Mutual Funding

Mutual Funding.

Javier Gil-Bazo Peter Hoffmann Sergio Mayordomo

Universitat Pompeu Fabra ECB Banco de España

and Barcelona GSE

January 2019

Abstract

Using data on Spanish mutual funds, we show that bank-affiliated funds provide funding support to their parent company via purchases of bonds in the primary market. Funding from affiliated funds increases when it is most valuable, i.e., in times of financial stress and to riskier banks with limited access to funding. Funding support is not limited to bonds but also takes place through bank deposits and purchases of short- term paper. These trades benefit banks at the expense of their affiliated mutual fund investors. To minimize the negative impact on their asset management business, banks steer funding support towards funds without a performance fee and funds that cater to the less-performance sensitive investors. Finally, we provide evidence that the ability to use affiliated funds as a source of funding enabled Spanish banks to limit credit rationing to the nonfinancial sector during the 2008-2012 period.

JEL Codes: G23, G32, G21Keywords: Financial Conglomerates, Mutual Funds, Agency

Conflicts, Financial Crisis, Sovereign Debt Crisis, Credit Dry Ups

. The views expressed are those of the authors and do not necessarily reflect those of the ECB, Banco de España, and/or the ESCB and should not be represented as such. We would like to thank John C. Adams, Geert Bekaert, Claire Célérier, José Miguel Gaspar, Augustin Landier, Steven Ongena, Pedro Matos, José Luis Peydró, Jean-Charles Rochet, Rafael Repullo, María Rodríguez-Moreno, Neal Stoughton, Javier Suárez as well as seminar/conference audiences at the Conference of Institutional and Individual Investors: Saving for Old Age (University of Bath), the 2015 Asset Management Conference (ESMT Berlin) ESSEC, the 2018 Spanish Finance Association Meeting, 5th Banco de España – CEMFI Workshop, VU Amsterdam, University of Zurich, and Comisión Nacional del Mercado de Valores for useful comments and suggestions. Author contact: [email protected], [email protected], and [email protected]. 1. Introduction

As financial conglomerates expand the array of financial services they provide, new potential conflicts of interest arise between the conglomerate and its clients. However, the mere existence of conflicts of interest does not necessarily imply that they will be abused since abusing conflicts of interest entails potential legal costs, reputational costs, and loss of future business. In this paper, we study for the first time the use of affiliated mutual funds managed by bank-controlled firms as a source of funding for banks, and investigate whether conglomerates strategically balance the benefits of abusing this conflict of interest against its costs.

While conflicts of interest within financial conglomerates have already received attention in the academic literature (see Mehran and Stultz, 2007, for a survey), we unveil a novel channel. It is rooted in the observation that a large number of mutual funds around the world are not independent entities but controlled directly or indirectly by a financial intermediary that is engaged in banking and other financial activities.1 We posit that banks can use affiliated mutual funds as an alternative source of funding, in particular at times when liquidity is scarce and funding markets are under stress. Funds managed by bank-affiliated asset management firms can help their parent banks decrease funding costs by artificially increasing demand at the funding stage.

Unlike other conflicts of interest that have been investigated in the literature, the conflict of interest that we study in this paper is likely to emerge only under extraordinary circumstances, such as those experienced in the post-Lehman period by

European banks. Indeed, while funding from affiliated funds is of little value to banks in normal times, it becomes a valuable alternative source of funding in times of financial

1 Ferreira, Matos, and Pires (2018) show that around 40% of mutual funds around the world are run by commercial banks. The share of bank-affiliated funds in the United States is lower (around 20%), but has picked up in recent years with the repeal of the Glass-Steagal Act in 1999. 1 distress, when access to stable sources of funding is not guaranteed. In other words, the possibility of using affiliated mutual funds as an alternative source of funding for the parent bank is a potential conflict of interest that stays dormant until a banking crisis hits. Since banking crises are rare, this conflict of interest has historically remained under the radar of both investors and regulators. Moreover, the rare-event nature of this conflict of interest makes it particularly appropriate to investigate the strategic response of banking groups to a change in the trade-off between the benefits of abusing conflicts of interest and the costs of doing so. In particular, the global financial crisis sharply reduced credit supply, increased banks’ funding costs and therefore, raised banks’ incentives to use alternative sources of funding. Also, the impact of the crisis was very heterogeneous across banks (Cassola et al, 2013).

Furthermore, the conflict of interest that we analyze in this paper has potentially far- reaching consequences that go beyond the redistribution of wealth between different groups of stakeholders in the financial conglomerate. More specifically, since the motivation for banks to use affiliated mutual funds’ resources is to overcome temporary funding constraints, it is natural to ask whether the availability of such an alternative source of funding actually helps banks weather periods of financial distress, and limit the extent of credit rationing to the nonfinancial sector. In this sense, our paper also contributes to an emerging literature that investigates how financial institutions direct investment from healthy business units to business units in financial distress.

In order to achieve our goal, we investigate funding support by Spanish bank-affiliated mutual funds in the primary market using data from the 2000-2012 period. Spain is a particularly appropriate laboratory for our purposes for three reasons. First, its mutual fund industry is largely dominated by banks, whose asset management divisions account for roughly three quarters of the total assets under management (AUM).

Second, the Spanish banking sector was under severe stress for an extended period due to a bursting real estate bubble and the European sovereign debt crisis. Banks’ access to key wholesale markets was severely limited, which increased their reliance on central bank liquidity.2 Moreover, intense competition for stable funding sources drove up retail deposit rates.3 Consequently, the incentives to tap related entities such as affiliated mutual funds for extra funding were very high. While portfolio decisions are in the hands of asset managers, they have incentives to avoid the negative consequences of their parent bank’s financial distress, such as cost-cutting policies, loss of captive clients, and liquidation of the bank’s asset management division.4

Third, unlike in the US, related-party transactions involving mutual funds are not prohibited in Spain. Although Spanish law mandates that asset management companies act in the best interest of mutual fund investors, it gives asset management firms a great deal of discretion in dealing with conflicts of interest. Consequently, the only explicit constraints on purchases of affiliated firms’ securities by Spanish mutual funds are general position limits aimed at ensuring portfolio diversification.5

Consistent with our hypothesis, we show that bank-affiliated mutual funds purchase more of their parents’ bond issues in the primary market than their non-affiliated peers.

Moreover, we find that excess funding from affiliated mutual funds, “mutual funding,” strongly increases in times of financial stress, consistent with banks reacting to changes in incentives to use their affiliated mutual funds as a source of funding. Also, this

2 See Garcia-de-Andoain, Manganelli and Hoffmann (2014) for empirical evidence. 3 See e.g. “Savings wars from Italy to Portugal drive bank costs higher,” Bloomberg News (online), Oct 21, 2011. 4 The defaults of Bankia and Banco Popular are two illustrative examples. Thus, the total AUM by funds affiliated to Bankia dropped by 8.7% from March 2012 to June 2012 (a month after its default). This drop is 2.6 times higher than the one observed for the rest of asset management groups. Regarding Banco Popular, it lost 7% of its managed products, including mutual and pension funds, six months after its default in June 2017. 5 Appendix B provides a detailed description of the regulation of related-party transactions and minimum diversification limits in Spain.

3 support is more prevalent among banks that rely more on central bank liquidity, have a higher ratio of non-performing loans, and have recently experienced a rating downgrade. Funding support is particularly strong for subordinated debt, which is more sensitive to financial stress than senior debt. In sum, both our time-series and cross- sectional results support the hypothesis that banks resort more to affiliated funds as an alternative source of funding when the incentives to abuse the conflict of interest are stronger.

The funding support to parent banks by affiliated mutual funds documented in this paper is economically significant. When aggregating bond purchases across all funds of the same asset management group (AMG), we find that excess purchases of parent debt by affiliated funds account for 2.85% of the total amount issued throughout the entire sample period, corresponding to 514 million EUR per bank or 14.4 billion EUR on aggregate. While funding support is absent in normal times, it amounts to 7% of the amount issued in crisis times (11.9 billion EUR, representing 83% of total funding support during the entire sample period).

In addition to excess purchases of banks’ debt, funding support also takes place through term deposits and short-term paper (STP).6 In contrast, we find no evidence of funding support through purchases of preferred shares, which Spanish banks were able to place directly to retail clients, or through purchases of seasoned equity offerings.

6 Our main analysis focuses on funding support through bond purchases rather than other instruments. This decision is motivated by the fact that bonds, unlike short-term paper and deposits, provide medium and long-term financing that is more difficult to substitute through other sources, in particular central bank liquidity. Moreover, data availability limits our ability to perform some of the tests for the two other fixed income instruments.

Banks can also have affiliated funds trade in the secondary market in order to reduce yields around the time of new issues and hence decrease debt refinancing costs. We choose to focus on the primary market because the proceeds of the issue accrue directly to the parent bank and therefore constitute genuine debt financing. In any case, we do not find evidence of bond price support through purchases in the secondary market.

Funding support is detrimental to affiliated mutual fund investors’ interests. In particular, bond issues with primary market participation by affiliated mutual funds display 29 (33) basis points lower abnormal returns on the issuance day (week) than other issues of the same bank, which is indicative of overpricing. We also find that purchases of affiliated bonds underperform simultaneous purchases of non-affiliated bonds by the same fund. Similarly, deposits with the parent bank earn lower deposit rates than those of non-affiliated banks. Therefore, we discard the alternative explanation that the documented trading pattern is motivated by insider information or preferential treatment in the placement of underpriced bonds.

The fact that affiliated mutual fund investors are hurt by mutual funding suggests that abusing this conflict of interest is costly to the bank-controlled asset management firm for at least two reasons. First, fund underperformance reduces future fee revenues if flows of new money to mutual funds depend on past performance (see, e.g., Sirri and

Tufano, 1998). We confirm a positive and convex flow-performance relationship in our data. The relationship, however, is much weaker for funds catering to retail investors, consistent with more efficient monitoring by institutional investors (Del Guercio and

Tcak, 2002, James and Karceski, 2006, Evans and Fahlenbrach, 2012). Second, some funds charge performance fees, so that the asset management firm bears a direct cost of underperformance. Therefore, if asset management firms wish to minimize the cost of

“mutual funding,” they will be more likely to use retail funds and funds that do not

5 charge performance fees. We find that this is indeed the case, which supports the idea that mutual funding is the outcome of a strategic decision taken at the conglomerate level.

Finally, we ask whether the availability and use of funding from affiliated mutual funds allows banks to circumvent financial stress and ease credit rationing in crisis times.

Controlling for demand effects, we find that affiliated bond purchases are associated with higher (less negative) credit growth in the 2008:Q4-2012:Q2 period. While the median amount of credit from banks to non-financial firms in our sample decreased by

40% during this period, a one standard deviation increase in the amount of bonds sold to affiliated mutual funds relative to the bank’s total assets is associated with an increase in credit growth relative to otherwise similar banks of around 7%. To the best of our knowledge, we are the first to show evidence suggesting that the presence of an asset management division in financial conglomerates can help mitigate negative shocks to credit supply to the nonfinancial sector.

Taken together, our findings support the hypothesis that in the presence of lax regulation of related-party transactions, financial institutions will use affiliated mutual funds to obtain additional funding at better-than-market conditions, and will do so more when the incentives are larger and the costs are lower.

While our paper is the first to provide systematic evidence on the funding activities of bank-affiliated mutual funds, a number of recent episodes provide additional anecdotal evidence consistent with our findings. In August 2014, Portuguese lender Banco

Espirito Santo failed to comply with minimum capital requirements following severe losses from the secret purchase of debt securities issued by its troubled parent

6 company.7 Also, UBS has recently been ordered to compensate fund investors for losses after Puerto Rico’s debt restructuring. The bank failed to place the bulk of a large bond issue by Puerto Rico’s public employee pension agency in 2008 and subsequently sold it to its own closed-end funds in order to limit its own exposure.8

The remainder of the paper is organized as follows. Section 2 contains a review of the literature. Section 3 describes the data. Section 4 presents our results on funding support by affiliated mutual funds. Section 5 studies the cost of funding support for fund investors. Section 6 investigates the strategic selection of mutual funds for funding support by asset management companies. Section 7 studies the consequences of funding support for the supply of bank credit to the nonfinancial sector. Finally, section 8 concludes the paper.

2. Related Literature

Our paper contributes to the growing literature on conflicts of interest in the mutual fund industry and their influence on portfolio decisions.9 Most closely related to our work is the study of Golez and Marin (2015), who provide evidence consistent with bank-affiliated funds supporting the stock price of the umbrella institution, in particular after bad news and around seasoned equity offerings. Our analysis is complementary to theirs as we focus on bond purchases by bank-affiliated funds. However, our paper

7 See “Banco Espírito Santo secretly lent funds to controlling shareholder,” Financial Times (Online Edition), Sep 11, 2014. 8 Investment Companies in Puerto Rico were exempt from compliance with the Investment Company Act, which prohibits transactions between investment companies and affiliated firms. In September 2015, motivated by the UBS scandal, a member of US Congress proposed legislation that would extend to investment companies operating in Puerto Rico the same firewalls that govern investment activity in the United States. See “Bill Proposed to Give Regulatory Protection to Puerto Rico Mutual Fund Investors,” New York Times, Sep 25, 2015. 9 See, e.g., Lakonishok, Shleifer, Thaler, and Vishny (1991), Chevalier and Ellison (1997) and Carhart, Kaniel, Musto, and Reed (2002). 7 departs from Golez and Marin (2015) along several important dimensions. First, we focus on primary market activity, which ensures that the proceeds directly accrue to the controlling financial institution. This is different from the more indirect channel of price manipulation in the secondary market. The sheer magnitude of the funding support that we estimate in the paper together with the evidence that funding support may help banks limit the extent of credit rationing to nonfinancial sector during crises suggests that the direct channel is indeed important. Second, we exploit both time and series cross-sectional variation in financial distress and show that conglomerates respond strategically to changes in both the benefits and costs of using affiliated mutual funds.

Third, while Golez and Marin (2015) do not find that affiliated funds’ trades predict future equity returns, we find that funding support predicts lower bond returns, and therefore hurts investors. Interestingly, we find no evidence of mutual funding through seasoned equity offerings, which, together with lack of evidence of price support in the secondary bond market, suggests that purchases of parent banks’ debt and equity by affiliated mutual funds play very distinct roles within banks’ strategies. Finally, we study the consequences of funding support for bank credit, which is new to the literature.

Our paper is also closely related to other recent studies that investigate conflicts of interest in asset management firms that are part of a financial conglomerate. Ferreira,

Matos, and Pires (2018) document widespread underperformance of bank-affiliated mutual funds and link it to investments in client stocks and the subsequent chances of obtaining additional lending business. Bagattini, Fecht, and Weber (2018) document that German banks sold off risky euro-area sovereign bonds to their affiliated mutual funds during the sovereign debt crisis. Finally, Ghosh, Kale, and Panchapagesan (2014) find evidence of conflicts of interest in the trading decisions of mutual funds affiliated

8 to Indian business groups.10 In contrast, some authors show that close ties between asset managers and financial institutions can also be beneficial to fund investors. Massa and

Rehman (2008) find that bank-affiliated funds tend to overweight the stocks of companies that obtain a loan from the parent bank around the deal date. In turn, these investments deliver superior risk-adjusted performance, corroborating the existence of information flows within financial conglomerates. Kacperczyk and Schnabl (2013) show that money market funds that were part of a financial conglomerate were more likely to receive direct support from their sponsors in the week after the Lehman’s bankruptcy. In a similar vein, Franzoni and Giannetti (2018) provide evidence consistent with hedge funds affiliated with a financial conglomerate benefitting from more stable funding than stand-alone hedge funds in times of turmoil.

Finally, several papers study how profit maximization by asset management firms can lead to re-distribution of wealth across mutual fund investors. Gaspar, Massa and Matos

(2006) and Eisele, Nefedova, and Parise (2014) show how mutual fund families shift performance between funds in order to generate higher fees and “stars” that attract investor flows. Bhattacharya, Lee, and Pool (2013) find evidence for liquidity insurance within mutual fund families which appears to benefit both the investment firm and the investors of funds suffering liquidity withdrawals at the expense of the shareholders of the liquidity-supplying funds.

10 Other papers investigating how conflicts of interest stemming from the provision of different financial services by the same firm may hurt mutual fund investors include Cohen and Schmidt (2009), Ritter and Zhang (2007), and Hao and Yan (2012). 9

3. Data

To construct our dataset, we combine data drawn from different sources. The data on mutual funds’ portfolio holdings and mutual fund characteristics are provided by

CNMV (the Spanish securities markets regulator) and correspond to the entire universe of Spanish mutual funds at the quarterly frequency for the period 2000:Q1–2012:Q2.

The dataset includes fund-level information on the fee structure, investment style, number of fund investors, and net asset value (NAV). It also contains information on who is the ultimate parent of each AMG, which allows individual funds to be mapped to fund families and ultimate parent companies (banks, insurance companies, asset managers, etc.).

We consider that a given fund is affiliated to a given bank if the ultimate parent of the

AMG is a Spanish bank and is the same as the ultimate parent of the bond issuer.

Importantly, the parent companies of all financial institutions used in our paper are banks with just one exception.11 Bond purchases of a fund whose AMG ultimate parent is a foreign bank are not considered as affiliated purchases.12 Because our sample period is characterized by a significant number of bank mergers and takeovers, we hand-collect public information on these events and ensure that the bank-fund linkages are adjusted appropriately.

In addition, we obtain data on the bond issuance activity (excluding money market instruments) of Spanish financial institutions from Dealogic. We screen the database for

11 One of the banks in our sample, Bankia, was subject to an intervention by the Spanish Central Bank in May 2012. After the intervention, Bankia’s ultimate parent became the Fund for Orderly Bank Restructuring (FROB). However, this does not affect our results because there were not any bond issuances by Bankia after March 2012. Moreover, the end of the sample period is June 2012. In any case, we obtain similar results when excluding Bankia from our analysis. 12 One illustrative example is the AMG Gespastor, S.A., whose parent bank was initially Banco Pastor and then Banco Espirito Santo. Other asset management groups that are purchased by foreign banks during our sample period are Popular Gestión, S.A. (Allianz) and Bankoa Gestión (C.R. Pyrenees Gascogne). 10

EUR-denominated fixed-income securities issued between 2000:Q1 – 2012:Q2, where the issuer is a financial institution and both the issuer and the issuer parent are domiciled in Spain.13 We exclude perpetuities, non-rated issues, and securities issued by

Special Purpose Vehicles. This procedure yields a total of 1,092 Spanish bank bonds, corresponding to 1,155 distinct issues.14 For each offering, the dataset contains information on the issue date, amount issued, maturity, and a number of additional bond characteristics. For some our tests, we require data on bond yields, which we obtain from Datastream.

Bank-related information is obtained from a proprietary dataset at Bank of Spain. The dataset is at the monthly frequency, so given the quarterly frequency of our mutual fund dataset we use the information corresponding to the last month in each quarter.

We screen the universe of mutual fund portfolios for bonds identified in Dealogic and identify a total of 674 individual securities held by at least one mutual fund during our sample period. Given that our analysis aims at the identification of possible funding support in the primary market by mutual funds, we naturally discard securities that i) are not purchased by at least one fund in the quarter of issuance and/or ii) are issued by a bank that is not in direct control of at least one asset management company.15 This yields a final sample of 504 individual issues (468 securities) from 28 different financial

13 Our search encompasses the following categories: investment-grade bonds, high-yield bonds, mid-term notes, and covered bonds. 14 There are more issues than bonds because some securities are fungible and allow for the re-issuance of additional securities after the initial offering. For our analysis, we consider re-issued securities as new issues. 15 In some cases, several institutions (usually co-operative banks or savings banks) pool their asset management activities in a single umbrella organization (e.g. Ahorro Corporación). In this case, banks do not have direct and/or full control over the asset management company, so they are excluded from our analysis.

11 institutions.16 Table 1 compares our final sample of bank bonds with the universe of

Spanish bank bonds found in Dealogic across a number of bond characteristics.

[Insert Table 1 and Figure 1 about here.]

The size of the average bond issue in our sample is almost exactly 1 billion EUR, which is considerably larger than that of the average Spanish bank bond issued during the same period (682 million EUR). This is likely a reflection of funds’ preference for more liquid assets, as a larger amount outstanding facilitates a more active secondary market.

Notice that the cross-sectional standard deviation is rather large for either group (around

870 million EUR), which is driven by relatively few very large issues. The average maturity is slightly over 5 years, both for the bonds in our sample as well as for the entire universe. There are only minor differences concerning the remaining characteristics (collateralization, callability, fixed/floating coupons, and domestic governing law), with one exception. While 21% of the bonds in our sample are fungible, this property applies to only 12% of the bonds found in Dealogic. Once again, this is most likely connected to liquidity, as fungible bonds can experience an increase in the amount outstanding.

Figure 1 contrasts the bond issuance for the Dealogic universe and for our final sample over time. Given the heterogeneity of bond issuance in terms of issue size, we graph both the total amount issued as well as the number of individual issues. Issuance grew strongly in the first years of the sample period, from around 3 billion EUR in 2000 to a peak of 128 billion in 2006. The years 2007 and 2008 marked a significant decline, before the market began to recover in 2009. The yearly number of issues in fact peaked

16 These 28 banks are: Banca March, Bancaja, Banco de Sabadell, Banco Guipuzcoano, Banco Pastor, Banco Popular, Banco Santander, Bankia (Caja Madrid), Bankinter, BBVA, Kutxabank (Bilbao Bizkaia Kutxa), Caixabank (La Caixa), Caixa de Girona, Caixa Laietana, Caixa Manresa, Caixa Tarragona, Caixa Terrassa, Caja Asturias, Caja Cantabria, Caja Duero, Caja de Gipuzkoa y San Sebastián, Caja del Mediterraneo, Caja España, Caja Navarra, Caja Vital, Catalunya Caixa, Ibercaja, and Unicaja. 12 in 2010 at 179 bonds. By and large, our sample of bonds follows a similar trend over time. This is particularly true for the number of individual issues, where both lines move in parallel. For total issuance, the share of the bond universe covered by our sample declines in 2011 and 2012.

To investigate the bond purchase activities of Spanish mutual funds, we select the mutual fund sample as follows. We begin by excluding all funds that never invest into any Spanish bank bond. Moreover, we discard passive funds and all-equity funds, as well as funds with incomplete history. We also exclude the first quarter of each fund’s history to account for possible incubation bias (Evans, 2010), and furthermore drop the last quarter before a fund’s liquidation as well as the quarter in which a change in the fund’s parent company occurred. We also discard funds whose assets under management drop below 1 million EUR. Finally, we discard funds with less than four consecutive reporting quarters after the application of all these filters. The final sample consists of 1,875 funds and 62,629 fund-quarter observations.

[Insert Table 2 about here.]

Panel A of Table 2 provides an overview of how bond issuance and the scope of asset management operations vary across the 28 financial institutions covered in our sample.

The average bank accounts for 18 bond issues, ranging from a low of 1 bond to a maximum of 108 primary offerings. The average bond size (per issue) also varies considerably across banks, ranging from 89 million EUR to 1.54 billion EUR, with an average of 557 million EUR. Similarly, there is considerably heterogeneity across banks in terms of the asset management activities under their control. The average bank in our sample of 28 issuers controls roughly 25 funds and 3.9 billion EUR worth of assets.

However, some banks control more than 100 funds and have more than 30 billion EUR of assets under management, while others are very small.

Panel B contrasts the size of the asset management operations of these 28 banks with the remainder of the industry for our final sample of mutual funds. On average, they account for roughly half of all mutual funds (686 out of 1218), and nearly 80% of the total assets under management. The remaining assets are roughly equally split among other domestic banks, domestic non-banks, and foreign entities (including both banks and non-banks). The finding that the vast majority of the funds in our sample are bank- affiliated corroborates the numbers reported by Golez and Marin (2015) and Ferreira,

Matos, and Pires (2018) for Spanish all-equity funds.

[Insert Table 3 about here.]

Table 3 provides a breakdown of mutual funds’ trading activity in the 504 bank bonds under consideration. Crucially, we assume that purchases in the quarter of issuance always occur in the primary market.17 Similarly, bond sales in the quarter of maturity are classified as held until redemption, while trades in the remaining quarters are naturally attributed to the secondary market. In addition, we distinguish between four types of transactions: initiations (new positions), additions (increases of existing positions), reductions (decreases of existing positions), and terminations (full sale of existing positions).

A glance at Table 3 reveals that around 41% of all new investment positions (9,541 out of 23,213) in bank bonds are established in the quarter of issuance, consistent with a very significant role for the primary market. Similarly, around one third of all

17 While our data do not allow us to formally identify funds’ direct participation in the initial sale of a security, this assumption is likely to be an accurate reflection of reality in light of the low liquidity in the secondary market for European corporate bonds. Moreover, we show in the Internet Appendix that the main results are unchanged when restricting the sample to bonds issued in the last month of a quarter, which mechanically reduces the probability of mis-classifying secondary market trading as primary market trading.

14 terminations occur in the quarter of maturity.18 Taken together, this confirms the importance of buy-and-hold investment strategies among bond mutual funds. Moreover, there are relatively few additions and reductions, which is consistent with a low level of liquidity in the secondary market.

[Insert Table 4 about here.]

Panel A of Table 4 provides some summary statistics for a number of mutual fund characteristics. These numbers are computed across the 62,629 fund-quarter observations in our final sample. In addition, the last 4 rows of the table also provide a breakdown of the sample across different types of funds. Around 70% of all observations correspond to guaranteed funds (39%) or fixed income funds (32%).

Hybrid funds investing in both equity and fixed income assets account for 19%, while the remaining 11% are other funds (e.g. funds of funds).

Finally, Panel B of Table 4 reports the descriptive statistics of two of the variables that are intended to capture banks’ difficulties in obtaining funding: the ratio of non- performing loans and changes in credit ratings, as defined in Appendix A. Note the because of confidentiality issues, we may not report descriptive statistics on our third variable of stress at the bank level: the amount borrowed by banks from the Eurosystem.

Although our main focus is on bond purchase activity by affiliated funds, we also investigate funding support through other instruments. The CNMV data include mutual funds’ holdings of all instruments, as well as data on issuance of Short-Term-Paper.

Data on issuance activity of Preferred Shares is obtained from the Bank of Spain.

Seasoned equity offerings information is provided by the Spanish Stock Exchange

(Bolsas y Mercados Españoles). For these financial instruments, we construct the

18 Notice that there are fewer terminations than initiations because some bonds mature after the end of the sample period. 15 samples for the analysis analogously to how we construct the bond sample. The details are provided in Subsection 4.3. Panel C of Table 2 reports the number of bank deposits in our sample as well their average size. It also reports the distribution across banks of the number of issues per bank of short-term paper, preferred shares, and seasoned equity offerings, as well as the average per bank of the amount issued.

4. Empirical evidence on funding support

This section contains our main tests for the funding support activities of bank-affiliated mutual funds. While we focus on bond purchases in the primary market, we also present some results for bond market trading in the secondary market and funding through other instruments.

4.1. Funding support in the primary bond market

We begin our inquiry into the potential funding support activities of bank-affiliated mutual funds by examining their direct participation in new bond issues relative to other mutual funds. Our empirical strategy is based on the following linear panel regression

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛿𝑋,, 𝜀,,,, 1 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸,

where 𝑃,, denotes the EUR amount purchased in quarter t (the quarter of issuance) by fund f of bond issue i, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, denotes the total EUR amount issued of bond i in quarter t, and 𝐴𝐹𝐹,, is an indicator variable which takes the value of 1 if mutual fund f is affiliated with the issuer of bond i at time t (regardless of whether fund f buys the bank’s bond issue) and zero otherwise. 𝛼,𝛼 , and 𝛼 denote AMG, investment-

19 style, and time fixed effects. 𝑋,, is a vector of control variables that includes fund characteristics as well as bond characteristics that are likely to affect funds’ trading behavior. The precise definitions of these variables are provided in Appendix A. The coefficient 𝛽 captures the excess participation of bank-affiliated mutual funds in the bond issues of their parent companies relative to other mutual funds, over the entire sample period.20

As argued in the introduction, funding support can entail legal costs, reputational costs, and loss of future business. Therefore, we expect conglomerates to use this alternative source of findings more when the benefits are higher, i.e., in times of financial distress when credit becomes scarce and costly. In order to test this hypothesis, we augment the above regression model to

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛽𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆, 𝛽𝑆𝑇𝑅𝐸𝑆𝑆, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, 𝛿𝑋,, 𝜀,,, 2

Here, 𝑆𝑇𝑅𝐸𝑆𝑆 is a variable that captures difficult funding conditions for the bank issuing bond i. We use several alternative specifications for this variable. The first one relies on the bankruptcy of Lehman Brothers as an exogenous event that caused liquidity in key funding markets to evaporate. In particular, we define 𝑆𝑇𝑅𝐸𝑆𝑆, as a dummy variable that takes a value of one starting in 2008:Q3 until the end of our sample period, and zero before that date. Note that this post-Lehman dummy covers the

19 In a few cases, several fund families are affiliated with the same ultimate owner, and they are then pooled under the umbrella of one asset management group. We obtain equivalent results throughout the entire paper when instead using fixed effects at the fund family level or even the individual fund level. 20 Our analysis is conducted at the issue level, so the number of observations in a given quarter is equal to the number of active funds times the number of bond issues.

17 period of the European sovereign debt crisis, too.21 Alternatively, we define

𝑆𝑇𝑅𝐸𝑆𝑆 at the bank-quarter level as the outstanding amount borrowed by the bank from the Eurosystem in a given quarter over total assets in that quarter. Third, we define the stress variable as the ratio of gross non-performing loans over total loans. When we use the post-Lehman dummy as a measure of stress, the coefficient on the interaction term in equation (1), 𝛽, can be interpreted as a difference-in-differences estimator where the treatment is the financial crisis, the treatment group consists of all funds f that are affiliated with issuer i and the control group is comprised of all other funds in our sample. Note that, unlike in most diff-in-diff settings, both treatment and control groups vary across different observations, i.e. bond issues.

[Insert Table 5 about here.]

Table 5 contains the coefficient estimates associated with the above regressions, where all standard errors are clustered at the AMG level. Column (1) details the results obtained from the estimation of equation (1). The results suggest that, other things equal, bank-affiliated mutual funds purchase more of their parents’ debt in the primary market than non-affiliated funds. On average, each affiliated fund purchases 0.04% of the total amount issued in excess of the amount purchased by otherwise similar non- affiliated funds. These abnormal purchases are statistically significant at the 5% level.

While this result confirms the excess participation of affiliated funds in their parent banks’ bond issuance, we need to estimate the excess purchases of all funds in the same group for assessing the economic magnitude of funding support to parent banks. We perform this analysis in Table 6.

21 Importantly, the end of our sample period coincides with Mr. Draghi’s well-known “whatever it takes” promise, which led to a clear improvement in the euro zone’s weakest members’ borrowing costs.

Columns (2) – (6) of Table 5 correspond to the estimation of equation (2). In columns

(2) and (3), 𝑆𝑇𝑅𝐸𝑆𝑆, is defined as the post-Lehman dummy. The difference between column (2) and column (3) is that the latter reports estimation results when the sample is restricted to include only funds whose parent banks are bond issuers, whereas in column

(2) we use the full sample. In either case, we find a strong increase of funding support in crisis times. The coefficient 𝛽 on the interaction term is positive and significant at the

1% and 5% levels in columns (2) and (3), respectively. This implies that bank-affiliated mutual funds particularly engage in primary market purchases of parent debt in times of tight liquidity and restricted access to wholesale funding sources, namely, when they benefit the most from being able to use an alternative source of funding. The difference between crisis and non-crisis times is also quantitatively important. For example, using the estimated coefficients of column (2), funding support by the average fund increases from virtually zero in the pre-Lehman period to 0.09% (𝛽 𝛽) of the amount issued in the time after 2008:Q3.

In columns (4)-(6), we use different variables of financial stress at the bank level and the restricted sample used in column (3). In column (4), we use the bank’s reliance on the Eurosystem funds as a proxy for the bank’s financial stress. We find that banks that borrowed more from the Eurosystem, relative to their assets, also relied more on affiliated mutual funds for funding. In column (5) we use non-performing-loans as our proxy for financial distress and find also a statistically and economically significant association with banks’ reliance on their affiliated mutual funds. Finally, in column (6), we obtain a similar result using the change in the issuer’s average credit rating (a higher number corresponds to more rating downgrades, as explained in Appendix A).22

22 Our results are robust to other definitions of 𝑆𝑇𝑅𝐸𝑆𝑆,, all of which attempt to capture banks’ difficulties in accessing credit. In particular, we also employ the spread between the Spanish 10-year

Interestingly, other proxies for the banks’ funding structure, their asset structure, and their business model do not explain differences in the use of affiliated funds for funding support. For instance, in the Internet Appendix we show that listed credit institutions, despite their advantage in access to equity funding, do not exhibit higher propensity to use affiliated funds than non-listed credit institutions. Also, we observe no difference between banks and savings and loans institutions (“cajas”), which were typically captured by local governments and engaged in indiscriminate lending to real estate developers (Fernández-Villaverde, Garicano, and Santos, 2013). Similarly, neither banks’ exposure to the real estate sector nor holdings of Spanish sovereign bonds explain differences in funding support from affiliated mutual funds. Finally, a higher reliance on wholesale funding is not associated with either higher or lower use of affiliated funds for funding support. To interpret these results, it is important to note that our sample selection criteria leave out credit institutions that did not issue any bonds during our sample period and those that did not control an asset management division.

This could dilute differences between different types of institutions that are present in the whole population. In fact, differences in the average NPL ratios across different subsamples (e.g., banks v. cajas) are not statistically significant (unreported).

One possible concern with our results is the fact that funds’ trades are jointly determined by the characteristics of the bond issuers and the funds. Such endogenous matching could make it more likely for funds affiliated with a bank to purchase the bonds of banks that are similar to their parent bank. To alleviate this concern, in the

Internet Appendix we extend the specification in equation (2) to include the interaction

government bond yield and the German 10-year benchmark yield; the difference between the deposit rate offered by Spanish banks and the deposit rate offered by German banks; the 3-month EURIBOR-OIS spread; the bank’s liquidity ratio; and two measures of funding costs at the bank level (the bond’s yield and the issuer’s CDS spread). Results are reported in the Internet Appendix, which also contains the results of robustness checks for alternative definitions of participation in bond issues and different subsamples.

20 of bond issuer fixed effects with fund fixed effects and confirm that our results concerning the existence of funding support in times of financial distress and to riskier banks with limited access to funding are not driven by fund-bond issuer matching.

As mentioned above, in order to gauge the economic magnitude of total funding support from the asset management division to the parent bank, we aggregate bond purchases across all funds within a given AMG and repeat the analysis.23

[Insert Table 6 about here.]

The resulting coefficient estimates in Table 6 show that our conclusions also hold at the

AMG level. Moreover, they show that funding support is also economically significant.

From column (1), we deduce that, on average, bank-affiliated AMGs purchased an additional 2.85% of securities issued by their parent bank, relative to unaffiliated

AMGs. Based on 504 bond issues with an average size of 1,003 billion EUR (Table 1), this amounts to total funding support of 14.4 billion EUR during our sample period, or around 514 million EUR per bank.

Moreover, in line with the fund-level analysis, the results in columns (2) and (3) show that most of the funding support was provided in times of financial distress. On average, affiliated AMGs purchased an additional 7.0% (using the post-Lehman dummy) of each bond issue. Based on 226 bond issues with an average size of 750 million EUR, this amounts to total funding support of 11.9 billion EUR during the post-Lehman period

(i.e., 83% of total funding support during the whole sample period).

While parent banks benefit from the funding support of their affiliated funds through the purchase of debt, the potential benefits are larger for subordinated debt than for senior

23 For this purpose, we construct aggregate fund characteristics as AUM-weighted averages across all funds in the same AMG. 21 debt, given the higher cost of funding with subordinated debt, particularly in times of stress as subordinated debt is the one used to absorb losses first. To explore this possibility, in Table 7, we re-estimate equations (1) and (2) separately for senior and subordinated bond purchases. Estimated coefficients in columns (2) and (3) confirm our expectation that other things equal, affiliated funds buy more of the subordinated debt than of the senior debt issued by their parent banks in the primary market. Importantly, as shown in column (6), in the post-Lehman period, the difference in participation becomes 0.09%, which is statistically significant at the 1% level. This result further suggests that banks strategically resort to affiliated funds’ support in the times and circumstances in which it is most valuable to them.

[Insert Table 7 about here.]

4.2. Price support in the secondary bond market

Besides funding support in the primary market, banks can also encourage affiliated funds to trade in the secondary market in order to reduce yields (and thus financing costs) around the time of new issues. However, this constitutes only indirect support, as the proceeds accrue to the seller, not the issuer. Moreover, there is no guarantee that outright purchases will be able to affect yields, partly because European corporate bonds markets tend to be very opaque and illiquid.24 In any case, the reduced attractiveness of secondary market purchases is ultimately an empirical question. We address this question by studying changes in bond holdings between any two consecutive quarters and exclude from the sample bond-quarter observations

24 According to a recent survey by the Association for Financial Markets in Europe, 63.8% of the corporate bonds and 82.2% of the covered bonds investigated traded less than 20 times per month in the period from July 2010 to June 2011. Generally, corporate bond markets tend to be rather illiquid because standardization is low (Oehmke and Zawadowski, 2017), and many bond investors such as pension funds and insurance companies tend to follow buy-and-hold strategies (Biais et al., 2006). Moreover, these effects are amplified by the fact that there is no mandatory transaction reporting in Europe, unlike in the United States.

22 corresponding to the quarter in which the bond is issued and the quarter in which the bond matures. For simplicity, we aggregate funds’ trading activity in a given quarter across all outstanding bonds from the same issuer.25

Adapting our previous methodology, we estimate

∆𝐻,, 𝛼𝐴𝑀𝐺 𝛼𝑆 𝛼𝑡 𝛽𝐴𝐹𝐹,, 𝛽𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆, 𝛽𝑆𝑇𝑅𝐸𝑆𝑆, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸𝑖,𝑡

𝛿𝑋,, 𝜀,,, 3

where ∆𝐻,, denotes the quarterly change in fund f’s holdings of bonds from issuer i.

While we employ the same fund-level characteristics as before, 𝑋,, does not include bond-level characteristics due to our aggregation at the issuer level. However, we additionally control for the log of the total amount outstanding for issuer i at time t.

[Insert Table 8 about here.]

Table 8 contains the regression results, where column (1) corresponds to the regression specification without the interaction term 𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆, and columns (2) and (3) include the interaction term with the post-Lehman dummy for the unrestricted and restricted samples, respectively. In columns (4), (5), and (6), we use borrowing from the

Eurosystem, the ratio of non-performing loans as our measures of stress, and changes in the issuer’s credit rating. In all cases, we find no consistent evidence for price support in the secondary market, either in normal times or times of financial distress. Overall, the results corroborate our prior that funding support is most likely to occur in the primary market.

25 The results are qualitatively similar if we conduct the analysis at the bond level.

4.3 Other funding sources

So far, we have considered funding support only in the form of bond purchases. As explained above, this is a particularly valuable support channel because it constitutes long-term funding that was difficult to obtain for Spanish banks during much of the sovereign debt crisis. However, funding from affiliated mutual funds can also take place through the purchase of other instruments. In this subsection, we investigate this possibility by looking at funds’ investments into Spanish banks’ short-term paper (so- called “Pagarés,” henceforth STP), bank deposits, preferred shares (“preferentes”), and seasoned equity issues.26 For each of these instruments, we construct the corresponding sample analogously to the construction of the sample for studying bond purchases.

Thus, the sample for each type of asset consists of those funds that invest into that type of asset at any time during the whole sample period.27 To study funding support through purchases of banks’ STP, we screen mutual funds’ portfolios for STP issued by Spanish banks with affiliated mutual funds. We find a total of 14,230 securities issued by 33 banks. We then discard all funds that never hold any of these securities during our sample period, which leaves us with 1,593 funds and 56,667 fund-quarter observations.

We estimate the regression equation (2) replacing the dependent variable with mutual fund purchases of STP normalized by the amount issued. Given the lack of security- specific information, we conduct the analysis at the issuer level, i.e. we sum across all purchases of the same issuer in the same quarter. Accordingly, the set of control

26 We obtain data on the amount outstanding for each STP from CNMV. However, data on interest rates and other security characteristics is unfortunately not available. 27 For comparability, we repeat the same analysis in the Internet Appendix using the sample of mutual funds constructed in Section 4.1 and screen the portfolios for each type of asset issued by our 28 baseline sample banks. The results are fully consistent with those obtained in this subsection.

24 variables is identical to that used in Section 4.1, including AMG, fund investment style, and quarter fixed effects.

[Insert Table 9 about here.]

The estimation results are contained in Table 9, where we also include again the results for bond purchases in column (1). Column (2) of Table 9 corresponds to purchases of

STP using the post-Lehman dummy as our stress variable. The estimated coefficients on both the affiliated dummy and its interaction with the stress variable are positive and significant at the 1% level, consistent with affiliated funds being used for funding support also through purchases of STP. Before the crisis, a bank-affiliated mutual fund purchases an additional 0.14% of its parent bank’s short-term debt issuance, while excess purchases increase an additional 0.13% during the crisis.

Next, we study funds’ investments in term deposits. Although in all previous tests, we have normalized debt purchases by the amount issued, issuance cannot be defined for bank deposits in a way that is unambiguously analogous to bond issuance. Therefore, we normalize deposits by funds’ assets under management, instead. Unfortunately, information on deposits is only available after the introduction of a slightly extended reporting template in the last quarter of 2008. Accordingly, we are forced to restrict this analysis to the crisis period. The dataset contains information on the amount, the rate, and the maturity of each deposit. As we do not observe the starting date but just the quarter in which the deposit was made, we simply compute the maturity in multiples of quarters. The sample consists of all funds that deposit money in any bank with affiliated mutual funds over the sample period. The screening of the portfolios for deposits yields a total of 13,057 deposits made in 33 of institutions. We drop all funds for which we do not observe any term deposit, resulting in 801 funds and 37,326 fund-quarters.

Given that our data on deposits starts at the end of 2008, we effectively estimate equation (1), but define the dependent variable as the balance of new deposits plus increases in the balance of deposits in that quarter relative to the fund’s AUM. We employ the same fund-specific control variables as in the other specifications.

Column (3) of Table 9 reveals a positive and statistically significant coefficient estimate on 𝐴𝐹𝐹, indicating that bank-affiliated mutual funds indeed tend to favor their parent bank when investing in time deposits. In particular, each affiliated fund on average invests an excess 0.58% of its assets per quarter in deposits with its parent bank relative to non-affiliated funds during the crisis, which is statistically significant at the 1% level.28 Given that the average AMG in the regression sample manages assets worth 4.1 billion EUR per quarter, we estimate that total funding support to affiliated funds through new investment in deposits amounted to 10.1 billion EUR in the period from

2008:Q4 to 2012:Q2, or 0.4 billion EUR per bank.29

In Column (4), we investigate funding support through purchases of preferred shares.

To construct the sample, we screen mutual fund portfolios for preferred shares issued by one of our sample institutions and find a total of 16 ISIN codes referred to preferred shares from 7 banks. We then discard all funds that never hold any of these securities during their lifetime, which leaves us with 164 funds and 1,538 fund-quarter observations. In contrast to the findings for other instruments, we find no empirical

28 One potential concern is that most mutual funds in our sample (70% of all observations) only hold deposits in a single bank. The decision not to shop for the best deposits across banks could be associated with both poor asset management and bank-controlled asset management firms, but not necessarily a conscious attempt to provide funding support. However, our conclusions do not change in we control for the number of deposits or restrict the sample to funds with deposits in more than one bank (not tabulated).

29 Note that the economic magnitude of funding support through new deposits is not directly comparable with that occurring through bond purchases because both instruments differ considerably in terms of maturity. The average maturity of deposits is around seven months, whereas the one of bonds is more than five years.

26 evidence of funding support through purchases of preferred shares either before or during the crisis.

To interpret this result, it is important to note that Spanish banks sold preferred shares worth 22 billion EUR directly to retail investors.30 As explained by Santos (2017), in

Spain preferred shares were historically purchased by institutional investors. However, when the crisis hit, institutional investors fled the market and banks started marketing them to retail investors since preferred shares computed as regulatory capital and allowed banks to meet regulatory capital requirements. By selling preferred shares directly to retail investors instead of indirectly through mutual funds, Spanish banks avoided the costs of hurting their asset management business. The reason why the direct sale of preferred shares was easy for banks is that they were marketed as safe products, similar to long-term deposits, despite their true risk and their lack of a liquid secondary market (Conac, 2018; Zunzunegui, 2014). Such direct sale of bank debt to retail investors during the crisis was not specific to Spain and prompted regulators in Europe to step up their monitoring of bank debt sales.31

Unlike preferred shares, bonds are well understood by investors, which makes it difficult for banks to mischaracterize their nature. Moreover, in Spain there is very little tradition of retail investors owning non-deposit debt instruments in their portfolios.

According to the Spanish Survey of Household Finances (EFF), the fraction of Spanish households owning any type of corporate or government bonds, bills, or notes, has not exceeded 2% in any of the years 2002, 2005, 2008, and 2011, while 10.4% of Spanish households owned shares of listed firms. These figures are in striking contrast with those from the Eurosytem’s Household Finance and Consumption Network (HFCN)

30 “Estudio sobre Participaciones Preferentes,” Defensor del Pueblo, 2013. 31 See “Europe banks face clampdown on debt sales,” Financial Times (Online Edition), May 19, 2014. 27 survey for the euro area (7.3% of households held bonds, bills, or notes in 2008, and

8.2% of them owned shares of listed firms).

The comparison with Italy, another European periphery country in which banks also experienced funding difficulties, is particularly interesting. Italian banks managed to raise 31 billion EUR during the second and third quarter of 2008 by placing bonds directly to retail investors (Deloitte, 2017). According to the HFCN survey, 14.6% of

Italian households owned fixed income assets and only 4.6% owned stocks of listed firms in 2008.32 Regardless of the reasons behind such differences in household investment patterns across countries, it seems plausible that Spanish banks relied on affiliated mutual fund purchases of bank bonds because direct sales to retail investors were unlikely to succeed.

Finally, we repeat the analysis with equity issues. In this case, the sample consists of 61 seasoned equity offerings (SEOs) over the sample period 2000Q1 – 2012Q2 conducted by 9 banks.33 We discard all funds that never participate in those SEOs during their lifetime, which leaves us with 1,160 funds and 35,092 fund-quarter observations. The results are shown in Column 5 of Table 9 and indicate that affiliated funds are not more willing to participate in the SEO of their parent bank than in the SEO of any other bank.

The fact that, according to the SNL database, around 60% of the shares of Spanish banks were held by non-institutional investors might reflect that, as it happens with preferred shares, banks can place their equity directly to retail investors. In sum, bonds, and not equity, is the preferred vehicle for funding support from mutual funds to controlling banks.

32 Grasso et al. (2010) further report that bank bonds in Italy accounted for 10% of households’ portfolios in 2009, well above bank bond ownership by households in Spain, Germany, France, and the UK. 33 The analysis is implemented on SEOs and so, we discard initial public offerings but similar results are obtained when they are included in our analysis. Given that the analysis is implemented on a quarterly basis we aggregate all SEOs of each bank in a given quarter.

5. Cost for mutual fund investors

The results in the previous section suggest that bank-affiliated funds tend to buy debt of the parent institution in times of financial turmoil characterized by a deteriorated access to standard funding sources such as the interbank market. However, whether funding support is costly to affiliated mutual fund investors is an empirical question. For instance, it is possible that this excess investment is the result of insider information concerning the true financial health of the parent company, and attempts to exploit low market valuations. Massa and Rehman (2008) provide evidence of such information flows for U.S. financial conglomerates and show that the associated trades generate positive abnormal performance for fund investors. On the other hand, Golez and Marin

(2015) find no association between affiliated mutual fund trades in parent banks’ stocks and subsequent stock returns. Ferreira, Matos and Pires (2018) find that purchases of the banks’ clients’ stock actually hamper mutual fund performance. In this section, we investigate whether funding support via purchases of parent banks’ debt hurts or benefits affiliated mutual fund investors.

5.1. Cost of bond purchases

We start by examining bond purchases. First, we take the point of view of the issuer and check whether the participation of bank-affiliated mutual funds in the primary market is associated with bond mispricing in the primary market. If affiliated funds have the ability to detect if a bond issued by its parent bank is mispriced, they will purchase underpriced bonds in the primary market and shun overpriced bonds. However, as private information disseminates, prices should converge to fundamental values.

Therefore, we would expect to observe positive short-term risk-adjusted returns among those bonds that affiliated funds choose to purchase in the primary market and zero or

29 negative risk-adjusted returns among bonds avoided by affiliated funds. On the other hand, if affiliated funds have no informational advantage, we would expect no return differences between bonds purchased and comparable bonds avoided by affiliated funds. Finally, if purchases of affiliated funds contribute to inflating bond prices at issuance, then we would expect to observe negative short-term risk-adjusted returns among those that affiliated funds choose to purchase in the primary market. Therefore, this approach enables us to quantify the potential benefit for parent banks due to the funding support of their affiliated funds.

Second, we take the point of view of mutual fund investors and test whether purchases of parent bank debt in the primary market outperform simultaneous trades in other bank bonds with similar characteristics against the alternative of “pure” funding support in the case of zero or even negative abnormal performance. This approach allows us to assess the benefit or cost to mutual fund investors due to purchases of parent-bank bonds.

In order to assess bond performance, we need to compute abnormal returns. This is a challenging task because sector indices at the European level constitute poor benchmarks due to the high degree of financial fragmentation during the most intense crisis episodes. Accordingly, we are forced to construct our own benchmark for Spanish bank bonds. To this end, we collect daily bond returns from Datastream and the history of security ratings from the three major ratings agencies (S&P, Moody’s and Fitch).

Then, we compute the abnormal return for bond i on day t, denoted 𝐴𝑅,, as the bond’s total return minus the average total return of all Spanish bank bonds in the same rating category R and maturity category M.34 In order to mitigate the effects of outliers, we

34 We set M = {<5y, 5-10y, >10y} and R = {AAA, AA, A, BBB}. For securities with multiple ratings, we use the best score across the three major rating agencies at given point in time. Total returns are directly provided by Datastream.

30 winsorize bond returns at the 1st and the 99th percentile. Moreover, we require the benchmark portfolios used to compute abnormal returns to consist of at least 5 securities in a given quarter. Due to the low liquidity in the European corporate bond market, we are only able to obtain abnormal returns for 187 of our 504 bond issues. Based on these data, we compute the return on the day of the issuance (1-day abnormal return), as well as 1-week and 1-month cumulative abnormal returns.

We then regress these returns on a constant and an indicator variable,

𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸, that is equal to one if any of the funds affiliated with the issuer of bond i purchased the bond in the primary market, and zero otherwise. We also include issuer fixed effects (𝜇 in the regression to control for any time-invariant issuer characteristic that may affect bond returns.

𝐴𝑅,, 𝜇 𝜇𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸, 𝜁,,. 4

The results for different time horizons are tabulated in columns (1), (3), and (5) of Table

10, where standard errors are clustered at the AMG level. The estimated coefficient on

𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸 is negative and statistically significant at all three horizons.

Therefore, bond issuances where affiliated funds participate, not only fail to outperform those in which affiliated funds choose not to participate, but underperform by an amount that ranges from 29 basis points (on the day of issuance) to 44 basis points

(cumulative over one month). This evidence of bond overpricing at issuance provides further credence to the idea that affiliated fund purchases are aimed at providing funding to the parent bank at better-than-market conditions.

[Insert Table 10 about here.]

We then augment the regression with 𝑆𝑇𝑅𝐸𝑆𝑆, defined as the post-Lehman indicator variable, and the interaction term 𝑆𝑇𝑅𝐸𝑆𝑆 𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸 in order to see whether the observed bond overpricing is particularly concentrated in crisis times:

𝐴𝑅,, 𝜇 𝜇𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸, 𝜇𝑆𝑇𝑅𝐸𝑆𝑆

𝜇𝑆𝑇𝑅𝐸𝑆𝑆 𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸, 𝜁,,. 5

At one-day and one-week horizons, we find that most of the return differential between bond issuances with and without affiliated fund participation indeed realizes during times of financial stress. While the coefficient on 𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸 diminishes, the coefficient on the interaction term is large and statistically significant in two out of three cases. Given the reduced sample, the lack of significance at the 1-month horizon may be due to a lack of power. Based on the regression coefficients, Table 10 also reports the average abnormal returns of bonds purchased by affiliated funds in crisis times (𝜇

𝜇 𝜇. These estimated abnormal returns are always negative and statistically significant. In sum, we conclude that the data suggest that bank bonds are overpriced in the primary market when affiliated mutual funds participate at the issuance stage.

In order to study the relative performance of parent debt purchases at the fund level, we proceed as follows. We begin by screening all trades of bank-affiliated mutual funds for purchases of parent debt in the primary market. We then retain only those observations for which we additionally observe at least one purchase of a non-affiliated bank bond in the same quarter by the same fund. In order to increase the number of observations, we allow the simultaneous purchases of non-parent debt to also take place in the secondary markets.35 We then compute the abnormal performance of two value-weighted portfolios, one containing purchases of affiliated bonds and the other one containing the

35 We obtain qualitatively very similar results when considering only primary market purchases.

32 contemporaneous purchases of non-affiliated bank bonds. We denote these abnormal

portfolio returns for fund f affiliated to bank b in quarter t by 𝐴𝑅,, and 𝐴𝑅,, , respectively. As in the previous subsection, abnormal returns are computed relative to the universe of Spanish bank bonds in the same rating and maturity category. We then regress the differential abnormal performance of these two portfolios on a constant and the post-Lehman dummy:

𝐴𝑅,, 𝐴𝑅,, 𝜇𝑏 𝜇1𝑆𝑇𝑅𝐸𝑆𝑆 𝜀,,. (6)

In order to control for time-invariant unobserved heterogeneity, we additionally include

AMG fixed effects and also cluster standard errors at the AMG level. The associated coefficient estimates are displayed in Table 11. In columns (1) and (2), we use the entire sample of bonds but column (2) excludes purchases by guaranteed funds given that investors in these funds are guaranteed a minimum return by the asset management company provided they hold their shares until a specific date.

[Insert Table 11 about here.]

The estimate of µ is negative and highly significant in columns (1) and (2), which is direct evidence that mutual funding is costly to mutual fund investors. The negative impact of parent-debt purchases is also economically significant. Our estimates imply a

1-quarter ahead abnormal return of around -92 bps during the crisis. We would expect that placing lower quality bonds makes funding support more valuable, and therefore leads to more overpricing. To test this hypothesis, in column (3) we exclude from the sample purchases of AAA-rated securities, and find that the 1-quarter ahead abnormal return reduces to -137 bps in times of crisis. Statistical significance, however, declines to the 10% level, which could be due to the reduction in sample size.

In sum, we find no evidence that primary market purchases in times of financial stress are motivated by valuable insider information about the true health status of the parent bank. Instead, we find that these investments earn negative abnormal returns at the expense of fund investors.

5.2. Cost of deposits

Investing in term deposits with the parent bank could be justified from the point of view of mutual fund investors if affiliated funds were favored by the bank and earned higher interest rates on their investment. In this case, it would be difficult to claim that the observed patterns are necessarily funding support.

We first take again the perspective of the investor and ask whether mutual funds that deposit cash with their parent bank earn higher returns than on deposits with other banks. In order to answer this question, we compute the weighted average return on new deposits with the parent bank minus the weighted average return on new deposits in other banks, for each fund and quarter with new deposits of both types. We then compute the mean difference across all fund-quarter observations with both types of deposits. The result in column (1) of Table 12 suggest that on average funds earn about

24 basis points less on new deposits with their parent bank than with other banks, and this return spread is statistically significant at the 1% level (based on standard errors clustered at the AMG level). In column (2), we repeat the same analysis after excluding guaranteed funds, since investors in such funds do not necessarily bear the opportunity cost of lower rates of return on parent-bank deposits. The results are almost identical.

[Insert Table 12 about here.]

One potential concern is the possibility that these findings are driven by differences in bank risk. More specifically, our results could reflect the fact that the risk of non-

34 affiliated bank deposits is higher than that of parent banks. To address this concern, we re-run our analysis after adjusting deposit rates for bank risk by subtracting the average deposit rate within the same rating and maturity category. We distinguish seven bank rating categories (i.e. AAA, AA, A, BBB, BB, B, and CCC) and two maturity buckets

(maturities below and above six months). The results are given in columns (3) and (4).

We find that funds earn a risk-adjusted return on parent-bank deposits that is 6 basis points less than that earned on non-parent bank deposits, and this difference is statistically significant at the 10% level. When we exclude guaranteed funds (column

(4)), the spread is equal to –7 basis points. While the risk-adjustment has indeed led to a narrower spread, we still continue to find that bank-affiliated funds forego some return when depositing cash with the parent bank compared to their other bank deposits.

Finally, in columns (5) and (6) we take the perspective of the banks and ask whether they compensate affiliated funds with higher returns than those paid to non-affiliated funds. The advantage of this approach is that it enables us to quantify the cost (or benefit) to the bank of having its affiliated funds favor their parent over its competitors.

It also allows us to control for time-invariant bank characteristics that influence returns on deposits. Analogously to columns (1)-(4), we compute for each bank and quarter in which the bank receives deposits from both affiliated and non-affiliated funds, the weighted average return paid to affiliated funds minus the weighted average return paid on deposits to non-affiliated funds. The average return spread is -16 basis points when considering all funds, and –24 basis points when guaranteed funds are excluded. This negative spread constitutes a direct transfer from mutual fund investors to the parent institution, as the bank precisely saves the interest rates forgone by the investors.

6. Strategic selection of mutual funds for funding support

The fact that mutual funds overpay for their parent banks’ bonds and earn lower deposit rates implies that funding support is costly for affiliated mutual fund investors, but it also means that funding support carries an immediate benefit for the bank. A different question is whether and how the asset management firm is affected. There are several channels through which funding support to the parent bank could harm an asset management firm. First, the mutual fund literature has consistently shown that investors chase past performance (e.g., Sirri and Tufano, 1998). Therefore, for the majority of funds that charge a management fee as a fraction of assets under management, funding support can hurt future fee revenues indirectly through lower asset growth. Second, some mutual funds charge a performance fee which is calculated as a fraction of

(positive) portfolio returns. If funding support leads to lower fund returns, it also has a direct negative impact on fee revenues for the management firm. Third, violation of regulation of related-party transactions could lead to fines and, perhaps more importantly, a loss of reputation for the asset management firm, which may harm its business in the long run. Since such fines are rarely observed in Spain, in this section, we focus on the other two channels.

We start by studying how Spanish bond fund investors respond to past fund performance. The literature suggests that funds that cater to retail investors are subject to less monitoring and screening (Del Guercio and Tkac, 2002, Evans and Fahlenbrach,

2012). If that is true in the Spanish market, we would expect flows to exhibit more sensitivity to fund performance when they are targeted to institutional investors.

To explore this possibility, we define a dummy for retail funds, RETAIL, that equals 1 for funds whose average AUM per shareholder is below the 75th percentile (50,000

EUR, approximately) in the cross-section of funds for every quarter.36 To measure fund performance, we use the fund’s return in excess of the 3-month Spanish Treasury bill.

, Fund growth due to inflows of new money is computed as , where 𝐹𝐿𝑂𝑊, , denotes the flow in euros to fund f in quarter t and is provided by the supervisor.

Therefore, we estimate the regression equation:

𝐹𝐿𝑂𝑊, 𝜃 𝜃 𝜃 𝜃𝑃𝐸𝑅𝐹, 𝜃𝑃𝐸𝑅𝐹, 𝑅𝐸𝑇𝐴𝐼𝐿 𝐴𝑈𝑀,

𝜃𝑅𝐸𝑇𝐴𝐼𝐿 𝜃𝑋, 𝜉,. 7

where 𝑃𝐸𝑅𝐹, denotes lagged fund performance and 𝑋, is a vector of lagged control variables containing fund return volatility, fund flow, portfolio turnover, fund age, fund fees, (log) AUM, and (log) AMG’s AUM. We also include investment style

(𝜃), quarter (𝜃), and AMG (𝜃) fixed effects.

[Insert Table 13 about here.]

Column (1) of Table 13 contains the coefficient estimates for a model specification that includes RETAIL but not its interaction with lagged performance. Standard errors are clustered at the AMG level and the dependent variable, relative flows, is expressed in percentage points. Consistent with the literature on mutual fund flows, the estimated coefficient on lagged performance is positive and highly significant. This association is also economically significant: A 1% increase in the fund’s excess return is followed by an increase in fund assets due to new money of 38bps. Therefore, asset management companies have incentives to mitigate the negative impact of funding their parent bank.

36 We obtain qualitatively similar results if we increase the cut-off to the 90th percentile, or directly use the average AUM per shareholder (obviously yielding coefficients with the opposite sign).

Although retail funds receive less flows than funds with similar performance, the difference is not statistically significant.

In column (2) we estimate the full specification (7) and find that the coefficient on the interaction term, 𝜃, is negative and significant at the 1% level, which confirms our prior that retail investors are less sensitive to fund performance. Moreover, the economic magnitude of the difference in flow-performance sensitivity between retail and institutional funds is considerable: Flows to institutional funds are twice as sensitive to performance as flows to retail funds (𝜃 0.5476, 𝜃 𝜃 0.247).

Starting with Sirri and Tufano (1998), several authors have documented a convex flow- performance relationship in US equity funds. Ferreira et al. (2012) study a large sample of equity mutual funds domiciled in 28 countries and find a similarly convex relationship in most countries, including Spain. In contrast, Goldstein, Jiang and Ng

(2017) document a concave flow-performance relationship for US corporate bond funds. To allow for the possibility of a non-linear relationship between mutual fund flows and recent fund performance in our sample, we follow Sirri and Tufano (1998) and rank mutual funds every period within their investment objectives. Accordingly, we compute the variable RANK which is normalized by the number of funds in each objective and period, so it takes values between 0 and 1. We define the variable

LOWPERF as Min(RANK,0.2). MIDPERF is defined as Min(0.6, RANK-LOWPERF).

And finally, HIGHPERF is defined as RANK – (LOWPERF + MIDPERF). These three performance measures are rescaled (i.e., multiplied by 100) to be consistent with the scale of the dependent variable. We then estimate the resulting regression equation (7) using the three new measures of performance in place of the fund’s return in excess of the 3-month Spanish Treasury bill.

The estimated coefficients in column (3) suggest that the flow-performance relationship is also convex for Spanish bond funds. Although the slope in the middle performance range is slightly lower than the slope in the bottom range, new money flows disproportionately to funds in the top quintile of the performance distribution. In column (4), the performance variables are interacted with the retail dummy. The estimated regression coefficients suggest that the flow-performance relationship is clearly convex for institutional funds: the slope of the flow-performance relationship increases with performance. However, for retail funds the slope is lower in the medium- performance range than in the other two regions, and the difference in slope between the high and low performance ranges is not statistically significant. In other words, the flow-performance relationship for retail funds is not only flatter than for retail funds but also more linear. Consequently, managers of retail funds are rewarded less for performance, particularly for outstanding performance, than managers with institutional clients.

Based on these findings and our reasoning above, we hypothesize that funding support is more prevalent in retail funds and funds that do not charge a performance fee. To test this hypothesis, we augment equation (1) and estimate

𝑃,, 𝛼𝐴𝑀𝐺 𝛼𝑆 𝛼𝑡 𝛽𝐴𝐹𝐹,, 𝛽𝐹𝑈𝑁𝐷_𝐶𝐻𝐴𝑅𝐴𝐶𝑇𝐸𝑅𝐼𝑆𝑇𝐼𝐶, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, 𝛽𝐴𝐹𝐹,, 𝐹𝑈𝑁𝐷_𝐶𝐻𝐴𝑅𝐴𝐶𝑇𝐸𝑅𝐼𝑆𝑇𝐼𝐶, 𝛿𝑋,, 𝜀,,, 8

where 𝐹𝑈𝑁𝐷_𝐶𝐻𝐴𝑅𝐴𝐶𝑇𝐸𝑅𝐼𝑆𝑇𝐼𝐶 is either the RETAIL dummy or PERF_FEES, an indicator variable for the presence of a performance fee. Table 14 contains the regression results for both specifications. The coefficient on AFF in column (1) is positive and significant, which suggests that institutional funds are also used to provide funding support. However, the coefficient on the interaction term AFF RETAIL is

39 positive and statistically significant, which provides support to our hypothesis. The economic magnitude of the difference is also significant. While affiliated institutional funds purchase 0.04% of their parent banks’ bond issues more than non-affiliated institutional funds, the excess participation of affiliated retail funds is 0.08% (𝛽 𝛽) of the amount issued, relative non-affiliated retail funds.

[Insert Table 14 about here.]

In column (2), we find supporting evidence for our hypothesis that funds that charge a performance fee provide less funding support to their parent banks. Around 10% of the funds in our sample charge performance fees. Such funds have an excess participation in their parent banks’ bond issues of a statistically insignificant 0.004% (𝛽 𝛽), as opposed to a statistically significant 0.04% for funds without a performance fee.

In sum, the results of this section complement the previous findings by confirming that banks not only use funding support from their affiliated funds when the benefits of doing so are higher, but also strategically allocate the burden of support to funds for which the impact of poor performance is lower.

7. Consequences for credit supply

Banking crises can severely reduce credit supply, which in turn may have negative real effects on employment and investment. In Spain, this risk is potentially more severe than in other countries given the strong reliance of firms on bank credit. According to

Bentolila, Jansen, and Jiménez (2018), loans from credit institutions accounted for 86% of GDP in Spain and 62% in the European Union. If funding support from affiliated funds helps banks weather their reduced access to more conventional sources of funding

40 during crises and at a lower cost, it may also mitigate the negative shock to credit supply to the nonfinancial sector. To address this question, in this section we investigate whether firms experience less credit rationing from banks that sell more of their own debt to their affiliated funds. To isolate the effect of mutual funding on the supply of credit from potentially confounding demand effects, we follow closely the methodology introduced by Khwaja and Mian (2008). More specifically, we regress credit growth at the bank-firm level on affiliated funding support and other bank characteristics while controlling for firm fixed effects. The firm fixed effects absorb any heterogeneity across firms that is associated with both changes in their demand for credit and their borrowing from banks that use more or less mutual funding. By comparing the growth of credit from different banks to the same firm, we are effectively ruling out that changes in demand drive our results.

To construct our sample, we obtain data at the individual loan level from the Central

Credit Register of the Bank of Spain (CRR). The dataset contains information on every loan given to non-financial institutions above 6,000 euros, including the size of the credit instrument and other characteristics such as maturity, guarantees and creditworthiness. We then aggregate loan data at the firm-bank level and estimate the following credit growth equation at the bank-firm level:

log1 Credit,,log1Credit,,𝜆 𝜆AP

𝜆𝑋 𝜈,, 9

where Creditf,b,2012Q2 and Creditf,b,2008Q4 denote the total credit (in thousands of euros) committed by bank b to firm f, both drawn and undrawn, on June 2012 and December

2008, respectively.37 The variable AP (Affiliated Purchases) refers to the total purchases of bank b’s securities by its affiliated funds during the same period. We consider different definitions of AP: logarithm of the amount of bonds purchased

(Log(1+MF_Bond)); amount of bonds purchased relative to total assets

(MF_Bonds/TA); amount of bonds purchased relative to equity (MF_Bonds/Eq); and amount of bonds purchased relative to the AUM by affiliated funds (MF_Bonds/AUM).

𝑋 is a vector of bank-level controls which include the total amount of bonds issued from December 2008 to June 2012 (in logs), as well as bank characteristics as of

December 2008 (profitability, size, leverage, ratio of NPL, credit over deposits, borrowing from the Eurosystem, diversification ratio, and share of liquid assets). 𝜆 denotes the firm fixed effects.

The results are reported in Table 15. The estimated values of our coefficient of interest,

𝜆, are positive and statistically significant for all definitions of AP. These results are consistent with the hypothesis that banks’ usage of affiliated mutual funds as a source of funding limited credit rationing to the nonfinancial sector. The magnitudes of the association between affiliated purchases and credit growth are also economically significant. More specifically, a one standard deviation increase in normalized affiliated purchases (columns 1-3) is associated with an increase in (log) credit growth between

6.5% and 7%. This is a remarkable figure considering that during this period, the median (log) reduction in bank credit to non-financial firms was 40%. In column (4), where we use the logarithm of the amount of bonds purchased by affiliated funds as our measure of AP, the economic magnitude is higher: A one standard deviation in AP is associated with a 23% higher credit growth.

37 Some recent papers that study the real effects of credit supply shocks exploiting cross-sectional differences in lenders’ characteristics include Greenstone et al. (2014) and Chodorow-Reich (2014) and Bentolila et al. (2017). In fact, Bentolilla, Jiménez, and Jansen (2018) also use a 4-year time interval in their cross-sectional regression (2006-2010).

Note that the decision to purchase the parent bank’s bonds is endogenous, so we are not able to provide a causal interpretation of the strong association between affiliated purchases and credit growth. Instead, we take the results in Table 16 as suggestive that the ability to raise funds from mutual funds under the bank’s control may be an effective means of overcoming credit dry ups and mitigating its negative consequences both for credit institutions and the nonfinancial sector.

8. Conclusions

In this paper, we have unveiled a novel channel for conflicts of interest in financial conglomerates, namely the provision of funding support to the parent company. Using

Spanish data, we show that bank-affiliated funds purchase more of their parents’ bond issues in the primary market than non-affiliated funds. Further empirical evidence supports the notion that banks strategically resort to this unconventional source of funding when access to traditional funding is limited and costly. In particular, banks use their affiliated mutual funds for funding in times of crisis, when they are forced to borrow from the Eurosystem, when the fraction of non-performing loans is higher, and when their credit rating deteriorates.

Importantly, we show that funding support is associated with bond overpricing, so it appears to benefit parent banks at the expense of their affiliated mutual fund investors.

Since it is well known that poor fund performance hurts the asset management business of the bank, we ask whether banks strategically select the mutual funds that support them to minimize the impact. The evidence suggests that this is indeed the case: funding support is steered towards funds that do not charge a management fee linked to

43 performance and to retail funds whose investors are less responsive to fund performance.

In contrast to the US, related-party transactions are not banned as part of the European regulatory framework relating to conflicts of interest in asset management. Instead, regulation relies on the establishment of norms of conduct that prevent conflicts of interests or deal with them in a way that minimizes the impact on investors. Our paper’s results suggest that the European approach, at least in the way it has been implemented in Spain, may not be effective in protecting mutual fund investors.

But our results also suggest that banks are highly responsive to the costs of abusing conflicts of interest. Therefore, greater investor protection could be achieved by subjecting asset managers to higher potential liabilities associated with failure to comply with the regulation of related party transactions. This view is consistent with a very recent regulatory change in Spain. A new rule of the Spanish Securities and

Exchange Commission effectively defines every related-party transaction as a potential conflict of interest and forces asset management companies to explicitly declare that the transaction is in the best interest of investors and at a price and conditions that are better or equal to those prevailing in the market, prior to trading.38 A different way of raising the costs of abusing conflicts of interest is to facilitate investors’ monitoring of asset managers’ actions. This approach includes enhancing the disclosure of related-party transactions and their potential costs to investors as well as providing investors with tools that facilitate the comparison of fund performance with the fund’s benchmark portfolio and/or peers.

38 Technical Guide 1/2018 to related-party transactions of collective investment schemes and other transactions of collective investment scheme management companies.

Moreover, the results of our paper have two other implications to be taken under consideration by regulators. First, the fact that we find no funding support through preferred shares is suggestive that the purchase of banks’ securities by affiliated funds could be a substitute for the direct sale of securities to retail investors. The natural consequence is that mutual funds should be regulated jointly with other retail financial markets in order to avoid the risk that increasing the cost of affiliated funding support leads to more direct sales to retail investors. Second, our results suggest that the availability of funding from affiliated funds might help the banking system mitigate the impact of crises on credit supply. But the parent firm or its subsidiaries within a financial conglomerate could in principle also help the clients of the asset management division by providing funding in times of turmoil (Kacperczyk and Schnabl, 2013,

Franzoni and Giannetti, 2018). Therefore, banning within-conglomerate transactions would eliminate the ability of financial conglomerates to offset shocks internally and the associated benefits to investors and the system as a whole.

References

Bagattini, G., Fecht, F., and Weber, P. “The Fire-Sale Channels of Universal Banks in the European Sovereign Debt Crisis,” Working Paper, 2018.

Bentolila, S., Jansen, M., and Jiménez, J. “When Credit Dries Up: Job Losses in the

Great Recession.” Journal of the European Economic Association 16, 45–76, 2018.

Bhattacharya, U., Lee, J. H., and Pool, V. K., “Conflicting Family Values in Mutual

Fund Families,” Journal of Finance, 68, 173-200, 2013.

Biais, B., Declerck, F., Dow, J., Portes, R., and von Thadden, E. “European Corporate

Bond Markets: transparency, liquidity, efficiency,” CEPR Working Paper, 2006.

Cassola, N., Hortaçsu, A., and Kastl, J. “The 2007 Subprime Market Crisis Through

The Lens Of European Central Bank Auctions For Short-Term Funds,” Econometrica,

81(4), 1309-1345, 2013.

Carhart, M. M., Kaniel, R., Musto, D. K., and Reed, A. V. “Leaning for the Tape:

Evidence of Gaming Behavior in Equity Mutual Funds,” Journal of Finance, 57(2), 661-

693, 2002.

Chevalier, J. and G. Ellison, “Risk Taking by Mutual Funds as a Response to

Incentives,” Journal of Political Economy, 106, 1167-1200, 1997.

Chodorow-Reich, G. “The Employment Effects of Credit Market Disruptions: Firm- level Evidence from the 2008-09 Financial Crisis,” Quarterly Journal of Economics

129, 1-59, 2014.

Cohen, L., and Schmidt, B., “Conflicts of Interest and Mutual Fund Portfolio Choice:

Attracting Flows by Attracting 401(k) Plans,” Journal of Finance, 64, 1225-1252, 2009.

Conac, P. H.. “Subordinated Debt and Self-placement,” Directorate-General for Internal

Policies, European Parliament, 2018.

Del Guercio, D., and P. A. Tkac. “The determinants of the flow of funds of managed portfolios: mutual funds vs. pension funds,” Journal of Financial and Quantitative

Analysis 37, 523-557, 2002.

Deloitte, “Asset Management in Italy: A Snapshot in an Evolutive Context,” Report,

2017.

Eisele, A., Nefedova, T., and Parise, G., “Predation versus Cooperation in Mutual Fund

Families” Working Paper, 2013.

Evans, Richard, “Mutual Fund Incubation,” Journal of Finance, 65, 1581-1611, 2010.

Evans, R. B., and Fahlenbrach, R. “Institutional Investors and Mutual Fund

Governance: Evidence from Retail–Institutional Fund Twins,” Review of Financial

Studies, 25, 3530-3571, 2012.

Fernández-Villaverde, J., Garicano, L. and Santos, T. J. “Political Credit Cycles: The

Case of the Euro Zone,” Journal of Economic Perspectives, 27, 145-166, 2013.

Ferreira, M., Keswani, A., Miguel, A. and Ramos, S. “The Flow Performance

Relationship around the World,” Journal of Banking and Finance, 6, 1759-1780, 2012.

Ferreira, M., Matos, P., and Pires, P. “Asset Management within Commercial Banking

Groups: International Evidence,” Journal of Finance, 73, 2181-2227, 2018.

Franzoni F., and Giannetti, M. “Costs and Benefits of Financial Conglomerate

Affiliation: Evidence from Hedge Funds,” Journal of Financial Economics, forthcoming, 2018.

Garcia de Andoain, C., Heider, F., Hoerova, M., and Manganelli, S. “Lending-of-Last-

Resort is as Lending-of-Last-Resort-Does: Central Bank Liquidity Provision and

Interbank Market Functioning in the Euro Area,” Journal of Financial Intermediation,

28, 32-47, 2016.

Garcia de Andoain, C., Manganelli, S., and Hoffmann, P. “Fragmentation in the Euro

Overnight Unsecured Money Market,” Economics Letters, 125, 298–302, 2014.

Gaspar, J.M., Massa, M., and Matos, P. “Favoritism in Mutual Fund Families? Evidence on Strategic Cross-Fund Subsidization,” Journal of Finance, 61, 73-104, 2006.

Ghosh, P., Kale, J., and Panchapagesan, V. “Do Indian Business Groups owned mutual funds maximize value for their investors?” Working Paper, 2014.

Goldstein, I., Jiang, H., and Ng, D. T. “Investor Flows and Fragility in Corporate Bond

Funds.” Journal of Financial Economics, 126, 592-613, 2017.

Golez, B., and Marin, J., “Price Support by Bank-Affiliated Mutual Funds,” Journal of

Financial Economics, 115, 614-638, 2015.

Grasso, R., Linciano, N., Pierantoni, L., and Siciliano, G. “Le Obbligazioni Emesse da

Banche Italiane.” Unpublished working paper, Consob, Rome. 2011.

Greenstone, M. Mas, A., and Nguyen, H.-L. “Do Credit Market Shocks affect the Real

Economy? Quasi-Experimental Evidence from the Great Recession and 'Normal'

Economic Times” NBER Working Paper No. 20704, 2014.

Hao, Q., and Yan, X., “The Performance of Investment Bank Affiliated Mutual Funds:

Conflicts of Interest or Informational Advantage?,” Journal of Financial and

Quantitative Analysis, 47, 537-565, 2012.

James, C., & Karceski, J., “Investor Monitoring and Differences in Mutual Fund

Performance.,” Journal of Banking & Finance, 30, 2787-2808, 2006.

Kacperczyk, M., and Schnabl, P., “How Safe Are Money Market Funds?,” Quarterly

Journal of Economics, 128, 1073–1122, 2013.

Khwaja, Asim I. and Atif Mian, “Tracing the Impact of Bank Liquidity Shocks:

Evidence from an Emerging Market.” American Economic Review, 98, 1413–1442,

2008.

Lakonishok, J., Shleifer, A., Thaler, R., and Vishny, R., “Window Dressing by Pension

Fund Managers,” American Economic Review Papers and Proceedings 81, 227–231,

1991.

Massa, M., and Rehman, Z., “Information Flows within Financial Conglomerates:

Evidence from the Banks–Mutual Funds Relation,” Journal of Financial Economics, 89,

288-306, 2008.

Oehmke, M., and A. Zawadowski, “The Anatomy of the CDS Market,” Review of

Financial Studies, 30, 80-119, 2017.

Ritter, J. R., and Zhang, D., “Affiliated Mutual Funds and the Allocation of Initial

Public Offerings,” Journal of Financial Economics, 86, 337-368, 2007.

Santos, T., “El Diluvio: The Spanish Banking Crisis, 2008-2012,” Working Paper,

2017.

Sirri, E., and Tufano, P., “Costly Search and Mutual Fund Flows,” Journal of Finance,

53, 1589-1622, 1998.

Zunzunegui, F. “Mis-Selling of Preferred Shares to Spanish Retail Clients,” Journal of

International Banking Law and Regulation, 29, 174-186, 2014.

Appendix A – Definition of control variables

In the following, we define the control variables that are used throughout the paper.

A.1. Fund-level variables:

𝐿𝑂𝐺_𝐹𝑈𝑁𝐷_𝐴𝑈𝑀,: The natural logarithm of the assets under management of fund f in quarter t.

𝐿𝑂𝐺_𝐴𝑀𝐺_𝐴𝑈𝑀,: The natural logarithm of the total sum of assets under management for fund f’s asset management group in quarter t.

𝐿𝑂𝐺_𝐹𝐿𝑂𝑊,: The signed natural logarithm of the net in-/outflow of fund f in quarter t, computed as 𝑠𝑖𝑔𝑛𝐹𝐿𝑂𝑊,∗log𝐹𝐿𝑂𝑊 ,.

𝐿𝐴𝐺_𝑅𝐸𝑇,: The return of fund f in quarter 𝑡1.

LAG_PERFf,t: The return of fund f in quarter t-1 in excess of the average 3-month Spanish

Treasury bill that quarter.

LOWPERFf,t, MIDPERFf,t, and HIGHPERFf,t: These variables represent three fund performance measures: the bottom performance quintile (LOWPERF), the 2nd-4th performance quintiles (MIDPERF), and the upper performance quintile (HIGHPERF). The three variables are based on the fund’s fractional rank (RANK) which is defined as its percentile performance relative to other funds with the same investment objective in the same period, and ranges from 0 to 1. LOWPERF is defined as Min(RANK,0.2), MIDPERF is defined as Min(0.6, RANK-LOWPERF), and finally HIGHPERF is defined as RANK – (LOWPERF + MIDPERF). These three performance measures are rescaled (i.e., multiplied times 100) to be consistent with the scale of the dependent variable.

𝐹𝐸𝐸𝑆,: The total fees charged by fund f in quarter t, following Golez and Marin (2014) and Sirri and Tufano (1998). Total fees are computed as

𝐹𝐸𝐸𝑆, 𝑚𝑎𝑛𝑎𝑔𝑒𝑚𝑒𝑛𝑡, 𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒, 𝑐𝑢𝑠𝑡𝑜𝑑𝑖𝑎𝑛,

𝑢𝑝𝑓𝑟𝑜𝑛𝑡 𝑟𝑒𝑑𝑒𝑚𝑝𝑡𝑖𝑜𝑛, 𝑟𝑒𝑏𝑎𝑡𝑒,/7

PERF_FEESf,t: Dummy variable that is equal to one if the fund has performance fees.

𝑅𝐸𝑇𝐴𝐼𝐿: Dummy variable equal to one if the average AUM per fund shareholder is below the 75th percentile (corresponding to roughly 50,000 EUR) in the cross-section of funds for every quarter, and zero otherwise.

𝐿𝑂𝐺_𝐴𝐺𝐸,: The logarithm of 1 plus the fund age measured in quarters since its registration date.

𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅,: The total value of purchases (sales) during quarter t by fund f relative to the average total net assets of the fund during the last year (𝑇𝑁𝐴,). Following the methodology used in the CRSP database, we compute turnover as computed as:

min𝐵𝑢𝑦𝑠 ,, 𝑆𝑒𝑙𝑙𝑠, 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅, 𝑇𝑁𝐴,

𝐹𝑈𝑁𝐷_𝑅𝐸𝑇_𝑉𝑂𝐿𝐴,: The standard deviation of quarterly fund returns over the prior four quarters.

For our analyses at the AMG level (Section 4.1), these variables are aggregated across funds using weighted averages based on assets under management. In this section, we additionally use the following variable

𝐿𝑂𝐺_𝐴𝑀𝑇_𝑂𝑈𝑇𝑆𝑇𝐴𝑁𝐷𝐼𝑁𝐺,: The natural logarithm of the total principal amount outstanding at time t across all bonds of issuer i that are included in our sample.

A.2. Bond-level variables:

𝐿𝑂𝐺_𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸,: The natural logarithm of the total amount issued of bond i in quarter t.

𝑀𝐴𝑇𝑈𝑅𝐼𝑇𝑌,: The residual maturity of bond i at issuance in quarter t (measured in years).

𝑆𝐸𝐶𝑈𝑅𝐸𝐷: Dummy variable that takes the value of 1 if bond i is collateralized, and 0 otherwise.

𝐶𝐴𝐿𝐿𝐴𝐵𝐿𝐸: Dummy variable that takes the value of 1 if bond i issued in quarter t is callable and 0 otherwise.

𝐹𝐿𝑂𝐴𝑇𝐼𝑁𝐺: Dummy variable that takes the value of 1 if bond i has a floating rate coupon and 0 otherwise.

𝐹𝑈𝑁𝐺𝐼𝐵𝐿𝐸: Dummy variable that takes the value of 1 if bond i is fungible and 0 otherwise.

𝐷𝑂𝑀_𝐿𝐴𝑊: Dummy variable that takes the value of 1 if bond i is issued under Spanish law and 0 otherwise.

A.3: Bank level variables

EUROSYSTEMf,b,t: Outstanding amount borrowed from the Eurosystem by the parent bank of fund f in a given quarter over total assets in that quarter.

NPLf,b,t: Ratio of gross non-performing loans over total loans.

RATINGf,b,t: Change in the issuer’s average credit rating based on the three major rating agencies (S&P, Moody’s and Fitch) over the past two quarters. We first convert credit ratings to an ordinal scale according to the following schedule: 21 for the lowest rating (C) and 1 for the highest rating (AAA).

Appendix B –Regulation of related-party transactions and minimum diversification limits in Spain

Under Spanish law, asset management companies have the obligation to put in place procedures and internal control mechanisms to identify, monitor, and mitigate conflicts of interest. Accordingly, the firm must detect possible conflicts of interest. If a potential conflict of interest associated with related-party transactions is detected, an independent unit within the conglomerate must verify that the transaction is in the best interest of investors and at prices and conditions that are better than or equal to those prevailing in the market. When conflicts of interest cannot be avoided, the firm must disclose them to investors. Such decisions are ultimately subject to the supervisory authority’s ex-post scrutiny. We have found only two instances in which the supervisor fined an asset management company for failure to comply with the regulation of related-party transactions. Until February 2018, detection of potential conflicts of interest in related- party transactions was in the hands of the asset management firm itself. Since that date, the Spanish Securities and Exchanges Commission issued a new rule whereby all purchases of securities issued by related parties, including deposits, must be authorized by the independent unit prior to the transaction. This rule, in practice, defines every related-party transaction as a potential conflict of interest and forces the asset management firm to declare explicitly that the transaction is in the best interest of investors and at a price and conditions that are better than or equal to those prevailing in the market, before carrying out the trade.

The Undertakings for Collective Investment in Transferable Securities (UCITS)

Directive 2009/65/EC, in its article 52, establishes that a UCITS cannot invest more than 5% of its assets in transferable securities or money market instruments issued by the same body. (Deposits in a single body are subject to a 20% position limit).

However, the Directive allows “Member States to raise the 5% limit to a maximum of

10%,” provided that “the total value of the transferable securities and the money market instruments held by the UCITS in the issuing bodies in each of which it invests more than 5% of its assets does not exceed 40% of the value of its assets.” This is the so- called 5/10/40 rule. The Spanish Law (Law 35/2003) transposes the European Directive and raises the 5% limit to 10% as long the sum of investments in a single issuer that exceed the 5% limit does not account for more than 40% of the portfolio’s assets.

However, there are a number of exceptions to the 5/10/40 rule. Most important to our paper, the Spanish Law allows the Implementing Regulation to raise the 5% limit for specific types of securities and funds. Accordingly, the Implementing Regulation grants exemption from the 5% limit to funds whose investment objective is to replicate an equity or fixed-income index, provided that the index meets certain conditions. In this case, the limit on investments in securities of a single issuer is raised to 20% of assets under management or 35% under exceptional circumstances to be assessed by the market’s supervisor. Under exceptional circumstances, funds benchmarked to an index are also allowed to invest up to 35% of the portfolio in securities issued by a single issuer. Since such national-level regulation does not comply with the EU Directive, mutual funds that benefit from this exemption are considered non-UCITS.

Appendix C – Tables

Table 1: Bond characteristics This table compares the set of Spanish bank bonds with at least one purchase by a Spanish mutual fund in the quarter of issuance (“Sample Bonds”) with the set of Spanish bank bonds found in Dealogic (“Dealogic Universe”). Each row tabulates the group averages of the respective bond characteristic, with standard deviations in brackets whenever appropriate. The variable definitions are provided in Appendix A.

Sample Bonds Dealogic Universe

# ISSUES 504 1,155 AMOUNT (Mio. EUR) 1,003.4 681.7 (873.8) (866.6) MATURITY 5.13 5.15 (3.59) (4.16) SECURED 38% 35% CALLABLE 9% 11% FLOATING 42% 49% FUNGIBLE 21% 12% DOMESTIC LAW 84% 82%

Table 2: Descriptive statistics for funds and bond issuance activity Panel A of this table provides an overview of the issuance activity and the scope of the asset management operations across the 28 bond-issuing financial institutions (“sample banks”) with control over at least one asset management firm. All numbers are based on the cross-section of individual time-series averages. Panel B provides a breakdown of our sample of mutual funds into four different groups of affiliation: The 28 bond issuers (“sample banks”), other domestic banks, domestic non-banks, and foreign institutions (both banks and non-banks). Panel C provides an overview of the issuance activity across the banks that issue each specific instrument (number of banks in brackets) and with control over at least one asset management firm.

Panel A: Bond issuance and asset management operations across sample banks

Variable Mean Std. Dev. Min Max Number of Bonds Issued 18.0 25.0 1 108 Average Issuance (Mio. EUR) 556.7 443.4 89.0 1,538.1 Average AUM (Bln. EUR) 3.9 7.4 0.1 30.7 Average Number of Funds 25.2 26.4 4.0 112.5

Panel B: Distribution of mutual fund affiliations

Type of Fund Avg. # Funds Avg. AUM (Bln EUR) Sample banks 685.7 107.1 Other domestic banks 199.4 9.5 Domestic non-banks 176.0 8.7 Foreign (banks and non-banks) 157.6 9.9 Total 1,218.7 135.2

Panel C: Other instruments issuance across sample banks

Variable Mean Std. Dev. Min Max Deposits (33 banks) Number of Deposits 287.4 358.6 1 1331 Average Deposit Size (Mio. EUR) 3.3 2.0 0.2 8.5 STP (33 banks) Number of Issues 431.2 422.3 1 1463 Average Issuance (Mio. EUR) 31.2 19.6 3.7 80.1 Preferred Shares (7 banks) Number of Issues 2.3 1.7 1 5 Average Issuance (Mio. EUR) 300.0 165.0 75.0 507.0 Seassoned Equity Offerings (9 banks) Number of Issues 6.8 5.0 1.0 16.0 Average Issuance (Mio. EUR) 1030.0 1070.0 59.3 3300.0

Table 3: Mutual funds’ trading activity in the bond market This table provides a breakdown of mutual funds’ trading activity in our sample of 504 bank bonds. For each bond, we distinguish between trading in the quarter of issuance (Primary Mkt.), the quarter of maturity (Maturity), and the remaining quarters (Secondary Mkt.). Moreover, transactions are grouped into four types. “Initiations” (“Terminations”) refer to the establishment (termination) of a new (existing) investment position. Similarly, “Additions” (“Reductions”) refer to an increase (decrease) of an existing investment position.

Secondary Type of trade Primary Mkt. Mkt. Maturity Total

Initiations 9,541 13,672 0 23,213 Additions 0 4,068 0 4,068 Reductions 0 3,976 0 3,976 Terminations 0 11,210 5,758 16,968

Total 9,541 32,926 5,758 48,225

Table 4: Summary statistics for mutual fund and bank characteristics Panel A of this table contains summary statistics for the fund characteristics defined in Appendix A (based on the final sample of 62,629 fund-quarter observations): AUM (FUND_AUM), AMG’s AUM (AMG_AUM), quarterly returns (RET), fees (FEES), net in-/outflow relative to AUM (FLOW/AUM), indicator of retail investors’ fund (RETAIL), standard deviation of quarterly returns over the prior four quarters (FUND_RET_VOLA), purchases/sales turnover (TURNOVER), and fund age in quarters (AGE). We also provide a breakdown of funds across the four different types retained in the sample: Guaranteed funds (GUARANTEED), fixed income funds (FIXED_INCOME), hybrid funds (HYBRID), and other funds excluding index funds and equity funds (OTHER). In addition, we report a breakdown of funds depending on the typology of the AMG parent firm: Domestic bank (AMG_BANK), domestic saving bank (AMG_SAVING_BANK), other domestic non-bank (AMG_OTHER_DOM), and foreign (AMG_FOREIGN). Panel B contains descriptive statistics on the ratio of non-performing loans (NPL) and the number of notches of rating changes for downgrades (RATING) of the 29 banks to which the funds in our sample are affiliated. The units are reported in brackets next to the variable names.

Panel A: Funds' descriptive statistics Mean Median Std. Dev. P5 P95 FUND_AUM (€000,000) 112.00 31.70 307.00 4.20 528.00 AMG_AUM (€000,000) 7,740.00 2,360.00 11,600.00 122.00 43,600.00 RET (%) 0.39 0.50 2.84 -3.58 4.20 FEES (%) 1.44 1.43 0.62 0.30 2.66 FLOW/AUM (%) -0.29 -1.78 20.42 -21.13 27.52 RETAIL (%) 58.83 100.00 49.23 0 100 FUND_RET_VOLA (%) 1.68 0.99 2.09 0.06 5.78 TURNOVER (%) 19.53 11.18 32.42 0.86 75.91 AGE (# quarters) 26.52 22.25 18.32 4.22 74.24 GUARANTEED (%) 39.08 - - - - OTHER (%) 10.81 - - - - FIXED_INCOME (%) 31.55 - - - - HYBRID (%) 18.56 - - - - AMG_BANK (%) 31.27 - - - - AMG_SAVING_BANK (%) 44.34 - - - - AMG_OTHER_DOM (%) 12.86 - - - - AMG_FOREIGN (%) 11.53 - - - -

Panel B: Banks' descriptive statistics Mean Median Std. Dev. P5 P95 NPL (%) 2.46 1.06 2.48 0.37 7.12 RATING 0.17 0.00 0.61 -1.00 5.50

Table 5: Funding support in the primary market Column (1) of this table presents the coefficient estimates of the regression equation

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛿𝑋,, 𝜀,,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, where P,, denotes the EUR amount purchased in quarter t (the quarter of issuance) by fund f of bond i, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, denotes the total EUR amount issued of bond i in quarter t, and AFF,, is an indicator variable which takes the value of 1 if mutual fund f is affiliated with the issuer of bond i at time t and zero otherwise. X,, is a vector of control variables that are defined in Appendix A. Columns (2) – (6) contain the coefficient estimates for the regression

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛽𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆, 𝛿𝑋,, 𝜀,,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, where STRESS, denotes a time-varying variable capturing times of financial stress. In column (2), this variable is defined as a dummy that takes the value of one in the period after the Lehman bankruptcy (i.e. starting in 2008:Q3), and zero otherwise. In column (3), we use the same measure of financial stress but the sample of funds consists of those with affiliated banks that are bond issuers. In columns (4) – (6), the measures of financial stress are defined at bank level and are the total borrowing from the Eurosystem over total assets, the ratio of non-performing loans, and the change in the issuer’s average credit ratings over the past two quarters, respectively. All regressions include AMG, style, and year-quarter fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively. (1) (2) (3) (4) (5) (6)

AFF 0.0418** -0.0018 0.0013 0.0437** 0.0010 0.0399** (0.019) (0.002) (0.002) (0.018) (0.008) (0.019) AFF × STRESS 0.0952*** 0.0878** 0.0274** 0.0166** 0.0386* (0.034) (0.036) (0.012) (0.008) (0.023) STRESS -0.0006 -0.0043*** 0.0004 (0.001) (0.002) (0.003) LOG_FUND_AUM 0.0093*** 0.0094*** 0.0108*** 0.0115*** 0.0109*** 0.0108*** (0.002) (0.002) (0.003) (0.003) (0.003) (0.003) LOG_AMG_AUM -0.0072* -0.0029 -0.0033 -0.0090 -0.0067 -0.0112 (0.004) (0.002) (0.005) (0.006) (0.006) (0.007) LAG_RET 0.0001 0.0001 0.0001 0.0000 0.0001 0.0001 (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) FLOW/AUM 0.0095 0.0092 0.0133 0.0144 0.0130 0.0138 (0.009) (0.009) (0.021) (0.022) (0.021) (0.021) FEES 0.0003 0.0001 -0.0002 -0.0002 -0.0001 -0.0001 (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) RETAIL -0.0034 -0.0034 -0.0036 -0.0039 -0.0036 -0.0037 (0.003) (0.003) (0.005) (0.005) (0.005) (0.005) LOG_ISSUANCE -0.0084** -0.0085** -0.0113*** -0.0113*** -0.0114*** -0.0111*** (0.003) (0.003) (0.004) (0.004) (0.004) (0.004)

Table 5, continued

MATURITY -0.0001 -0.0002 -0.0007* -0.0008* -0.0007* -0.0007* (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) SECURED 0.0070 0.0072 0.0142** 0.0132** 0.0148** 0.0137** (0.005) (0.005) (0.006) (0.006) (0.007) (0.006) CALLABLE 0.0039 0.0043 -0.0020 -0.0017 -0.0022 -0.0023 (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) FLOATING 0.0029 0.0021 0.0034 0.0041 0.0042 0.0045 (0.005) (0.005) (0.007) (0.007) (0.007) (0.006) FUNGIBLE -0.0066 -0.0073 -0.0123** -0.0115** -0.0123** -0.0112** (0.004) (0.005) (0.005) (0.005) (0.005) (0.005) DOM_LAW 0.0028** 0.0024 0.0011 0.0026 0.0013 0.0019 (0.001) (0.001) (0.002) (0.003) (0.002) (0.003) AMG FE Yes Yes Yes Yes Yes Yes Style FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Observations 642,957 642,957 366,129 339,168 362,528 366,129 Adj. R-squared 0.002 0.002 0.003 0.003 0.003 0.002

Table 6. Funding support in the primary market – aggregation at AMG level

Column (1) of this table presents the coefficient estimates of the regression equation

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛿𝑋,, 𝜀,,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, where 𝑃,, denotes the EUR amount purchased in quarter t (the quarter of issuance) by all funds belonging to AMG g of bond issue i, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, denotes the total EUR amount issued of bond i in quarter t, , and 𝐴𝐹𝐹,, is an indicator variable which takes the value of 1 if AMG g is affiliated with the issuer of bond i at time t and zero otherwise. 𝑋,, is a vector of control variables aggregated at the AMG level that are defined in Appendix A. Columns (2) – (6) contain the coefficient estimates for the regression

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛽𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆, 𝛿𝑋,, 𝜀,,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, where 𝑆𝑇𝑅𝐸𝑆𝑆, denotes a time-varying variable capturing times of financial stress. In column (2), this variable is defined as a dummy that takes the value of one in the period after the Lehman bankruptcy (i.e. starting in 2008:Q3), and zero otherwise. In column (3), we use the same measure of financial stress but the sample of funds consists of those with affiliated banks that are bond issuers. In columns (4) - (6), the measures of financial stress are defined at bank level and are the total borrowing from the Eurosystem over total assets, the ratio of non-performing loans, and the change in the issuer’s average credit ratings over the past two quarters, respectively. All regressions include AMG, style, and year-quarter fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

(1) (2) (3) (4) (5) (6)

AFF 2.8514** -0.5670** -0.5330** 2.8976* -0.1308 2.7124** (1.441) (0.269) (0.226) (1.540) (0.668) (1.377) AFF × STRESS 7.5854** 7.3867** 1.7454** 1.1858* 1.7947* (3.613) (3.305) (0.827) (0.668) (1.048) STRESS -0.0031 -0.2001** -0.0951 (0.039) (0.093) (0.182) LOG_FUND_AUM -0.0504 0.0093 0.0304 0.1715 -0.0849 0.0290 (0.071) (0.037) (0.229) (0.240) (0.256) (0.234) LAG_RET 0.0230* 0.0192* 0.2089 0.2297 0.2074 0.2030 (0.013) (0.011) (0.137) (0.143) (0.133) (0.153) FLOW/AUM 0.0670 0.0486 -0.6158 -0.8233 -0.7229 -0.6220 (0.086) (0.077) (1.008) (1.055) (1.005) (0.688) FEES 0.1236 0.0852 -0.2471 -0.3151 -0.1888 -0.2448 (0.122) (0.085) (0.314) (0.357) (0.317) (0.237) RETAIL -0.3107 -0.2267 -1.1375 -0.9180 -0.3066 -1.1295 (0.417) (0.331) (0.972) (0.933) (0.792) (1.586) LOG_ISSUANCE -0.1604** -0.1603** -0.3981*** -0.4010*** -0.4019*** -0.3972* (0.076) (0.076) (0.125) (0.128) (0.126) (0.221)

Table 6, continued

MATURITY -0.0022 -0.0022 -0.0201* -0.0270* -0.0203* -0.0206 (0.007) (0.007) (0.011) (0.014) (0.011) (0.023) SECURED 0.1191 0.1182 0.4849** 0.4617** 0.5216** 0.4916* (0.088) (0.091) (0.237) (0.233) (0.250) (0.259) CALLABLE 0.0450 0.0451 -0.1215 -0.0835 -0.1190 -0.1172 (0.092) (0.092) (0.159) (0.166) (0.159) (0.116) FLOATING 0.0099 0.0095 0.0639 0.0292 0.0766 0.0683 (0.093) (0.092) (0.225) (0.231) (0.230) (0.292) FUNGIBLE -0.1215 -0.1209 -0.3965** -0.4114** -0.4118** -0.3946* (0.075) (0.079) (0.160) (0.170) (0.168) (0.221) DOM_LAW 0.0296 0.0303 0.0212 0.0355 0.0111 0.0179 (0.024) (0.025) (0.063) (0.067) (0.060) (0.064) AMG FE Yes Yes Yes Yes Yes Yes Style FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Observations 40,078 40,078 12,222 11,228 12,128 12,222 Adj. R-squared 0.018 0.040 0.076 0.088 0.089 0.077

Table 7: Funding support in the primary market through senior and subordinated bonds Columns (1) – (3) of this table present the coefficient estimates of the regression equation

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛿𝑋,, 𝜀,,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, where 𝑃,, denotes the EUR amount purchased in quarter t (the quarter of issuance) by fund f of bond i, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, denotes the total EUR amount issued of bond i in quarter t, and AFF,, is an indicator variable which takes the value of 1 if mutual fund f is affiliated with the issuer of bond i at time t and zero otherwise. 𝑋,, is a vector of control variables that are defined in Appendix A. Column (1) contains the results obtained for the whole sample of bonds whereas columns (2) and (3) refer to the subsamples of senior and subordinated bonds, respectively. Columns (3) – (6) contain the coefficient estimates for the regression

𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛽𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆, 𝛿𝑋,, 𝜀,,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, where 𝑆𝑇𝑅𝐸𝑆𝑆, is a dummy that takes the value of one in the period after the Lehman bankruptcy (i.e. starting in 2008:Q3), and zero otherwise. Column (4) contains the results for the whole sample of bonds whereas columns (5) and (6) refer to the subsamples of senior and subordinated bonds, respectively. All regressions include AMG, style, and year-quarter fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively. (1) (2) (3) (4) (5) (6) ALL SENIOR SUBORD. ALL SENIOR SUBORD. AFF 0.0418** 0.0224** 0.0601* -0.0018 -0.0022 -0.0036 (0.019) (0.009) (0.036) (0.002) (0.002) (0.006) AFF × STRESS 0.0952*** 0.0524*** 0.1440** (0.034) (0.018) (0.073) Control Variables YES YES YES YES YES YES AMG FE YES YES YES YES YES YES Fund Inv Style FE YES YES YES YES YES YES Time FE YES YES YES YES YES YES Observations 642,957 346,101 296,856 642,957 346,101 296,856 R-squared 0.002 0.002 0.002 0.002 0.002 0.002

Table 8: Funding support in the secondary market This table contains the coefficient estimates for the regression equations

∆𝐻,, 𝛼𝐴𝑀𝐺 𝛼𝑆𝑡 𝛼𝑡 𝛽𝐴𝐹𝐹,, 𝛽𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆, 𝛿𝑋,, 𝜀,,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, where ∆𝐻,, denotes the quarterly change in fund f’s holdings of bonds from issuer i, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, denotes the total EUR amount issued of bond i in quarter t, 𝐴𝐹𝐹,, is an indicator variable which takes the value of 1 if mutual fund f is affiliated with issuer i at time t and zero otherwise, and 𝑆𝑇𝑅𝐸𝑆𝑆 denotes a time-varying variable capturing times of financial stress. 𝑋,, is a vector of control variables that are defined in Appendix A. In column (1), 𝛽 is restricted to be zero. Column (2) presents the results when defining 𝑆𝑇𝑅𝐸𝑆𝑆 via a dummy that takes the value of one in the period after the Lehman bankruptcy (i.e. starting in 2008:Q3), and zero otherwise. In column (3), we use the same measure of financial stress but the sample of funds consists of those with affiliated banks that are bond issuers. In columns (4) - (6), the measures of financial stress are defined at bank level and are the total borrowing from the Eurosystem over total assets, the ratio of non-performing loans, and the change in the issuer’s average credit ratings over the past two quarters, respectively. All regressions include AMG, style, and year-quarter fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

(1) (2) (3) (4) (5) (6)

AFF 0.0001 -0.0004* -0.0005 0.0002 -0.0004 -0.0003 (0.000) (0.000) (0.000) (0.000) (0.001) (0.000) AFF × STRESS 0.0009 0.0011 0.0002 0.0002 0.0010 (0.001) (0.001) (0.000) (0.000) (0.001) STRESS 0.0000 0.0005 -0.0027 (0.000) (0.000) (0.002) LOG_FUND_AUM -0.0001 -0.0001 -0.0000 -0.0001 -0.0000 0.0001 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) LOG_AMG_AUM 0.0006 0.0006 0.0014 0.0013 0.0016 0.0013 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) LAG_RET -0.0000 -0.0000 -0.0001 -0.0001 -0.0001 -0.0001 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) FLOW/AUM 0.0048** 0.0048** 0.0030*** 0.0029*** 0.0029*** 0.0030*** (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) FEES 0.0002 0.0002 0.0000 0.0001 0.0000 0.0000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) RETAIL 0.0001 0.0001 -0.0001 -0.0001 -0.0001 0.0000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) LOG_AMT_OUTSTANDIN G 0.0001 0.0001 0.0001 0.0002 0.0002 0.0001 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) AMG FE Yes Yes Yes Yes Yes Yes Style FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Observations 942,901 942,901 525,758 505,715 518,563 546,800 Adj. R-squared 0.000 0.000 0.000 0.000 0.000 0.000

Table 9: Funding support through other financial instruments This table presents the coefficient estimates of the regression equation

𝑃,, 𝛼, 𝛼, 𝛼 𝛽𝐴𝐹𝐹,, 𝛽𝐴𝐹𝐹,, 𝑆𝑇𝑅𝐸𝑆𝑆 𝛽𝑆𝑇𝑅𝐸𝑆𝑆 𝛿𝑋,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, 𝜀,,,

where 𝑃,, denotes the EUR amount purchased in quarter t by fund f of a given financial instrument sold by issuer i, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, denotes the total EUR amount issued by issuer i in quarter t, and 𝐴𝐹𝐹,, is an indicator variable which takes the value of 1 if mutual fund f is affiliated with issuer of i at time t and zero otherwise. 𝑆𝑇𝑅𝐸𝑆𝑆 is a dummy that takes the value of one in the period after the Lehman bankruptcy (i.e. starting in 2008:Q3), and zero otherwise. 𝑋,, is a vector of control variables that are defined in Appendix A. Column (1) reports the results obtained when we measure funding support through bonds as in the baseline analysis whereas in columns (2), (4), (5) funding support is measured through short-term paper (STP), preferred shares and equity. The dependent variable used to measure funding support in column (3) is based on deposits and it is obtained as the EUR amount deposited in quarter t by fund f in bank i relative to the fund’s contemporaneous EUR amount of assets under management. Each column is estimated excluding all funds that never invest into the corresponding type of instrument under analysis. The information on deposits is only available after the last quarter of 2008 whereas for the other alternative instruments it is available from 2000. All regressions include AMG, style, and year-quarter fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

Bonds Notes Deposits Preferred Sh Equity Issuances AFF -0.0018 0.1377*** 0.0007 0.2435 (0.002) (0.027) (0.025) (0.491) AFF x POST 0.0952*** 0.1270*** 0.5837*** -0.0172 -0.8065 (0.034) (0.043) (0.054) (0.023) (0.492) Style FE YES YES YES YES YES AMG FE YES YES YES YES YES Time FE YES YES YES YES YES Fund + Bond Controls YES YES YES YES YES Observations 642,957 992,113 714,321 2,144 79,391 R-squared 0.002 0.040 0.021 0.146 0.181

Table 10: Post-issuance abnormal returns This Table presents the coefficient estimates from the linear regression

𝐴𝑅,, 𝜇 𝜇𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸, 𝜇𝑆𝑇𝑅𝐸𝑆𝑆 𝜇𝑆𝑇𝑅𝐸𝑆𝑆 𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸, 𝜁,,

where 𝐴𝑅, denotes the abnormal return on bond i on the issuance date (first three columns), the cumulative abnormal return during the bond’s first week after issuance, (middle columns), or the cumulative abnormal return during the bond’s first month after issuance (last three columns). Abnormal returns are computed relative to other Spanish banks bonds from the same maturity and ratings category. See section 5.1. for details. AFF_PURCHASE is a dummy variable equal to one if at least one mutual fund affiliated with the issuer of bond i has purchased the bond in the primary market, and zero otherwise. STRESS is a time-varying variable capturing times of financial stress, which is defined as a dummy that takes the value of one in the period after the Lehman bankruptcy (i.e. starting in 2008:Q3), and zero otherwise. The row “Estimated abnormal returns in crisis times” gives the estimated abnormal returns in crisis times for bonds purchased by affiliated funds, i.e. the fitted values for 𝐴𝐹𝐹_𝑃𝑈𝑅𝐶𝐻𝐴𝑆𝐸 1 and STRESS 1, respectively. All regressions include issuer fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

1-DAY ABNORMAL RETURN 1-WEEK ABNORMAL RETURN 1-MONTH ABNORMAL RETURN AFF_PURCHASE -0.0029*** -0.0009 -0.0033*** 0.0002 -0.0044** -0.0006 (0.001) (0.001) (0.001) (0.001) (0.002) (0.001) STRESS -0.0005 0.0006 0.0014 (0.001) (0.001) (0.003) STRESS × AFF_PURCHASE -0.0023** -0.0050** -0.0060 (0.001) (0.002) (0.004)

Estimated abnormal return in crisis times -0.0038*** -0.0042** -0.0053** Issuer FE YES YES YES YES YES YES Observations 187 187 187 187 187 187 Adj. R-squared 0.028 0.031 0.020 0.031 0.017 0.018

Table 11: Relative performance of funding support purchases This table contains the coefficient estimates from the linear regression 𝐴𝑅,, 𝐴𝑅,, 𝜇𝑏 𝜇1𝑆𝑇𝑅𝐸𝑆𝑆 𝜀,, where 𝐴𝑅,, and 𝐴𝑅,, denote the abnormal (portfolio) returns earned by fund f in quarter t from affiliated and non-affiliated bonds, respectively. Abnormal returns are computed relative to other Spanish banks bonds from the same maturity and ratings category. See section 5.1 for details. 𝑆𝑇𝑅𝐸𝑆𝑆 is a time-varying variable capturing times of financial stress, which is defined as a dummy that takes the value of one in the period after the Lehman bankruptcy (i.e. starting in 2008:Q3), and zero otherwise. Column (1) is estimated on the entire sample, while column (2) excludes purchases by guaranteed funds. Finally, column (3) excludes purchases of AAA-rated securities. All regressions include AMG fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

Variables (1) (2) (3)

STRESS -0.0092*** -0.0096*** -0.0137* (0.003) (0.003) (0.007) AMG FE YES YES YES Observations 307 294 122 Adj. R-squared 0.014 0.015 0.027

Table 12: Deposit rates Columns (1) – (4) of this table report the difference in value-weighted average deposit rates between deposits in affiliated and non-affiliated banks. The computation is based on all fund-quarter observations where fund f makes a new deposit in at least one affiliated and one non-affiliated bank. The results in columns (1) and (2) are based on raw deposit rates, while the numbers in columns (3) and (4) are computed after adjusting for differences in bank risk by subtracting the average deposit rate in the same rating and maturity category. We distinguish seven rating categories (i.e. AAA, AA, A, BBB, BB, B, and CCC) and two maturity buckets (i.e. maturities below and above six months). Columns (5) and (6) report the average difference between the interest rates paid by bank i in quarter t on deposits of affiliated and non-affiliated mutual funds. Columns with odd numbers refer to the entire sample, while the remaining columns refer to results obtained when discarding guaranteed funds. Standard errors clustered at the AMG/Bank level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

(1) (2) (3) (4) (5) (6)

MEAN -0.2441*** -0.2516*** -0.0611* -0.0687** -0.1646* -0.2410** (0.032) (0.033) (0.033) (0.032) (0.098) (0.101) St-Err. Cluster AMG AMG AMG AMG BANK BANK Observations 492 459 510 495 162 141 Adj. R-squared 0.244 0.250 0.162 0.177 0.227 0.185

Table 13: Flow-performance relationship Columns (1) and (2) of this table contains the coefficient estimates from the regression equation 𝐹𝐿𝑂𝑊, 𝜃𝐴𝑀𝐺 𝜃𝑆𝑡 𝜃𝑡 𝜃𝑃𝐸𝑅𝐹, 𝜃𝑃𝐸𝑅𝐹, 𝑅𝐸𝑇𝐴𝐼𝐿 𝜃𝑅𝐸𝑇𝐴𝐼𝐿 𝐴𝑈𝑀,

𝜃𝑋, 𝜉,,

where 𝐹𝐿𝑂𝑊, measures the flow in or out of fund f in quarter t, 𝐴𝑈𝑀, denotes lagged fund size, 𝑃𝐸𝑅𝐹, denotes lagged fund performance, and 𝑅𝐸𝑇𝐴𝐼𝐿 is a dummy variable equal to one for retail funds (defined as funds that are in the lower 75 percentile of the cross-sectional distribution of AUM per shareholder during every quarter, see Appendix A). 𝑋, is a vector of lagged control variables, defined in Appendix A. The dependent variable is expressed in percentage units. Columns (1) and (2) use the return in excess of the 3-month Spanish treasury bill. The performance measure in columns (3) and (4) is based on the fund’s fractional rank (RANK) which is defined as its percentile performance relative to other funds with the same investment objective in the same period, and ranges from 0 to 1. In columns (3) and (4), we estimate the previous regression equation using three new measures of performance in place of the fund’s return in excess of the 3-month Spanish Treasury bill: LOWPERF is defined as Min(RANK,0.2), MIDPERF is defined as Min(0.6, RANK- LOWPERF) and finally HIGHPERF is defined as RANK – (LOWPERF + MIDPERF) with RANK defined as fund’s rank in terms of lagged performance within all funds with same investment objective and in the same period. These three performance measures are rescaled (i.e., multiplied times 100) to be consistent with the scale of the dependent variable. In the rows immediately after the regression coefficient of columns (2) and (4), we report the linear combinations of a series of coefficients. All regressions include AMG, style, and year-quarter fixed effects. Standard errors are clustered at the AMG level and given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

(1) (2) (3) (4) LAG_PERF 0.3798*** 0.5476*** (0.059) (0.082) LAG_PERF * RETAIL -0.3005*** (0.088) RETAIL -0.3666 -0.4355 -0.1783 0.5922 (0.322) (0.325) (0.320) (1.002) LOWPERF 0.0617*** 0.0313 (0.023) (0.050) MIDPERF 0.0352*** 0.0666*** (0.007) (0.010) HIGHPERF 0.1736*** 0.2270*** (0.035) (0.043) RETAIL * LOWPERF 0.0526 (0.054) RETAIL * MIDPERF -0.0524*** (0.012) RETAIL * HIGHPERF -0.1324* (0.080)

Table 13, continued

(0.053) HIGHPERF + RETAIL * HIGHPERF 0.011 - LOWPERF - RETAIL * LOWPERF (0.054)

Control Variables YES YES YES YES AMG FE YES YES YES YES Fund Inv Style FE YES YES YES YES Time FE YES YES YES YES Observations 51,200 51,200 51,200 51,200 Adj. R-squared 0.085 0.086 0.088 0.089

Table 14: Funding support in the primary market and the role fund characteristics This table contains the coefficient estimates of the regression equation 𝑃,, 𝛼 𝛼 𝛼 𝛽𝐴𝐹𝐹,, 𝛽𝐹𝑈𝑁𝐷_𝐶𝐻𝐴𝑅𝐴𝐶𝑇𝐸𝑅𝐼𝑆𝑇𝐼𝐶, 𝛽𝐴𝐹𝐹,, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, 𝐹𝑈𝑁𝐷_𝐶𝐻𝐴𝑅𝐴𝐶𝑇𝐸𝑅𝐼𝑆𝑇𝐼𝐶, 𝛿𝑋,, 𝜀,,, where 𝑃,, denotes the EUR amount purchased in quarter t (the quarter of issuance) by fund f of bond issue i, 𝐼𝑆𝑆𝑈𝐴𝑁𝐶𝐸, denotes the total EUR amount issued of bond i in quarter t, 𝐴𝐹𝐹,, is an indicator variable which takes the value of 1 if mutual fund f is affiliated with the issuer of bond i at time t and zero otherwise. 𝐹𝑈𝑁𝐷_𝐶𝐻𝐴𝑅𝐴𝐶𝑇𝐸𝑅𝐼𝑆𝑇𝐼𝐶, denotes the fund characteristics of interest that is interacted with the dummy 𝐴𝐹𝐹,,. In column (1) the characteristic of interest is denoted as 𝑅𝐸𝑇𝐴𝐼𝐿 and corresponds to a dummy variable equal to one for retail funds (defined as funds that are in the lower 75 percentile of the cross-sectional distribution of AUM per shareholder during every quarter, see Appendix A). In column (2) we use 𝑃𝐸𝑅𝐹_𝐹𝐸𝐸𝑆, which is a dummy variable that is equal to one if the fund has performance fees. All regressions include AMG, style, and year-quarter fixed effects, and standard errors clustered at the AMG level are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively. (1) (2)

AFF 0.0413*** 0.0438** (0.015) (0.020) RETAIL -0.0019 (0.001) AFF × RETAIL 0.0362** (0.018) PERF_FEES -0.0036* (0.002) AFF × PERF_FEES -0.0394* (0.021) Control Variables Yes Yes AMG FE Yes Yes Style FE Yes Yes Time FE Yes Yes Observations 642,957 642,957 Adj. R-squared 0.002 0.002

Table 15: Mutual funding and credit supply This table presents the coefficient estimates of the regression equation

log1 Credit,,log1Credit,,𝜆 𝜆AP 𝜆𝑋 𝜈,, where Credit,, and Credit,, denote the total credit committed (in thousands of euros) by bank b to firm f, both drawn and undrawn, at June 2012 and December 2008, respectively. The variable AP (Affiliated Purchases) denotes the total purchases of bank b’s securities by its affiliated funds. Each column of the table refers to a different definition of AP: logarithm of the amount of bonds purchased (Log(1+MF_Bond)) in column (1); amount of bonds purchased relative to total assets (MF_Bonds/TA) in column (2); amount of bonds purchased relative to equity (MF_Bonds/Eq) in column (3); amount of bonds purchased relative to the AUM by affiliated funds (MF_Bonds/AUM) in column (4). 𝑋 is a vector of bank-level controls including the total amount of bonds issued from December 2008 to June 2012 (in logs), as well as bank characteristics as of December 2008: profitability, size, leverage, ratio of NPL, credit over deposits, borrowing from the Eurosystem, diversification ratio, and share of liquid assets. 𝜆 denotes firm fixed effects. Standard errors are robust to heteroskedasticiy and are given in parentheses. ***, **, and * denotes statistical significance at 1%, 5%, and 10% level, respectively.

Log(MF_Bonds) MF_Bonds / TA MF_Bonds / Eq MF_Bonds / AUM (1) (2) (3) (4)

AP 0.0386*** 50.0857*** 2.3463*** 2.4626*** (0.002) (5.948) (0.294) (0.263) Bank Characteristics YES YES YES YES Firm FE YES YES YES YES Observations 864,563 864,563 864,563 864,563 R-squared 0.509 0.508 0.508 0.508

Figure 1: Bond issuance over time This figure plots the evolution of the bond issuance of Spanish financial institutions over time, based on the securities found in Dealogic (dashed line). The figure also includes the yearly number of issues that form part of our sample of 504 bonds held by Spanish mutual funds (dashed line with markers).

Total Issuance # of Issues (Bln. EUR) 210 140

180 120

150 100

120 80

90 60

60 40

30 20

0 0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Total # Issues (LHS) Sample # Issues (LHS) Total Issuance (RHS) Sample Issuance (RHS)