

Year FE

Sample

N

× Cross-selling

[3] UB × No CS × Underwriting(1994/95)

[4] UB

[5] UB

×

×

No CS

Underwriting(1994/95)

9.4%

3.1%

0.12

0.01

0.067*

(0.04)

0.027

(0.04)

0.120

(0.09)

-0.020

(0.04)

-0.136

(0.09)

N

Y

N

Y

Y ln( σ ( return i

) 6 y )

12.1%

5.0%

0.07

0.00

0.089**

(0.04)

0.032

(0.04)

0.115

(0.09)

-0.020

(0.04)

-0.134

(0.09)

Y

Y

N

Y

Y

11.0%

2.4%

0.03

0.01

0.104**

(0.04)

0.006

(0.04)

0.092

(0.11)

-0.050

(0.05)

-0.122

(0.10)

Y

Y

N

Y

Y

8.1%

2.8%

0.40

0.10

Post 1996 Post 1996 Post 1996, Post 1996

2,404 2,404 rel. w. UB before 1996

1,846 2,404

0.055

(0.05)

0.026

(0.04)

0.140

(0.09)

0.015

(0.05)

-0.182**

(0.09)

Y

N

Y

Y

N

Notes: All regressions include firm fixed effects. In general, the sample consists of two observations per year during which a firm received at least one loan, where the loans sample consists of all completed syndicated loans (package level) of publicly listed firms, subject to availability of the dependent variable. Furthermore, we limit the sample to the years including and after 1996.

σ ( return i

)

6 y is the six-year standard deviation of firm i ’s stock return from t − 7 to t − 2 for the first, pre-loan(s)-year observation, and from t + 2 to t + 7 for the second, post-loan(s)-year observation.

U B is an indicator variable for whether, given any loan transactions in a year, at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the firm’s loan year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan in year t was associated with a cross-sold underwriting product by the same bank from t − 2 to t + 2.

Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product from t − 2 to t + 2 which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. The first, pre-loan(s)-year observation uses information from the last trading day of year t − 3, and the second, post-loan(s)-year observation uses information from the last trading day of year t + 2. Control variables include the log of the firm’s sales, the log of its number of employees, the log of the ratio of the average deal size across all loans in a given year over the firm’s assets, and the average value of the refinancing indicator (the latter two loans-related variables are always zero for the first, pre-loan(s)-year observation). The sample in the third column comprises only firms that did not enter into loan agreements with universal banks only in or after 1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together (omitted category). Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the lead-arranger level for both observations of each firm’s loan year, treating each (eventual) universal bank individually and pooling all pure commercial banks) are in parentheses.

41

Table 6: Impact of Universal-bank Financing on Option-implied Volatility – Firmloan-years Sample, Within-firm Effects

Treatment ([1]+[2])

Control ([1]+[3]+[4]+[5])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

Controls

Bank FE

Bank-year FE

Firm FE

Year FE

Sample

N

× Cross-selling

[3] UB × No CS × Underwriting(1994/95)

[4] UB

[5] UB

×

×

No CS

Underwriting(1994/95)

-6.0%

-11.6%

0.05

0.05

ln( σ ( implied i

) 5 y )

2.9%

-4.0%

0.01

0.35

2.1%

-3.0%

0.06

0.55

-0.123***

(0.03)

-0.030

(0.03)

0.063*** 0.059***

(0.02) (0.02)

-0.117

(0.11)

-0.000

(0.04)

0.124

-0.148

(0.12)

0.002

(0.03)

0.136

(0.12)

N

Y

N

Y

Y

(0.12)

Y

Y

N

Y

Y

-0.031

(0.04)

0.052**

(0.02)

-0.281**

(0.12)

0.021

(0.03)

0.261**

(0.12)

Y

Y

N

Y

Y

-3.2%

-11.1%

0.00

0.44

Post 1996 Post 1996 Post 1996, Post 1996

2,614 2,614 rel. w. UB before 1996

1,644 2,614

-0.114**

(0.04)

0.082***

(0.03)

-0.089

(0.11)

0.004

(0.03)

0.088

(0.11)

Y

N

Y

Y

N

Notes: All regressions include firm fixed effects. In general, the sample consists of two observations per year during which a firm received at least one loan, where the loans sample consists of all completed syndicated loans (package level) of publicly listed firms, subject to availability of the dependent variable. The sample is limited to the years including and after 1996 due to the availability of the dependent variable.

σ ( implied i

)

5 y is firm i ’s five-year implied volatility calculated using the volatility surface from option prices (source: Option

Metrics) in t − 2 for the first, pre-loan(s)-year observation, and in t + 2 for the second, post-loan(s)-year observation.

U B is an indicator variable for whether, given any loan transactions in a year, at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the firm’s loan year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan in year t was associated with a cross-sold underwriting product by the same bank from t − 2 to t + 2. Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product from t − 2 to t + 2 which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or

1995. The first, pre-loan(s)-year observation uses information from the last trading day of year t − 3, and the second, post-loan(s)-year observation uses information from the last trading day of year t + 2. Control variables include the log of the firm’s sales, the log of its number of employees, the log of the ratio of the average deal size across all loans in a given year over the firm’s assets, and the average value of the refinancing indicator (the latter two loans-related variables are always zero for the first, pre-loan(s)-year observation).

The sample in the third column comprises only firms that did not enter into loan agreements with universal banks only in or after 1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together

(omitted category). Public-service, energy, and financial-services firms are dropped. Robust standard errors

(clustered at the lead-arranger level for both observations of each firm’s loan year, treating each (eventual) universal bank individually and pooling all pure commercial banks) are in parentheses.

42

Table 7: Impact of Universal-bank Financing on Loan Characteristics – Loans

Sample

Treatment ([1]+[2]+[3])

Control ([1]+[2]+[4]+[5]+[6])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

Bankruptcy

-4.6%

-2.1%

0.37

0.09

-0.020

(0.03)

0.092**

-3.1%

-0.9%

0.44

0.27

-0.020

(0.03)

0.087** ln(All-in-drawn spread)

-24.0%

-11.2%

0.01

0.00

0.070

(0.09)

0.012

[2] UB × A.(1996)

[3] UB × A.(1996) × Cross-selling

(0.04)

-0.118***

(0.04)

[4] UB × A.(1996) × No CS × Underwriting(1994/95) -0.148*

(0.04)

-0.098**

(0.04)

-0.142*

(0.06)

-0.322***

(0.05)

-0.129**

-9.0%

-2.2%

0.07

0.09

0.016

(0.06)

-0.028

(0.04)

-0.078*

(0.05)

-0.062

[5] UB

[6] UB

×

×

A.(1996)

A.(1996)

×

×

No CS

Underwriting(1994/95)

(0.08)

-0.045

(0.03)

0.100

(0.09)

(0.08)

-0.036

(0.04)

0.102

(0.06)

-0.133***

(0.04)

0.068

(0.06)

(0.07)

-0.006

(0.03)

0.058

Log of sales at close in 2010 $

Log of no. employees

(0.09)

-0.023***

(0.00)

0.000

(0.00)

(0.07)

-0.199***

(0.01)

-0.083***

(0.01)

Log of deal size/assets

Refinancing indicator

Bank FE

Industry FE

Year FE

Y

Y

Y

-0.006*

(0.00)

0.000

(0.01)

Y

Y

Y

Y

Y

Y

0.032*

(0.02)

0.047***

(0.01)

Y

Y

Y

Sample

N

All

8,627

All

8,627

All

13,859

All

13,859

Notes: The sample consists of all completed syndicated loans (package level) of publicly listed firms, subject to availability of the dependent variable. Furthermore, we limit the sample to loans with at most one lead arranger that was a universal bank. The dependent variable in the first two columns is an indicator variable for whether the borrowing company went bankrupt (according to CRSP delisting codes) in the ten years following the cross-selling period (i.e., t + 3 to t + 12), and the dependent variable in the last two columns is the natural logarithm of the all-in-drawn spread (in bps), which is the sum of the spread over LIBOR and any annual fees paid to the lender syndicate.

U B is an indicator variable for whether, given any loan transactions in a year, at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the firm’s loan year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan in year t was associated with a cross-sold underwriting product by the same bank from t − 2 to t + 2. Conversely,

N o CS indicates whether a firm that received a loan in year t also received an underwriting product from t − 2 to t + 2 which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. Bank fixed effects are included for all lead arrangers of the respective loan that are or eventually become universal banks, whereas all remaining commercial banks are grouped together

(omitted category). Industry fixed effects are based on two-digit SIC codes. Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the lead-arranger level, treating each (eventual) universal bank individually and pooling all pure commercial banks) are in parentheses.

43

Table 8: Impact of Universal-bank Financing on Total Factor Productivity – Compustat Sample, Long-run Within-firm Effects

Treatment ([1]+[2]+[3])

Control ([1]+[2]+[4]+[5]+[6])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

[3] UB

Controls

Bank FE

Firm FE

Year FE

Sample

N

×

×

A.(1996)

A.(1996) × Cross-selling

[4] UB × A.(1996) × No CS × Underwriting(1994/95)

[5] UB

[6] UB

×

×

A.(1996)

A.(1996)

×

×

No CS

Underwriting(1994/95) ln( T F P i t +1

)

2.3% 2.4%

-0.6%

0.01

0.00

-0.8%

0.00

0.00

3.8%

-0.1%

0.00

0.00

-0.015** -0.010

(0.01)

0.019**

(0.01)

(0.01)

-0.000

(0.01)

-0.010

(0.01)

-0.003

(0.01)

0.033*** 0.037***

-0.006

(0.01)

-0.003

(0.01)

0.047***

(0.01) (0.01) (0.01)

-0.052** -0.050** -0.081***

(0.02) (0.02)

0.021*** 0.022***

(0.01)

0.035*

(0.01)

0.033

(0.03)

0.030***

(0.01)

0.059**

N

N

Y

Y

All

66,986

(0.02)

N

Y

Y

Y

All

66,986

(0.02)

Y

Y

Y

Y

All

66,986

(0.02)

Y

Y

Y

Y

Rel. w. UB before 1996

50,242

Notes: All regressions include firm fixed effects. In general, the sample consists of all available firm-year observations from Compustat, the unit of observation is the firm-year level.

T F P i t +1 is firm i ’s average total factor productivity in year t + 1 from Imrohoroglu and Tuzel (2014).

U B is an indicator variable for whether, given any loan transactions from (and including) year t − 4 to (and including) year t , at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan from year t − 4 to t was associated with a cross-sold underwriting product by the same bank anytime from t − 4 to t . Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product anytime from t − 4 to t which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. Control variables are measured in year t , and include the log of the firm’s sales, the log of its number of employees, the log of the average ratio of deal size across all loans over the firm’s assets from t − 4 to t , and the average value of the refinancing indicator from t − 4 to t . The sample in the last column comprises only firms that did not enter into loan agreements with universal banks only in or after

1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together (omitted category).

Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the firm-year level) are in parentheses.

44

Table 9: Impact of Universal-bank Financing on Capital Expenditure – Compustat Sample, Long-run Within-firm Effects

Treatment ([1]+[2]+[3])

Control ([1]+[2]+[4]+[5]+[6])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

[3] UB

Controls

Bank FE

Firm FE

Year FE

Sample

N

×

×

A.(1996)

A.(1996) × Cross-selling

[4] UB × A.(1996) × No CS × Underwriting(1994/95)

[5] UB

[6] UB

×

×

A.(1996)

A.(1996)

×

×

No CS

Underwriting(1994/95)

0.159***

(0.02)

0.002

(0.02)

0.00

0.00

ln( CapEx t i

)

28.4%

18.6%

0.189***

(0.02)

-0.094***

(0.02)

0.189***

(0.02)

-0.024

10.1%

3.2%

0.00

0.00

0.061***

(0.01)

-0.022

(0.02)

0.062***

(0.01)

12.4%

6.7%

0.01

0.00

0.075***

(0.02)

-0.039**

(0.02)

0.088***

(0.01)

-0.138*** -0.186***

(0.06)

0.055***

(0.04)

0.025**

(0.05)

0.035**

N

N

Y

Y

(0.02)

0.060

(0.05)

N

Y

Y

Y

(0.01)

0.106**

(0.04)

Y

Y

Y

Y

(0.02)

0.182***

(0.05)

Y

Y

Y

Y

All

121,377

All

121,377

All

121,377

Rel. w. UB before 1996

93,513

Notes: All regressions include firm fixed effects. In general, the sample consists of all available firm-year observations from Compustat, the unit of observation is the firm-year level.

CapEx t i is firm i ’s capital expenditure in year t .

U B is an indicator variable for whether, given any loan transactions from (and including) year t − 4 to (and including) year t , at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan from year t − 4 to t was associated with a cross-sold underwriting product by the same bank anytime from t − 4 to t . Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product anytime from t − 4 to t which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. Control variables are measured in year t , and include the log of the firm’s sales, the log of its number of employees, the log of the average ratio of deal size across all loans over the firm’s assets from t − 4 to t , and the average value of the refinancing indicator from t − 4 to t . The sample in the last column comprises only firms that did not enter into loan agreements with universal banks only in or after 1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together (omitted category). Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the firm-year level) are in parentheses.

45

Table 10: Impact of Universal-bank Financing on Market Capitalization – Compustat Sample, Long-run Within-firm Effects

Treatment ([1]+[2]+[3])

Control ([1]+[2]+[4]+[5]+[6])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

[3] UB

Controls

Bank FE

Firm FE

Year FE

Sample

×

×

A.(1996)

A.(1996) × Cross-selling

[4] UB × A.(1996) × No CS × Underwriting(1994/95)

[5] UB

[6] UB

×

×

A.(1996)

A.(1996)

×

×

No CS

Underwriting(1994/95) ln( M arketCap t i

)

27.7% 16.0%

21.3% 11.4%

0.01

0.00

0.04

0.00

0.091*** 0.128*** 0.045***

(0.02) (0.02) (0.02)

0.053*** -0.054***

(0.02) (0.02)

-0.005

(0.02)

0.203*** 0.120***

(0.02) (0.01)

0.142** 0.065

(0.06) (0.05)

0.077*** 0.057***

N

N

Y

Y

(0.02)

-0.080

(0.05)

N

Y

Y

Y

(0.01)

-0.048

(0.05)

Y

Y

Y

Y

All All All

N 122,760 122,760 122,760

23.0%

14.0%

0.00

0.00

0.061***

(0.02)

0.005

(0.02)

0.164***

(0.02)

-0.007

(0.06)

0.095***

(0.02)

-0.014

(0.06)

Y

Y

Y

Y

Rel. w. UB before 1996

94,706

Notes: All regressions include firm fixed effects. In general, the sample consists of all available firm-year observations from Compustat, the unit of observation is the firm-year level.

M arketCap t i is firm i ’s market value of equity in year t .

U B is an indicator variable for whether, given any loan transactions from (and including) year t − 4 to (and including) year t , at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan from year t − 4 to t was associated with a cross-sold underwriting product by the same bank anytime from t − 4 to t . Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product anytime from t − 4 to t which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. Control variables are measured in year t , and include the log of the firm’s sales, the log of its number of employees, the log of the average ratio of deal size across all loans over the firm’s assets from t − 4 to t , and the average value of the refinancing indicator from t − 4 to t . The sample in the last column comprises only firms that did not enter into loan agreements with universal banks only in or after 1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together (omitted category). Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the firm-year level) are in parentheses.

46

Table 11: Impact of Universal-bank Underwriting on Age of Firms at their IPOs

UB

UB

UB

Industry FE

IPO-year FE

N

×

×

Est.(1996)

Est.(1996)

UB M&A

UB M&A

UB M&A

UB M&A

×

×

×

UB Section 20

UB Section 20

After(1996)

×

×

Gross spread in %

×

Book-value leverage

A.(1996)

Est.(1996)

Est.(1996)

IPO count

UB Section 20

UB Section 20 × Est.(1996) × A.(1996)

Log of sales in 2010 $

Log of no. employees

× A.(1996)

Est.(1996)

IPO count

Eventually UB through M&A

IPO age in years

-3.939*

(2.34)

-8.429***

(3.03)

-6.938***

(2.19)

-5.335*

(3.03)

-6.862***

(2.27)

-5.213*

(3.00)

12.291*** 9.372*** 9.227***

(3.50) (3.47) (3.48)

1.345

(1.53)

2.345*

(1.38)

Y

Y

-8.825*** -9.032***

(2.16) (1.97)

-5.824*

(3.45)

-7.037**

(3.20)

11.983*** 11.544***

(3.61) (3.54)

0.017

(0.01)

-4.215** -5.134***

(1.71) (1.59)

-1.052

(3.18)

-3.273

(3.26)

2.190

(3.52)

2.219

(3.62)

0.119*

0.944

(1.57)

-1.189

(0.83)

0.804

(1.62)

-0.919

(0.88)

0.603

(1.63)

-0.924

(0.88)

(0.07)

0.783

(1.63)

-0.955

(0.87)

2.260*** 2.189*** 2.176*** 2.170***

(0.32) (0.31) (0.31) (0.32)

2.514*** 2.409*** 2.414*** 2.402***

(0.50) (0.51) (0.51) (0.52)

6.473*** 6.440*** 6.415***

Y

Y

(1.84)

0.117

(0.27)

Y

Y

(1.85)

0.108

(0.27)

Y

Y

(1.85)

0.080

(0.27)

Y

Y

3,827 3,827 3,827 3,827 3,827

Notes: IP O age is firm i ’s age in years at the time of its IPO. The unit of observation is a firm’s IPO.

U B ( through M & A or U B through Section 20) is an indicator variable for whether the bookrunner was a universal bank (formed through a merger or through opening a Section 20 subsidiary).

Af ter (1996) is an indicator for whether the IPO date was on or after August 1, 1996.

Est.

(1996) indicates whether a universal bank (through M&A or Section 20) was established prior to August 1, 1996.

Eventually U B through M & A is an indicator variable for whether the bookrunner, which was still an investment bank, eventually becomes a universal bank through M&A.

IP O count denotes the number of IPOs accompanied by universal banks, up to and including the current IPO. Book-value leverage is winsorized at the 1 st and 99 th percentiles. All firm-level explanatory variables are measured at the end of the IPO year. Industry fixed effects are based on two-digit SIC codes. Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the bookrunner level) are in parentheses.

47

Supplementary Appendix (Not for Publication)

A Supplementary Figures

Figure A.1: Loan-weighted Average Six-year [t,t+5] Stock-return Volatility associated with Loans granted to Public Firms by Commercial and Universal Banks

(1987-2005).

Post-1996 loans by universal banks are split into cross-sold and non-cross-sold loans, where cross-sold loans are defined as loans whose debtor firms also received an underwriting product from the same universal bank anytime within the last five years. Source: own analysis based on CRSP/Compustat, DealScan loan data, and SDC underwriting data.

48

Figure A.2: Ratios between the Mean for the Treatment Group and the Mean for the Control Group in the Pre-deregulation Period.

Variables are the same as in

Table 2.

49

B Supplementary Tables

Table B.1: Summary Statistics for Universal Banks Established through M&A and Section 20 Subsidiaries

M&A t = − 1 t = 0

Section 20 t = 1 t = − 1 t = 0 t = 1

Total assets in 2010 $bn

Total equity/assets in %

Cash balance/assets in %

513.7

1,110.2

1,101.0

47.82

51.54

52.33

(129.3) (305.3) (230.4) (33.11) (35.26) (36.22)

7.547

6.977

7.959

8.670

8.648

8.903

(1.032) (0.833) (1.455) (1.994) (1.772) (2.659)

4.944

5.177

4.776

5.484

5.768

5.338

(2.571) (2.764) (2.040) (1.663) (2.320) (2.027)

65.78

51.77

51.86

66.95

67.03

67.96

Total loans/assets in %

Net income in 2010 $bn

(7.724) (23.02) (20.76) (6.654) (6.744) (6.349)

6.675

5.227

12.69

0.330

0.333

0.350

(2.542) (3.591) (0.308) (0.275) (0.285) (0.375)

10.17

8.415

20.15

0.525

0.544

0.560

EBIT in 2010 $bn

(3.248) (6.264) (1.618) (0.458) (0.482) (0.607)

No. of employees in thousands 132.2

227.9

228.2

14.20

15.44

15.34

(39.06) (60.74) (55.33) (9.815) (10.13) (10.18)

N 5 30

Notes: This table reports means with standard deviations in parentheses, for universal banks established through M&A in the first three columns and for Section 20 subsidiaries in the last three columns. The data are taken from the respective banks’ call reports.

t indicates the year of the respective call report, and t = 0 denotes the first call report after the bank becomes a universal bank, and t = − 1 and t = 1 correspond to the call reports one year before and after the call report used for t = 0, respectively. Cash balance is the sum of non-interest-bearing balances and currency and coin, and interest-bearing balances in U.S. offices. EBIT is net income before income taxes, extraordinary items, and other adjustments on a fully taxable equivalent basis.

50

Table B.2: Impact of Universal-bank Financing on Future Capital Expenditure –

Compustat Sample, Long-run Within-firm Effects

Treatment ([1]+[2]+[3])

Control ([1]+[2]+[4]+[5]+[6])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

[3] UB

Controls

Bank FE

Firm FE

Year FE

Sample

N

×

×

A.(1996)

A.(1996) × Cross-selling

[4] UB × A.(1996) × No CS × Underwriting(1994/95)

[5] UB

[6] UB

×

×

A.(1996)

A.(1996)

×

×

No CS

Underwriting(1994/95)

0.111***

(0.02)

-0.002

(0.02) ln( CapEx t +1 i

)

22.0%

12.7%

0.00

0.00

0.136***

(0.02)

-0.085***

(0.02)

0.169***

(0.02)

-0.043

6.0%

1.2%

0.02

0.00

0.026*

(0.02)

-0.006

(0.02)

0.040***

(0.01)

7.5%

4.2%

0.15

0.00

0.035**

(0.02)

-0.019

(0.02)

0.059***

(0.02)

-0.173*** -0.249***

(0.06)

0.035**

(0.05)

0.015

(0.05)

0.037**

(0.02)

0.084

(0.06)

N

(0.01)

0.150***

(0.04)

Y

(0.02)

0.238***

(0.05)

Y N

N

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

All

107,572

All

107,572

All

107,572

Rel. w. UB before 1996

82,353

Notes: All regressions include firm fixed effects. In general, the sample consists of all available firm-year observations from Compustat, the unit of observation is the firm-year level.

CapEx t +1 i is firm i ’s capital expenditure in year t + 1.

U B is an indicator variable for whether, given any loan transactions from (and including) year t − 4 to (and including) year t , at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan from year t − 4 to t was associated with a cross-sold underwriting product by the same bank anytime from t − 4 to t . Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product anytime from t − 4 to t which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. Control variables include the log of the firm’s sales, the log of its number of employees (both measured at the end of year t + 1), the log of the average ratio of deal size across all loans over the firm’s assets from t − 4 to t , and the average value of the refinancing indicator from t − 4 to t . The sample in the last column comprises only firms that did not enter into loan agreements with universal banks only in or after 1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together (omitted category). Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the firm-year level) are in parentheses.

51

Table B.3: Impact of Universal-bank Financing on Future Market Capitalization

– Compustat Sample, Long-run Within-firm Effects

Treatment ([1]+[2]+[3])

Control ([1]+[2]+[4]+[5]+[6])

Test of Treatment = Control (p-value)

Test of Treatment = 0 (p-value)

[1] UB

[2] UB

[3] UB

Controls

Bank FE

Firm FE

Year FE

Sample

N

×

×

A.(1996)

A.(1996) × Cross-selling

[4] UB × A.(1996) × No CS × Underwriting(1994/95)

[5] UB

[6] UB

×

×

A.(1996)

A.(1996)

×

×

No CS

Underwriting(1994/95) ln( M arketCap t +1 i

)

22.5% 11.3%

15.1%

0.01

0.00

6.7%

0.05

0.00

0.049*** 0.075***

(0.02)

0.069***

(0.02)

(0.02)

-0.010

(0.02)

-0.000

(0.02)

0.042**

(0.02)

0.160*** 0.071***

(0.02)

0.107*

(0.01)

0.013

(0.06) (0.05)

0.045*** 0.032**

N

N

Y

Y

All

(0.02)

-0.066

(0.06)

N

Y

Y

Y

All

(0.01)

-0.020

(0.05)

Y

Y

Y

Y

All

109,064 109,064 109,064

16.6%

9.4%

0.01

0.00

0.013

(0.02)

0.045**

(0.02)

0.108***

(0.02)

-0.046

(0.07)

0.064***

(0.02)

0.018

(0.06)

Y

Y

Y

Y

Rel. w. UB before 1996

83,627

Notes: All regressions include firm fixed effects. In general, the sample consists of all available firm-year observations from Compustat, the unit of observation is the firm-year level.

M arketCap t +1 i is firm i ’s market value of equity in year t + 1.

U B is an indicator variable for whether, given any loan transactions from (and including) year t − 4 to (and including) year t , at the time of any loan transaction any one of the lead arrangers was a universal bank formed through a merger or through opening a Section 20 subsidiary.

Af ter (1996) is an indicator for whether the year in question was in 1996 or later.

Cross − selling is an indicator for whether any loan from year t − 4 to t was associated with a cross-sold underwriting product by the same bank anytime from t − 4 to t . Conversely, N o CS indicates whether a firm that received a loan in year t also received an underwriting product anytime from t − 4 to t which was not issued by the same bank.

U nderwriting (1994 / 95) is an indicator for whether the firm in question did not receive a cross-sold loan but, instead, an underwriting product from an investment bank in 1994 or 1995. Control variables include the log of the firm’s sales, the log of its number of employees (both measured at the end of year t + 1), the log of the average ratio of deal size across all loans over the firm’s assets from t − 4 to t , and the average value of the refinancing indicator from t − 4 to t . The sample in the last column comprises only firms that did not enter into loan agreements with universal banks only in or after 1996. Bank fixed effects are included for all lead arrangers of all loans in a given year that are or eventually become universal banks, whereas all remaining commercial banks are grouped together (omitted category). Public-service, energy, and financial-services firms are dropped. Robust standard errors (clustered at the firm-year level) are in parentheses.

52

Download