- Email: [email protected]
Bank business models and liquidity creation
Journal Pre-proof Bank business models and liquidity creation Dung Viet Tran (Conceptualization) (Methodology) (Software) (Data curation)Writing- Original draft) (Writing review and editing)
PII:
S0275-5319(19)30643-9
DOI:
https://doi.org/10.1016/j.ribaf.2020.101205
Reference:
RIBAF 101205
To appear in:
Research in International Business and Finance
Received Date:
4 June 2019
Revised Date:
14 February 2020
Accepted Date:
20 February 2020
Please cite this article as: Viet Tran D, Bank business models and liquidity creation, Research in International Business and Finance (2020), doi: https://doi.org/10.1016/j.ribaf.2020.101205
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier.
Bank business models and liquidity creation Dung Viet Tran1
1
of
Banking University Ho Chi Minh City, Ho Chi Minh city 700000, Vietnam. Mail: [email protected]. The author thanks the editor (John W. Goodell), two anonymous reviewers for their constructive comments and suggestions. The author gratefully acknowledges financial support of the National Foundation for Science and Technology Development of Vietnam (NAFOSTED).
Bank business model and liquidity creation
Expansion hypothesis Bank
re
business
-p
Diversification-Liquidity
model
ro
Graphical abstract
Liquidity creation
lP
Diversification-Liquidity
Abstract
ur na
Contraction hypothesis
Using a large panel of US bank holding companies from 2001 to 2015, this study investigates the
Jo
association between functional diversification and bank liquidity creation. I document evidence of lower liquidity creation for higher diversification. The effect of moving into nontraditional activities on liquidity creation is more apparent with large banks and less pronounced with small banks. The impact of diversification on liquidity creation is less significant during the late stage of crisis and is more clearly observed in small and medium-sized banks. Low liquidity creation banks, leveraged by a higher share of non-interest income, are more likely to further decrease
1
their liquidity creation. The study is of interest to regulators and policymakers who are concerned about bank business models.
Keywords: Bank liquidity creation; business models; diversification; crisis JEL Classification Codes: G21; G28; G34; G38
1. Introduction
of
Liquidity creation is a principal reason for the existence of banks. As liquidity providers, banks play an important role for the macroeconomy and the financial system. Banks provide necessary
ro
liquidity to informationally opaque borrowers without capital market opportunities (Levine and Zervos 1998) and supply liquid funds and payment services to households, which is the main
-p
driver for the functioning of the economy (Kashyap, Rajan, and Stein 2002). In other words, banks simultaneously satisfy the demand for liquidity by savers and the demand for longer-term
re
financing commitments by firms. Banks also provide loan commitments and other off-balance sheet guarantees that allow customers to plan their investments and expenditures, knowing that
and Stein 2002).
lP
the required funds are forthcoming when needed (Holmström and Tirole 1998; Kashyap, Rajan,
Deregulation, which started during the 1970s, allows banks to operate with fewer
ur na
constraints and to expand into highly volatile and complex non-bank activities, such as trading and market making (Tran et al. 2019), which, in turn, adds ever greater complexity to banks’ balance sheets (Financial Crisis Inquiry Commission 2011), increasing problems of information asymmetry in diversified banks. Most of the literature on bank diversification focuses on the relation between the bank business models and bank risk taking and performance (e.g. Stiroh
Jo
2004; Stiroh and Rumble 2006; DeYoung and Torna 2013; Tran et al. 2019). Generally, these studies point out that banks can benefit from diversification through a reduction of idiosyncratic risk and total risk, but this comes with the cost of heightened agency problems, leading to dispersed managerial resources and reduced stability. With the failure of a large number of banks and subsequent economic recessions, many claim the casino-style gambling on Wall Street was one of the main contributors to the last financial crisis. The banking industry is a particularly important sector in our economies, serving as a channel whose disruption could lead to adverse 2
fluctuations in the real economy. A sound banking system is a primary objective of regulators and policymakers (Tran et al. 2019). Given the importance of banks for the real economy, the lack of research on the activity strategies determining bank liquidity creation is surprising and potentially significant. This study is one of the first empirical investigations of whether liquidity creation differs across banks with different activity strategies. It provides empirical evidence on differing theoretical perspectives concerning the impact of functional diversification on bank liquidity
of
creation. On the one hand, under the framework of portfolio theory, the combination of different activities is generally believed to reduce the variance of returns, since they are not perfectly
correlated. This results in a coinsurance effect, diversification gains, and a more stable revenue
ro
stream (DeYoung and Roland 2001; Tran 2019), reducing the total risk and probability of failure of diversified banks (Brewer 1989; Saunders and Cornett 2008). Furthermore, in the absence of
-p
agency conflicts between banks and borrowers, diversified banks can retrieve borrower information and reuse it profitably for nontraditional banking activities, such as securities
re
underwriting (Diamond 1984; Yasuda 2005; Bharath et al. 2007). In turn, the information retrieved from non-interest income–generating activities also helps diversified banks make their
lP
loan-making decisions, resulting in better credit risk management. Diamond's (1984) seminal model suggests that banks that move toward nontraditional banking activities experience a more stable credit supply under aggregate shocks. When assuming the liquidity creation role, banks
ur na
are subject to liquidity risk, since they have to meet depositors’ demand for liquidity while providing funding to borrowers (Diamond and Rajan 2000, 2001). The more liquidity banks create, the greater the probability and severity of losses associated with the disposal of illiquid assets needed to meet customers’ liquidity demands (Berger and Bouwman 2009). Therefore, I suggest that diversified banks should create more liquidity than other banks, since they are
Jo
financially stronger in terms of meeting the demand from depositors and providing funds to borrowers. I call this hypothesis the diversification–liquidity expansion hypothesis. On the other hand, diversification raises concerns of intensified agency problems
associated with bank size, as well as banks’ opaqueness and complexity, leading to discretionary decisions to undertake value-decreasing investments (Berger and Ofek 1995). Managers derive private benefits from diversification that exceed their private costs (Denis, Denis, and Sarin 1997). They can adopt and maintain a strategy of inefficient diversification, even if it can 3
negatively affect shareholder value. Furthermore, more diversified activities do not translate into risk reduction if there is lack of expertise in the newly adopted business (Jiménez and Saurina 2004). Diversification can then spread out managerial resources, leading to higher costs to coordinate corporate policies and, in turn, inducing failure to meet the liquidity demands of both depositors and borrowers and negatively affecting liquidity creation. I call this hypothesis the diversification–liquidity contraction hypothesis. In this paper, I examine the interplay of diversification and bank liquidity creation for a
of
large sample of US bank holding companies (BHCs) from 2000:Q1 to 2015:Q4. Using the richest and most complete database related to banking institutions that provides the details of thousands of banks operating in hundreds of local markets, I can analyze the data at the highest
ro
frequency possible. I use the preferred measure of Berger and Bouwman (2009), catfat, as the main proxy for bank liquidity creation (LC), which is also used in numerous other studies
-p
(Distinguin, Roulet, and Tarazi 2013; Horvath, Seidler, and Weill 2016; Berger and Bouwman 2017; Jiang, Levine, and Lin 2017). In robustness checks, I also use different measures and still
re
find similar findings. I use the share of non-interest income (NII) as the proxy for diversification. A large share of non-interest income means greater diversification into nontraditional activities.
lP
Controlling for the effects of different bank characteristics and bank and time fixed effects, the empirical analysis provides consistent evidence of lower liquidity creation for diversified banks. The evidence is economically significant. For example, the baseline model
ur na
indicates that a one standard deviation increase in NII will lead to a decrease of LC of 0.014. Given LC’s mean of 0.445 and standard deviation of 0.186, this decrease is equivalent to 3.15% of the average bank’s assets. The evidence provides support for the diversification–liquidity contraction hypothesis, which suggests that agency problems derived from the complexity of diversified banks lead banks to fail to meet their clients’ liquidity demands.
Jo
To ensure the robustness of the findings, I run the baseline model with a single cross-
sectional regression (average analysis), with a balanced panel (excluding banks that only existed during part of the sample period), excluding merger and acquisition (M&A) banks (i.e., banks with large changes in assets over a certain periods, where I use a threshold of a 20% change in assets over a quarter), and with alternative econometric models, such as Newey–West, Fama– MacBeth, and two-way clustering by bank and time. The results of these specifications lend support to the diversification–liquidity contraction hypothesis. 4
I focus on the distribution of liquidity creation and use quantile regression to examine how the impact of activity strategies varies across the distribution of liquidity creation. Rather than relying on a single description of the central behavior of the sample, the quantile approach explores a range of conditional quantile functions, which, in turn, allows potential forms of conditional heterogeneity to be explored. The evidence indicates that diversification not only affects the conditional average of liquidity creation, but also influences the dispersion of liquidity creation. Low liquidity creation banks, leveraged by a higher share of non-interest income, are
of
more likely to decrease their liquidity creation. In other words, the impact on diversification appears to be more profound for low liquidity creation banks.
Since banks of different size create liquidity in different ways (Berger and Bouwman
ro
2009), I run the baseline model for banks with a range of different sizes: small banks, with assets of up to $1 billion; medium-sized banks, with assets between $1 billion and $5 billion; and large
-p
banks, with assets of over $5 billion. I document variations of the effect of diversification on bank liquidity creation across bank sizes. Large banks that engage in non-interest income–
re
generating activities, reduce their liquidity creation the most, followed by small diversified banks. Medium-sized banks reduce their liquidity creation the least when increasing their share
lP
of non-interest income. Potential explanations are as follows. Larger banks are more likely to gain from nontraditional activities for a long time and to then push their diversification level to saturation, no longer gaining benefit from marginal increases in diversification but potentially
ur na
incurring greater agency problems. This implies a large reduction in liquidity creation by large banks. Even though they have more to gain from diversification, small banks can lack experience in new and more complex nontraditional activities, thereby reducing their operating stability and liquidity creation.
I investigate whether diversification affects liquidity creation differently during financial
Jo
crises. Diversified banks create more (less negative) liquidity than other banks during crises than under normal circumstances. The evidence highlights the bright side of diversification, which could make banks less risky during times of turmoil, as documented by DeYoung and Torna (2013) and Tran et al. (2019), giving rise to more liquidity creation. When performing analyses across bank sizes, I find no differences in the effect of diversification on liquidity creation across bank sizes during times of crisis.
5
Since differences could exist in deposit flows between the early and late stages of the global financial crisis that affected banks’ ability to create liquidity, I split the crisis period into two subperiods—CRISIS 1, from 2007Q3 to 2008Q2, and CRISIS 2, from 2008Q3 to 2009Q2— and rerun the analysis separately for each stage of the crisis. I find that the bright side of diversification is mostly driven by the second stage of the crisis (CRISIS 2). Small and mediumsized banks that diversify into nontraditional banking activities create more liquidity during the late stage of the crisis, and this effect seems to be stronger for medium-sized banks. I find no
of
differences in the effect of diversification on liquidity creation at large banks during times of crisis.
In additional robustness tests, I use alternative measures of liquidity creation (deviating a
ro
bank’s liquidity creation from the industry average, on- and off-balance sheet components of liquidity creation). I also use alternative measures of diversification, such as the share of trading
-p
income, the ratio of loans over total assets, and an adjusted Herfindahl–Hirschman index. The results remain similar.
re
I also assess the net effect of moving toward nontraditional activities on bank liquidity creation by evaluating the 10th, 25th, 50th, 75th, 90th percentiles of the share of non-interest
lP
income. I find that banks benefit from diversification, but these gains are quickly offset by adverse effects from reliance on riskier assets. Banks with a relatively small share of non-interest income do not seem to change their liquidity creation when relying more on non-interest
ur na
income–generating activities. However, with greater reliance on non-interest income–generating activities, banks reduce their liquidity creation with an increase in the share of non-interest income.
Next, I address endogeneity concerns, since the decision to diversify is a deliberate decision by bank managers. They could choose to diversify when they receive more benefits than
Jo
the cost of diversification (Campa and Kedia 2002). I begin with propensity score matching (PSM). Next, I use a Heckman two-step model and an instrumental variable (IV) approach to examine whether the private information of the decision to diversify is correlated with the propensity to create liquidity. In all specifications, the findings remain quantitatively similar to the main results. This study contributes to the literature in several ways. First, to the best of my knowledge, this study is the first investigation of the impact of functional diversification on 6
liquidity creation within the US banking industry. In a concurrent study, Hou et al. (2018) document how an increase in the degree of bank diversification between traditional bank activities generating net interest income and nontraditional bank activities generating noninterest income reduces the liquidity creation of Chinese banks. Unlike the US banking system, the banking sector in China, as well as regulations regarding market access and the range of products, remains under strict government supervision. Bank’s net interest margin earnings are the main component of their profits. This study’s focus on one of the most regulated banking
of
industries, that in the United States, where banks have been allowed to extensively expand their activities into nontraditional activities for decades, and its comprehensive sample of long
duration strengthen its results, increase their robustness, and reduce sample selection bias. The
ro
study contributes to the liquidity creation literature by providing evidence of the dark side of
diversification for bank liquidity creation during normal circumstances, as well as a bright side
-p
of diversification during times of crisis. Furthermore, this study documents how the impact of diversification on bank liquidity creation varies by bank size.
re
Second, this study provides evidence of the effects of diversification over the entire range of liquidity creation distribution. The traditional inference approach (i.e., ordinary least squares,
lP
or OLS) reflects the mean behavior of the sample under the assumption of homogeneity in the relation between diversification and liquidity creation. However, with a sample as heterogeneous as this study’s, this approach could be a poor method for examining the relation between
ur na
diversification and liquidity creation across the entire industry. Rather than relying on a single description of the central behavior of the sample, the quantile approach explores a range of conditional quantile functions, which, in turn, allows potential forms of conditional heterogeneity to be explored (Tran, Hassan, and Houston 2019). I document that activity strategies not only affect the conditional average level of liquidity creation, but also influence the distribution of
Jo
liquidity creation.
This paper is organized as follows. The next section describes the data and variables.
Section 3 reports the main results and those of alternative tests. Section 4 provides the results of additional tests. Section 5 concludes the study.
7
2. Theoretical framework and hypothesis development Banks assume two central roles in our economy—creating liquidity and transforming risk— which are often jointly referred to as the qualitative asset transformation of banks (Bhattacharya and Thakor 1993). Despite a vast literature dedicated to the risk transformation role of banks (i.e., issuing riskless deposits to finance riskier loans), few empirical studies have focused on the liquidity creation role of banks. Banks create liquidity on balance sheets by financing illiquid loans, for example, business loans with relatively liquid liabilities, such as transaction deposits (Bryant 1980; Diamond and Dybvig 1983). They provide borrowers the necessary funds to
of
invest, while delivering deposits to assume liquidity and payment services to the public. In other words, the liquidity creation role of banks simultaneously satisfies the demand for liquidity from
ro
savers and the demand for longer-term financing commitments from firms. Banks also create liquidity off the balance sheets by providing loan commitments and other off-balance sheet
-p
guarantees that allow firms to develop and modify long-run investment strategies efficiently (Holmström and Tirole 1998; Kashyap, Rajan, and Stein 2002). Loan commitments allow
re
customers to plan their investments and expenditures, knowing that the required funds are forthcoming when needed (Berger and Bouwman 2017).
lP
Until recently, liquidity creation was mostly relegated to a theoretical concept and was not often used in empirical studies. Since Berger and Bouwman (2009) filled an important gap in bank research by providing a comprehensive measure of liquidity creation, a body of literature
ur na
has emerged on the association between bank liquidity creation and other bank financial decisions, such as equity capital (Horváth, Seidler, and Weill 2013), risk taking (Andreou, Philip, and Robejsek 2016), and other financial instruments and phenomena, such as monetary policy (Berger and Bouwman 2017), economic output (Berger and Sedunov 2017), government intervention (Berger et al. 2016), and competition (Horvath, Seidler, and Weill 2016; Jiang,
Jo
Levine, and Lin 2017). However, the present study is one of the first to investigate the impacts of the bank business models on liquidity creation. The bank business models, involving the integration of traditional and nontraditional
banking activities, has also been thoroughly examined. Most of the studies focus on the association between activity diversification and bank riskiness (Stiroh 2004; Stiroh and Rumble 2006; Berger, Hasan, and Zhou 2010; De Jonghe, Diepstraten, and Schepens 2015; Tran et al. 2019). 8
Within the framework of modern portfolio theory, the combination of different activities is generally believed to reduce the variance of returns, since they are not perfectly correlated, reducing the probability of failure (Brewer 1989; Saunders and Cornett 2008). Focused banks might have to reject positive net present value investment opportunities if external finance is costly (Kashyap and Stein 2000; Campello 2002) or if they are experiencing low income levels (Houston, James, and Marcus 1997). Diversified banks can reduce this problem, since diversification helps banks decrease volatility and the correlation between funding and
of
investment opportunities. By moving away from traditional intermediation activities, banks earn less interest income and, at the same time, earn less interest and take on less credit risk (Saunders and Walter 1994). Diversification could reduce the correlation of loan returns and, subsequently,
ro
idiosyncratic credit risk (Acharya, Hasan, and Saunders 2006), which decreases the risk of large liquidity outflows and the risk of early liquidation (Diamond and Dybvig 1983).
-p
Additionally, assuming the absence of agency conflicts between banks and borrowers, Diamond (1984) argues that diversified banks can enhance credibility in their loan making
re
decisions and in their borrowers’ monitoring by overcoming the information asymmetry between depositors and borrowers. Diversified banks can retrieve borrower information and reuse it
lP
profitably for nontraditional banking activities such as securities underwriting (Diamond 1984; Yasuda 2005; Bharath et al. 2007). In turn, the information retrieved from non-interest income– generating activities also helps diversified banks in their loan making decisions, resulting in
ur na
better credit risk management. Diamond (1984) suggests that diversified banks have a more stable credit supply under aggregate shocks, which could, in turn, lead to lower volatility of the cash flow from loan portfolios. In brief, these arguments suggest that banks engaging in nontraditional activities reduce the need for self-insurance, benefit from economies of scale and scope (Santomero and Eckles 2000), increase profitability (Sanya and Wolfe 2011), achieve
Jo
capital savings (Shim 2013), and experience a reduction in idiosyncratic risk exposure (Kwast 1989), total risk (Tran 2019), and the probability of failure (DeYoung and Torna 2013). When assuming the liquidity creation role, banks are subject to liquidity risk, since they
must meet depositors’ demand for liquidity when providing funding to borrowers (Diamond and Rajan 2000, 2001). In other words, the more liquidity banks create, the greater the probability and the severity of the losses associated with having the disposal of illiquid assets to meet their customers’ liquidity demands (Berger and Bouwman 2009). Kashyap, Rajan, and Stein (2002) 9
argue banks can still create liquidity as long as an imperfect correlation exists between drawdowns on demandable deposits and credit lines. Diversification can make simultaneous drawdowns less likely, allowing banks to create more liquidity. Therefore, I suggest that diversified banks should create more liquidity than other banks, since they are financially better able to meet the demand from depositors and to provide funds to borrowers. I call this hypothesis the diversification–liquidity expansion hypothesis. H1a (diversification-liquidity expansion hypothesis): Banks engaging in nontraditional
of
activities create more liquidity than other banks. On the other hand, diversification raises concerns of intensified agency problems, since it can increase a bank’s size, as well as its opaqueness and complexity, leading to discretionary
ro
decisions to undertake value-decreasing investments (Berger and Ofek 1995; Krishnaswami, Spindt, and Subramaniam 1999). The information asymmetry could prevent shareholders from
-p
rigorously monitoring managers, leading to increased agency problems. Managers derive private benefits from diversification that exceeds their private costs (Denis, Denis, and Sarin 1997).
re
They can adopt and maintain a strategy of inefficient diversification, even if it could negatively affect shareholder value (i.e., the risk shifting problem and the underinvestment problem), since
lP
managing a diversified bank can increase power and prestige (Jensen 1986), entrenchment (Shleifer and Vishny 1989), and compensation arrangements (Jensen and Murphy 1990), as well as reduce the risk of managers’ undiversified personal portfolios (Amihud and Lev 1981).
ur na
Furthermore, more diversified activities do not translate into risk reduction if there is a lack of expertise in the newly adopted business (Jiménez and Saurina 2004). Due to the moral hazard related to fixed-rate deposit insurance or the risk-seeking behaviors of bank management (Litan 1985), banks could support risky investments by using federally insured deposits. The seminal theoretical model of Shleifer and Vishny (2010) suggests banks that move toward nontraditional
Jo
banking activities tend to expand their balance sheets and transmit fluctuations from the security markets to the real economy; the volatility of sentiment thus turns into the volatility of real activity through the banking sector. The authors emphasize that such profit maximization practices could induce system instability. Diversification could then spread out managerial resources, leading to higher costs of coordinating corporate policies, which, in turn, induces failure to meet the liquidity demands of
10
both depositors and borrowers and negatively affects liquidity creation. I call this hypothesis the diversification–liquidity contraction hypothesis. H1b (diversification–liquidity contraction hypothesis): Banks engaging in nontraditional activities create less liquidity than other banks. The banking literature tends to presume that diversification and size go hand in hand. However, larger banks that are usually better diversified than smaller banks are not less risky. Studies document that large banks benefit from diversification advantages to pursue riskier
of
activities (Demsetz and Strahan 1997). Nevertheless, large banks are typically exposed to more stringent monitoring from regulators and other stakeholders. Therefore, I propose the following hypothesis.
ro
H2: The effect of diversification on bank liquidity creation varies with bank size.
Another interesting question relates to the association between diversification and
-p
liquidity creation during times of crisis. The financial crisis of 2007–2009 led to the failure of a large number of banks, forced recapitalizations, distressed mergers with healthier banks, and
re
significant government intervention around the world. In this context, it is of particular interest to investigate whether this extraordinary situation led to a reassessment of the role of bank
lP
diversification. On the one hand, diversification potentially enables banks to better resist shocks in one area of their business activity. This suggests that diversified banks have incentives to increase liquidity creation during crises to take advantage of market share. On the other hand,
ur na
banks that engage in more nontraditional banking activities could be the most clearly exposed, due to commercial banks expanding into investment banking activities (Elsas, Hackethal, and Holzhäuser 2010). It therefore follows that diversified banks will curtail their intermediation activity to protect financial stability. The following hypotheses will therefore be tested. H3a: The effect of diversification on bank liquidity creation is mitigated during a crisis.
Jo
H3b: The effect of diversification on bank liquidity creation is stronger during a crisis.
3. Data and variables 3.1 Sample banks
We obtain from the Federal Reserve System quarterly FR Y-9C regulatory reports filed by BHCs with assets of $150 million and over. The raw data cover the period from 2001:Q1 to 2015:Q4. Bank–quarter observations with missing or incomplete financial data for the accounting variables 11
in the main regression model are removed. Following Berger and Bouwman (2013), I replace all observations with a ratio of total equity to total assets of less than 1% by 1% to avoid distortions in ratios involving equity. I also exclude observations with negative or nonexistent outstanding loans or deposits. The data set contains 32,946 observations for 1,767 BHCs. All financial ratios are winsorized at the 1% level at the top and bottom of their distribution to mitigate the effects of outliers.
of
3.2 Measures of bank liquidity creation I use the liquidity creation measure of Berger and Bouwman (2009). Since banks are able to
securitize or sell loans, they prefer to classify loans by category (cat) rather than by maturity
ro
(mat). Additionally, since banks create liquidity both on and off the balance sheet, they prefer including off-balance sheet items (fat) to excluding them (nonfat). I therefore use their preferred
-p
measure, the catfat measure, which includes both on- and off-balance sheet items and classifies items by category. In robustness tests, I use alternative measures and still obtain similar
re
evidence. The data on liquidity creation were retrieved from Christa Bouwman’s website.1 The catfat and other liquidity measures are normalized by gross total assets, since they measure the
lP
dollar amount of liquidity creation. 3.3 Activity strategies and control variables
ur na
Following prior literature (e.g., Stiroh and Rumble, 2006; De Jonghe, Diepstraten, and Schepens, 2015; Tran et al., 2019), I capture the activity strategies of banks using bank income structure, which is measured as the ratio of non-interest income to net operating income (NII). In robustness tests, I use alternative measures and still obtain similar evidence. To mitigate potential omitted variable bias, I control for various bank-specific variables,
Jo
including bank size (SIZE), capital ratio (CAPITAL), bank performance (EARNINGS), asset growth (GROWTH), different indicators of bank risk such as earnings volatility (SD(EARNINGS)), loan portfolio quality (NPL), and bank risk (ZSCORE). See Table 1 for the variable definitions and Table 2 for a descriptive summary.
See https://sites.google.com/a/tamu.edu/bouwman/data. See Berger and Bouwman (2009) for the measure’s explanation. 1
12
4. Do activity strategies affect bank liquidity creation? 4.1 Main findings In this section, I conduct multivariate analysis to formally investigate the magnitude of activity strategies on bank liquidity creation, after controlling for the other control variables. Specifically, the empirical specification I estimate is 𝑌𝑖𝑡 = 𝛼𝑖 + 𝑁𝐼𝐼𝑖𝑡−1 + 𝑍𝑖𝑡−1 + 𝜇𝑖 + 𝜃𝑡 + 𝜀𝑖𝑡
(1)
of
where 𝑌𝑖𝑡 is the measure of liquidity created by bank i at time t. Following Berger and Bouwman (2009), I use their preferred measure, catfat, as a proxy of bank liquidity creation that includes
ro
both on- and off-balance sheet items. As argued above, this measure classifies loans by category (cat) instead of maturity (mat) and gauges how much liquidity banks create off the balance sheet
-p
(fat). Let LC denote the catfat measure. The variable of interest is NII, which is the ratio of noninterest income to net operating income, and 𝑍𝑖𝑡 is the vector of control variables described
re
above.
In all specifications, I lag bank factors by one period to mitigate simultaneity and
lP
endogeneity problems. I include bank fixed effects, 𝜇𝑖 , motivated by the fact that differences in bank liquidity creation are partly related to unobservable but time-invariant characteristics of banks, such as corporate culture and bank management. The estimation of a separate intercept
ur na
(𝛼𝑖 ) for each bank before fitting the slope coefficients controls for unobservable, time-invariant sources of bank heterogeneity, allowing a focus on the variation in liquidity creation at the level of individual banks over time. I also include time fixed effects, 𝜃𝑡 , to control for time effects that can affect the liquidity creation of banks. The term 𝜀𝑖𝑡 is the error term. Since LC is likely to be correlated within a bank over time, standard errors used to assess significance are corrected for
Jo
heteroscedasticity and bank-level clustering.2 The main results from the multivariate analysis are shown in Table 3. Model (1) reports
the baseline model. The estimated constant indicates that the expected average LC value of
2
To avoid the problem of spurious regressions, the model should include all variables that are stationary. I run unit root tests, not tabulated here. If I can reject the null hypothesis of a unit root, the findings from these unit root tests will allow an investigation into the association between diversification and liquidity creation. If not, the estimated parameters could be biased, due to the spurious regression problem. I employ a panel-based unit root test, a Fishertype test (Choi 2001), to test the null of the unit root of all the variables. The untabulated results strongly suggest that all the variables are stationary. I thank an anonymous referee for suggesting this complementary test.
13
banks is 0.850 and statistically significant. The coefficient of the main variable of interest, NII, is negative and statistically significant at the 1% level, suggesting that moving toward non-interest income–generating activities reduces liquidity creation. The economic magnitude of this effect is also significant. A one standard deviation increase of NII, holding all others equal, results in a decrease in LC of 0.014 (i.e. the coefficient on NII, -0.102, times the standard deviation of NII, 0.13). With LC having a mean of 0.445 and a standard deviation of 0.186, this decrease is equivalent to 3.15% of the average bank’s assets. This evidence is consistent with the
of
diversification–liquidity contraction hypothesis (H1b), showing that liquidity creation is decreased in diversified banks. The evidence is also in line with the findings of Hou et al. (2018) for the Chinese banking system.
ro
In Model (2), I run the baseline model with a single cross-sectional regression (average analysis) to address the potential error dependence problem. By performing this time series mean
-p
regression (one observation per bank), I eliminate the problem of serially correlated errors. This estimation maintains heterogeneity across banks but does not exploit the time series variation in
re
the observations (Tran, Hassan, and Houston 2019). The findings are comparable to the earlier results.
lP
For Model (3), I report the results for a balanced panel (excluding banks that only existed during part of the sample period). This exclusion mitigates the effects of M&A activities and bank defaults on the investigation, although at the price of overrepresenting “successful” banks.
ur na
In Model (4), I exclude M&A banks (banks with a large change in assets over a period, where I use a threshold of 20% of assets changing over a quarter). The results show the coefficients of NII remain negative and statistically significant for all specifications. Next, I use alternative econometric methods, such as Newey–West (Model (5)), Fama– MacBeth (Model (6)), and two-way clustering by bank and time, to allow for correlations among
Jo
different banks in the same quarter and across quarters for the same bank (Model (7)). The results are still similar. In untabulated tests, instead of including all three risk measures (NPL, SD(EARNINGS),
ZSCORE) at one time, I rerun the baseline model by alternatively including these measures individually. I also use a lag of four periods instead of one period, including the state dummy, to take into account characteristics of the local environment. I also exclude the top largest banks
14
(top 10th percentile of bank assets, too big to fail banks). Again, I observe that diversified banks still experience lower liquidity creation than other banks. Regardless of the control variables (Model (1)), I observe that large and well capitalized banks are more likely to create less liquidity.3 I documented that highly profitable banks are more likely to create liquidity than other banks. The evidence also shows that banks with a high proportion of nonperforming loans experience lower liquidity creation than other banks. I find no evidence of growth and other risk proxies.
of
In brief, the findings support the diversification–liquidity contraction hypothesis (H1b), in that diversification will make banks more complex, heighten moral hazard problems, and disperse managerial resources and operating stability, leading to failure to satisfy clients’
ro
liquidity demands. The negative effects of dispersing managerial resources and strengthening moral hazard due to diversification dominate the bright side of the risk mitigation and economies
-p
of scale and scope from diversification.
re
4.2 Quantile regression
The main purpose of this study is to investigate the relation between banks’ business models and
lP
their liquidity creation. Regulators and policymakers seem to be more interested in bank behaviors at the tails of the distribution of liquidity creation, due to the link between liquidity creation and real economic output (Berger and Sedunov 2017).
ur na
Table 4 shows the results of quantile regression—a generalization of median regression analysis to other quantiles—to assess whether the interplay between diversification and liquidity creation differ across quantiles of liquidity creation. The traditional inference approach (i.e., OLS) used in earlier sections describes the average behavior of the sample, with the assumption of the homogeneity of the effects of diversification on bank liquidity creation. However, when
Jo
heterogeneity in the sample is important, the traditional OLS approach might not be ideal. Rather than relying on a single description of the central behavior of the sample, the quantile approach
3
In an untabulated test, I use alternative measures of bank size. First, since size is largely an outcome of bank decision making and highly correlated with other independent and dependent variables, I decompose bank size with respect to all the other independent variables into two components: an organic growth component that is measured by the fitted value and a historical size component that equals the residual. Orthogonalizing size allows the pure effects of size to be derived (De Jonghe 2010). Second, I also check for a nonlinear relation between earnings management and size by including size–decile fixed effects to control for unobserved heterogeneity across banks in different size categories, as suggested in Ellul and Yerramilli (2013). I obtain similar results.
15
explores a range of conditional quantile functions, which, in turn, allows potential forms of conditional heterogeneity to be explored and an understanding of the behavior of banks across the distribution of liquidity creation (Tran, Hassan, and Houston 2019). Furthermore, the quantile regression approach avoids the restrictive assumption that the error terms are identically distributed at different locations of bank liquidity creation (Klomp and Haan 2012). I report the relations between NII and LC across quantiles. The coefficients on NII in Models (1) to (9) demonstrate that the impact of diversification on bank liquidity creation is
of
indeed uniform in sign (negative), suggesting that NII decreases banks’ liquidity creation in banks of all levels of liquidity creation. I plot the coefficients for NII across the LC distribution quantiles in Figure (1) and observe that these coefficients do not vary significantly across the
ro
distribution of liquidity creation, except for the 10th quantile of LC. This finding implies that, for banks with low liquidity creation (10th quantile of LC), moving toward non-interest income–
-p
generating activities would decrease their liquidity creation.
Overall, the empirical findings indicate activity strategies not only affect the conditional
re
average liquidity creation but also influence the liquidity creation distribution.
lP
4.3 Net effect of diversification on liquidity creation for large, medium-sized, and small banks Table 5 contains the regression results for different size ranges, that is, small banks, with assets of up to $1 billion; medium-sized banks, with assets between $1 billion and $5 billion; and large
ur na
banks, with assets of over $5 billion. The results for all three models, Models (1) to (3), indicate that diversification is associated with lower liquidity creation across all size classes, but with different magnitudes. The magnitude of the coefficient on NII is -0.109, significant at 1%, for the sample of small banks, suggesting that a one percentage point higher level of diversification will decrease a small bank’s liquidity creation by 0.1% of its assets. This effect seems to be mitigated
Jo
in the sample of medium-sized banks, where the coefficient on NII is -0.079, significant at 5%. For the sample of large banks, a one percentage point higher level of diversification will decrease a large bank’s liquidity creation by 0.15% of its assets, which appears to be a substantial effect. Potential explanations are as follows. Larger banks are more likely to gain from nontraditional activities for a long time and to then push their diversification levels to saturation, no longer gaining benefits from marginal increases in diversification but potentially incurring greater agency problems. This suggests a large reduction in the liquidity creation of large banks. Despite 16
having more to gain from diversification, small banks can lack experience in new and more complex nontraditional activities, thereby reducing their operating stability and liquidity creation. In summary, I document a variation of the effect of diversification on bank liquidity creation across bank sizes. Large banks that engage in non-interest income–generating activities reduce their liquidity creation the most, followed by small diversified banks. Medium-sized banks reduce their liquidity creation the least when increasing their share of non-interest income.
of
The evidence supports H2. 4.4 Effects of diversification on liquidity creation during banking crises
ro
In this section, I investigate whether the association between NII and LC changes during crisis periods. This study starts in 2001:Q1 and includes the last crisis, from 2007:Q3 to 2009:Q2,
-p
following Acharya and Mora (2015). I rerun the baseline model by adding the crisis dummy and its interaction with NII (NII*CRISIS). The focus is the coefficient of the interaction term, which
The results are reported in Table 6.
re
measures how the crisis affects the association between diversification and liquidity creation.
lP
Model (1) in Table 6 shows that, during times of crisis, banks create less liquidity than in normal times, as indicated by the negative and statistically significant coefficient on CRISIS. The interaction term NII*CRISIS is positive and statistically significant, indicating that
ur na
diversified banks created more (less negative) liquidity than other banks during the crisis than during normal circumstances. The results support H3a, that is, banks that move toward nontraditional activities create more liquidity during crisis times. The evidence points out the bright side of diversification, which can induce banks to take on less risk during times of turmoil, as documented recently by DeYoung and Torna (2013) and Tran et al. (2019), giving rise to
Jo
greater liquidity creation.
However, differences in deposit flows could exist between the early and late stages of the
crisis. Indeed, there is deposit funding pressure in the first phase of the crisis, starting August 9, 2007, due to the freezing of asset-backed commercial paper markets. This reflects investors’ perception of the greater risk of bank deposits relative to other instruments offering similar liquidity and payment services (Acharya and Mora 2015). The situation changed when the government explicitly backed the depository system through an increase in deposit insurance to 17
$250,000, among other measures. These factors could have also affected banks’ ability to create liquidity. I therefore divide the crisis period into two subperiods: CRISIS 1, from 2007Q3 to 2008Q2, and CRISIS 2, from 2008Q3 to 2009Q2. I rerun the analysis separately for each stage of the crisis. The results in Models (5) and (9) show that the results reported in Model (1) are mostly driven by the second stage of the crisis (CRISIS 2). I further investigate the effects of the crisis on the impacts of diversification on liquidity creation across bank size classes. The coefficients on NII in Models (2) to (4) are negative and
of
statistically significant, indicating that, during normal times, diversification will lead to lower liquidity creation across bank size classes. The coefficients on NII*CRISIS are positive but not
creation across bank size classes during the crisis period.
ro
statistically significant, implying no difference in the effects of diversification on liquidity
When the crisis is split into two subperiods, I find no evidence during the early stage of
-p
the crisis (CRISIS 1), but there is evidence during the late stage of the crisis (CRISIS 2). In summary, the results shown in Table 6 show evidence of the bright side of
re
diversification during a crisis; that is, diversified banks create more liquidity during crisis times. This effect can be clearly observed among medium-sized and small banks during the late stage of
5. Robustness checks
lP
the crisis.
ur na
5.1 Alternative measures of liquidity creation
In Table 7, Panel A, I rerun the baseline model with alternative measures of bank liquidity creation. In Model (1), inspired by Foos, Norden, and Weber (2010) and Tran, Hassan, and Reza (2019), I use the deviation of LC of bank i at time t from the industry average at time t as a measure of bank liquidity creation.
Jo
In Models (2) and (3), I divide the LC measure into on- and off-balance sheet
components (LC_ON and LC_OFF, respectively) to investigate the effect of diversification on different types of liquidity creation activities. The results show that the negative association between diversification and liquidity creation is mainly driven by the on-balance sheet component of liquidity creation.
18
5.2 Alternative measures of diversification In Table 7, Panel B, I rerun the baseline model with alternative measures of bank diversification. In Model (1), I first rank the NII variable by quartiles and create the variable NII_QUARTILE, which takes a value ranging from one (low) to four (high). This approach generates greater variation in the distribution of the extent of diversification. I next focus on the most controversial type of non-interest income, that is, the trading income in Model (2). I compute trading income following Gandhi, Kiefer, and Plazzi (2016),
of
who suggest totaling trading revenue, interest income from trading assets, and gains or losses realized from held-to-maturity and available-for-sale securities. Using this measure mitigates
ro
concerns of non-interest income including income derived from traditional activities. I use the ratio of loans over total assets as the proxy for a bank’s functional
-p
diversification (LOAN) in Model (3). In contrast to other measures, this measure is interpreted as an inverse diversification measure, since the higher value of loan ratio means that banks focus more on traditional banking activities.
re
The results shown in Models (1) to (3) in Panel B of Table 7 confirm earlier findings, suggesting that a move toward nontraditional activities will make banks create less liquidity.
lP
Following Stiroh and Rumble (2006), in Model (4), I use an adjusted Herfindahl– Hirschman index to measure diversification (HHI), which accounts for variations in the breakdown of net operating income (NOI) into the two main categories of net interest income
ur na
(NIM) and non-interest income (NII). The findings in Model (4) highlight that an increase of diversification leads to a reduction in bank liquidity creation: 𝐻𝐻𝐼 = 1 − [(𝑁𝐼𝑀)2 + (𝑁𝐼𝐼)2 ]
(2)
Jo
For each appropriate value of HHI, there are two possible values of NII, each reflecting
different specific underlying activities of banks (with greater reliance on either NIM or NII) and each potentially having different effects on LC. Then, in Model (5), I simultaneously include HHI and NII to account for the potential impacts of the heterogeneity of activities on LC. I find a positive coefficient for HHI and a negative coefficient for NII, both statistically significant. Economically, holding all other variables constant, an increase of HHI from zero to its mean value (i.e., 0.325) enhances LC by about 0.029 (= 0.088*0.325, i.e., from 0.456 to 0.485). Banks 19
experience a decrease in LC of about 0.033 (= -0.140*0.234), from 0.456 to 0.423, when moving from pure lending activities toward an activity generating $0.234 of NII per $1 of NOI. It is worth noting that, since HHI is a quadratic function of NII, it is clear that the effects of a variation in NII can be disentangled into a direct exposure effect (i.e., the coefficient on NII) and an indirect exposure effect (i.e., the coefficient on HHI times the first derivatives of HHI on NII). I then calculate the net effect by evaluating at the 10th, 25th, 50th, 75th, and 90th percentiles of NII the direct, indirect, and net effects on changes in NII on LC. The results are
of
shown in Panel C of Table 7. The net effect, which combines both effects, offers interesting evidence. For banks with a heavily concentrated interest income (at the 10th and 20th percentiles of NII), the diversification gains do not differentiate from the negative direct effects of an
ro
increase in NII, suggesting that the direct and indirect exposure effects on LC come close to
canceling each other out. The net effect is not statistically different from zero. However, the net
-p
effect becomes statistically negative with greater reliance on NII as the indirect positive effects progressively decrease. The net effect becomes significantly negative after the 50th percentile of
re
NII. At the 90th percentile of NII, one standard deviation in NII (i.e. 0.132) induces to a jump of -0.013 in LC, which corresponds to 2.8% of the mean of LC.
lP
In summary, the results with alternative measures of diversification confirm earlier findings, suggesting that a move toward nontraditional activities would make banks create less liquidity. I also find that banks benefit from diversification, but these gains are quickly offset by
ur na
the adverse effects from reliance on riskier assets. Banks with a relatively small share of noninterest income do not seem to change their liquidity creation when relying more on non-interest income–generating activities. However, with greater reliance on non-interest income–generating activities, banks reduce their liquidity creation with an increase in the share of non-interest
Jo
income.
5.3 Endogeneity concerns This investigation could be subject to reverse causality or endogeneity among the variables. The specification in the baseline model is based on the assumption that a bank’s decision to diversify is exogenous. However, diversification is not random, but a deliberate decision made by bank managers; hence, the same bank-level characteristics that drive the decision to diversify could also affect the liquidity creation of banks. Failure to control for factors that drive banks to 20
diversify leads to biased econometric results that inappropriately attribute the diversification costs and benefits to diversification per se, rather than to the underlying traits that lead banks to diversify (Campa and Kedia 2002). I therefore the OLS estimation with different approaches: the Heckman selection model, the IV approach, and PSM. These procedures should control for any potential selection bias in the above estimation. The results are tabulated in Table 8. First, I employ the PSM method developed by Rosenbaum and Rubin (1983) and
of
extended by Heckman et al. (1997). To conduct PSM, I separate the full sample into two groups: diversified (treated) and focused (untreated) banks. Following Laeven and Levine (2007), I
classify banks into two separate groups: (i) Banks with a share of net interest income between
ro
10% and 90% are classified as diversified, whereas (ii) banks with a share of net interest income either below 10% or above 90% are classified as focused (or specialized). I measure the
-p
propensity of undergoing treatment (i.e., the probability of diversification) by using a logit model for both treated and untreated samples. The dependent variable in this logit model is a binary
re
variable that equals one if the bank is classified as diversified, and zero otherwise. The logit model is similar to that in the first stage of the Heckman selection model described below. I
lP
match each diversified bank with one focused bank that shares similar characteristics, as reflected in their propensity scores. I retain only untreated observations whose propensity scores fall inside the interval defined for the treated group. I impose a tolerance level of 0.5% on the
ur na
maximum propensity score distance allowed (caliper) to minimize the risk of bad matches. I use one-to-one matching without replacement, which requires each focused bank to be used exactly once. The result in Model (1) confirms the earlier finding, that is, banks create less liquidity when engaging in nontraditional activities. The matching estimator presented above mitigates selection bias. However, there could
Jo
be unobservable factors that explain decisions to diversify. I use Heckman’s two-step approach to eliminate bias due to unobservable variables. I first model the selection of diversification by using the logit selection model and then obtain the inverse Mills ratio (IMR), the omitted variable in Equation (1). Following Laeven and Levine (2007), I use the average non-interest income of other banks as an IV. I then estimate the logit diversification choice model and calculate IMR. The IMR measure is the conditional expectation of the model selection error term, given banks’ observable characteristics and decision to diversify. In the second stage, I re21
estimate Equation (1) by including IMR as an additional control variable to correct for potential self-selection bias. Models (2) and (3) in Table 8 present the maximum likelihood estimates of the logit diversification choice and baseline model augmented by IMR. Consistent with the core findings, I still document a negative and significant coefficient of NII. I continue the inquiry into the endogeneity of the diversification decision with an IV estimation. As above, the instrument is the average non-interest income of other banks. I report the first- and second-stage IV regression results in Models (4) and (5) of Table 8, respectively. The Kleibergen–Paap underidentification and Cragg–Donald weak identification test statistics
of
further indicate that the selected IV is relevant.4 The results of the second stage also support the earlier finding. I observe that the coefficient in the IV estimation is much larger than in the OLS
ro
estimate (-0.621 vs. -0.106, both significant at 1%), which is consistent with concerns regarding reverse causality and, hence, with the need to use an IV approach to identify the impact of
-p
diversification on bank liquidity creation (Tran, Hassan, and Houston 2019). The OLS estimation could yield coefficient estimates of the impact of NII on LC that are biased toward zero, whereas
re
the IV estimation yields a more accurate (and larger) impact of NII on LC. It is worth noting that, in Models (2) and (4), the IV is positively and significantly associated with the decision to
lP
diversify. This confirms the relevance of the IV.
I end the examination with an autoregressive distributed lag model using system generalized method of moments (system-GMM) estimators to address the persistence of the
ur na
endogeneity bias.5 The regression also includes time fixed effects, following Roodman (2009). The results are reported in Table 8, Panel B. I first discuss the diagnostic tests. The Hansen test (chi-squared test) of overidentification checks the validity of the instruments (i.e., exogeneity). The p-value of the Hansen test suggests that the null hypotheses of overidentification cannot be rejected, that is, the instruments are valid. The results of tests for autocorrelation in the residuals
Jo
AR(1) and AR(2) are also reported (the null hypothesis being no serial correlation of order 1 or 2 in the first-difference residuals), and the residuals show no evidence of second-order autocorrelation. I still find a negative and statistically significant coefficient on NII, the variable of interest, consistent with the previous findings.
4 5
I did not perform the overidentification test, since my model is only identified. I am grateful to an anonymous referee for this suggestion.
22
In brief, whether estimating a Heckman selection model, employing an IV approach, using PSM, or using system-GMM, the findings remain unchanged: diversification per se lowers bank liquidity creation. 6. Conclusions This study investigates the impacts of functional diversification on banks’ liquidity creation, using a large sample of US banks from 2001:Q1 to 2015:Q4. The basic regressions suggest that banks with an increased share of non-interest income decrease their liquidity creation, which is
of
consistent with the diversification–liquidity contraction channel. Banks benefit from
diversification, but these gains are quickly offset by adverse effects from reliance on riskier
ro
assets. Banks with a relatively small share of non-interest income do not seem to change their liquidity creation when relying more on non-interest income–generating activities. However,
-p
with greater reliance on non-interest income–generating activities, banks reduce their liquidity creation with increases in the share of non-interest income. The empirical findings indicate
re
activity strategies not only affect the conditional average of liquidity creation, but also influence the distribution of liquidity creation. When performing analysis across bank sizes, I observe that large banks that engage in non-interest income–generating activities reduce their liquidity
lP
creation the most, followed by small diversified banks. Medium-sized banks reduce their liquidity creation the least when they increase their share of non-interest income. Finally, I
ur na
document evidence of the bright side of diversification during a crisis. Diversified banks create more liquidity during times of crisis. This effect can be clearly observed among medium-sized and small banks during the late stage of crisis. From a policy perspective, I believe that this study is of interest to regulators and policymakers. After the last financial crisis, various initiatives, such as the Volcker Rule in the
Jo
United States, Vickers Initiatives in the United Kingdom, the Liikanen Report in the European Union, and, recently, the call for the 21st Century Glass–Steagall Act propose narrow banking policies that aim to diminish banks’ ability to diversify across product lines (Tran et al. 2019). The evidence shows that, during the normal times, diversified banks, especially large ones, create less liquidity. However, during crises, there could be a bright side, allowing medium-sized and small banks to expand into non-interest–generating activities, since banks could become less risky during times of turmoil, leading to a rise in liquidity creation and thereby contributing to the recovery of the real economy. Taken together, this study suggests that ring-fencing initiatives 23
could help mitigate the negative effects of diversification on liquidity creation during normal
Jo
ur na
lP
re
-p
ro
of
times, but at the price of lowering the benefits of diversification in times of crisis.
24
References Acharya, Viral V., and Nada Mora, 2015, A Crisis of Banks as Liquidity Providers, The Journal of Finance 70, 1–43. Acharya, Viral V., Iftekhar Hasan, and Anthony Saunders, 2006, Should Banks Be Diversified? Evidence from Individual Bank Loan Portfolios, The Journal of Business 79, 1355–1412. Amihud, Yakov, and Baruch Lev, 1981, Risk reduction as a managerial motive for conglomerate mergers, The bell journal of economics, 605–617.
of
Andreou, Panayiotis C., Dennis Philip, and Peter Robejsek, 2016, Bank Liquidity Creation and Risk-Taking: Does Managerial Ability Matter?, Journal of Business Finance & Accounting 43, 226–259.
ro
Berger, Allen N., and Christa H. S. Bouwman, 2013, How does capital affect bank performance during financial crises?, Journal of Financial Economics 109, 146–176.
-p
Berger, Allen N., and Christa H. S. Bouwman, 2017, Bank liquidity creation, monetary policy, and financial crises, Journal of Financial Stability 30, 139–155.
re
Berger, Allen N., Christa H. S. Bouwman, Thomas Kick, and Klaus Schaeck, 2016, Bank liquidity creation following regulatory interventions and capital support, Journal of Financial Intermediation 26, 115–141.
lP
Berger, Allen N., and Christa HS Bouwman, 2009, Bank liquidity creation, Review of Financial Studies 22, 3779–3837.
ur na
Berger, Allen N., Iftekhar Hasan, and Mingming Zhou, 2010, The effects of focus versus diversification on bank performance: Evidence from Chinese banks, Journal of Banking & Finance 34. Performance Measurement in the Financial Services Sector, 1417–1435. Berger, Allen N., and John Sedunov, 2017, Bank liquidity creation and real economic output, Journal of Banking & Finance 81, 1–19.
Jo
Berger, Philip G., and Eli Ofek, 1995, Diversification’s effect on firm value, Journal of financial economics 37, 39–65. Bharath, Sreedhar, Sandeep Dahiya, Anthony Saunders, and Anand Srinivasan, 2007, So what do I get? The bank’s view of lending relationships, Journal of Financial Economics 85. The economics of conflicts of interest financial institutions, 368–419. Bhattacharya, Sudipto, and Anjan V. Thakor, 1993, Contemporary banking theory, Journal of financial Intermediation 3, 2–50. Brewer, Elijah, 1989, Relationship between bank holding company risk and nonbank activity, Journal of Economics and Business 41, 337–353. 25
Bryant, John, 1980, A model of reserves, bank runs, and deposit insurance, Journal of Banking & Finance 4, 335–344. Campa, Jose Manuel, and Simi Kedia, 2002, Explaining the Diversification Discount, The Journal of Finance 57, 1731–1762. Campello, Murillo, 2002, Internal Capital Markets in Financial Conglomerates: Evidence from Small Bank Responses to Monetary Policy, The Journal of Finance 57, 2773–2805. Choi, In, 2001, Unit root tests for panel data, Journal of International Money and Finance 20, 249–272.
ro
of
De Jonghe, Olivier, 2010, Back to the basics in banking? A micro-analysis of banking system stability, Journal of Financial Intermediation 19. Risk Transfer Mechanisms and Financial Stability, 387–417.
-p
De Jonghe, Olivier, Maaike Diepstraten, and Glenn Schepens, 2015, Banks’ size, scope and systemic risk: What role for conflicts of interest?, Journal of Banking & Finance 61, Supplement 1. Global Trends in Banking, Regulations, and Financial Markets, S3–S13.
re
Demsetz, Rebecca S., and Philip E. Strahan, 1997, Diversification, size, and risk at bank holding companies, Journal of money, credit, and banking, 300–313.
lP
Denis, David J., Diane K. Denis, and Atulya Sarin, 1997, Agency Problems, Equity Ownership, and Corporate Diversification, Journal of Finance 52, 135–160. DeYoung, Robert, and Karin P. Roland, 2001, Product Mix and Earnings Volatility at Commercial Banks: Evidence from a Degree of Total Leverage Model, Journal of Financial Intermediation 10, 54–84.
ur na
DeYoung, Robert, and Gökhan Torna, 2013, Nontraditional banking activities and bank failures during the financial crisis, Journal of Financial Intermediation 22, 397–421. Diamond, Douglas W., 1984, Financial intermediation and delegated monitoring, The Review of Economic Studies 51, 393–414.
Jo
Diamond, Douglas W., and Philip H. Dybvig, 1983, Bank runs, deposit insurance, and liquidity, The journal of political economy, 401–419. Diamond, Douglas W., and Raghuram G. Rajan, 2000, A theory of bank capital, The Journal of Finance 55, 2431–2465. Diamond, Douglas W., and Raghuram G. Rajan, 2001, Liquidity Risk, Liquidity Creation, and Financial Fragility: A Theory of Banking, Journal of Political Economy 109, 287. Distinguin, Isabelle, Caroline Roulet, and Amine Tarazi, 2013, Bank regulatory capital and liquidity: Evidence from US and European publicly traded banks, Journal of Banking & Finance 37, 3295–3317. 26
Ellul, Andrew, and Vijay Yerramilli, 2013, Stronger Risk Controls, Lower Risk: Evidence from U.S. Bank Holding Companies, The Journal of Finance 68, 1757–1803. Elsas, Ralf, Andreas Hackethal, and Markus Holzhäuser, 2010, The anatomy of bank diversification, Journal of Banking & Finance 34, 1274–1287. Financial Crisis Inquiry Commission, 2011, The Financial Crisis Inquiry Report (Cosimo, Inc.). Foos, Daniel, Lars Norden, and Martin Weber, 2010, Loan growth and riskiness of banks, Journal of Banking & Finance 34. International Financial Integration, 2929–2940.
of
Gandhi, Priyank, Patrick Christian Kiefer, and Alberto Plazzi, 2016, A False Sense of Security: Why U.S. Banks Diversify and Does it Help? SSRN Scholarly Paper, Social Science Research Network, Rochester, NY.
ro
Heckman, James J., Hidehiko Ichimura, and Petra E. Todd, 1997, Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme, The review of economic studies 64, 605–654.
-p
Holmström, Bengt, and Jean Tirole, 1998, Private and public supply of liquidity, Journal of political Economy 106, 1–40.
re
Horváth, Roman, Jakub Seidler, and Laurent Weill, 2013, Bank Capital and Liquidity Creation: Granger-Causality Evidence, Journal of Financial Services Research 45, 341–361.
lP
Horvath, Roman, Jakub Seidler, and Laurent Weill, 2016, How bank competition influences liquidity creation, Economic Modelling 52. Special Issue on Recent Developments in DecisionMaking, Monetary Policy and Financial Markets, 155–161.
ur na
Hou, Xiaohui, Shuo Li, Wanli Li, and Qing Wang, 2018, Bank diversification and liquidity creation: Panel Granger-causality evidence from China, Economic Modelling 71, 87–98. Houston, Joel, Christopher James, and David Marcus, 1997, Capital market frictions and the role of internal capital markets in banking, Journal of Financial Economics 46, 135–164. Jensen, Michael C., 1986, Agency costs of free cash flow, corporate finance, and takeovers, The American economic review, 323–329.
Jo
Jensen, Michael C., and Kevin J. Murphy, 1990, Performance pay and top-management incentives, Journal of political economy, 225–264. Jiang, Liangliang, Ross Levine, and Chen Lin, 2017, Competition and Bank Liquidity Creation. SSRN Scholarly Paper, Social Science Research Network, Rochester, NY. Jiménez, Gabriel, and Jesus Saurina, 2004, Collateral, type of lender and relationship banking as determinants of credit risk, Journal of banking & Finance 28, 2191–2212.
27
Kashyap, Anil K., Raghuram Rajan, and Jeremy C. Stein, 2002, Banks as Liquidity Providers: An Explanation for the Coexistence of Lending and Deposit-taking, The Journal of Finance 57, 33–73. Kashyap, Anil K, and Jeremy C. Stein, 2000, What Do a Million Observations on Banks Say about the Transmission of Monetary Policy?, The American Economic Review 90, 407–428. Klomp, Jeroen, and Jakob de Haan, 2012, Banking risk and regulation: Does one size fit all?, Journal of Banking & Finance 36. Systemic risk, Basel III, global financial stability and regulation, 3197–3212.
of
Krishnaswami, Sudha, Paul A Spindt, and Venkat Subramaniam, 1999, Information asymmetry, monitoring, and the placement structure of corporate debt, Journal of Financial Economics 51, 407–434.
ro
Kwast, Myron L., 1989, The impact of underwriting and dealing on bank returns and risks, Journal of Banking & Finance 13, 101–125.
-p
Laeven, Luc, and Ross Levine, 2007, Is there a diversification discount in financial conglomerates?, Journal of Financial Economics 85, 331–367.
re
Levine, Ross, and Sara Zervos, 1998, Stock Markets, Banks, and Economic Growth, The American Economic Review 88, 537–558.
lP
Litan, Robert, 1985, Evaluating and Controlling the Risks of Financial Product Deregulation, Yale Journal on Regulation 3. Roodman, David, 2009, How to do Xtabond2: An Introduction to Difference and System GMM in Stata, The Stata Journal 9, 86–136.
ur na
Rosenbaum, Paul R., and Donald B. Rubin, 1983, The central role of the propensity score in observational studies for causal effects, Biometrika 70, 41–55. Santomero, Anthony M., and David L. Eckles, 2000, The Determinants of Success in the New Financial Services Environment: Now that Firms Can Do Everything, What Should They Do and Why Should Regulators Care? SSRN Scholarly Paper, Social Science Research Network, Rochester, NY.
Jo
Sanya, Sarah, and Simon Wolfe, 2011, Can Banks in Emerging Economies Benefit from Revenue Diversification?, Journal of Financial Services Research 40, 79–101. Saunders, Anthony, and Marcia Millon Cornett, 2008, Financial Institutions Management: A Risk Management Approach (McGraw-Hill Education). Saunders, Anthony, and Ingo Walter, 1994, Universal Banking in the United States: What Could We Gain? What Could We Lose? (Oxford University Press).
28
Shim, Jeungbo, 2013, Bank capital buffer and portfolio risk: The influence of business cycle and revenue diversification, Journal of Banking & Finance 37, 761–772. Shleifer, Andrei, and Robert W. Vishny, 1989, Management entrenchment: The case of manager-specific investments, Journal of financial economics 25, 123–139. Shleifer, Andrei, and Robert W. Vishny, 2010, Unstable banking, Journal of Financial Economics 97. The 2007-8 financial crisis: Lessons from corporate finance, 306–318. Stiroh, Kevin J., 2004, Diversification in Banking: Is Noninterest Income the Answer?, Journal of Money, Credit and Banking 36, 853–882.
of
Stiroh, Kevin J., and Adrienne Rumble, 2006, The dark side of diversification: The case of US financial holding companies, Journal of Banking & Finance 30, 2131–2161.
ro
Tran, Dung V., Isabelle Girerd-Potin, Pascal Louvet, and Kabir M. Hassan, 2017, Activity strategies, agency problems and bank risk. SSRN Scholarly Paper, Rochester, NY.
-p
Tran, Dung Viet, M.K Hassan, and Reza Houston, 2019, How does listing status affect bank risk? The effects of crisis, market discipline and regulatory pressure on listed and unlisted BHCs, The North American Journal of Economics and Finance 49, 85–103.
Jo
ur na
lP
re
Yasuda, Ayako, 2005, Do Bank Relationships Affect the Firm’s Underwriter Choice in the Corporate-Bond Underwriting Market?, The Journal of Finance 60, 1259–1292.
29
.1
.2
.3
.4
.5
.6
.7
Jo
ur na
lP
re
Quantile
-p
-0.35
ro
-0.30
of
NII
-0.25
-0.20
Figure 1. Plots of coefficients for non-interest income across quantiles based on quantile regression results in Table (4)
30
.8
.9
Table 1. Variables Definitions This table presents definitions of all main variables used in the analysis. Definitions
LOAN TRADING DIV SIZE
Book value of equity over gross total assets Income before taxes, provisions recognized in income over gross total assets
Growth rate of gross total assets
Standard deviation of EARNINGS over the previous 12 quarters
Nonperforming assets over the quarter, scaled by total loans 𝐶𝐴𝑃+𝜇𝑅𝑂𝐴 A bank measure of financial risk calculated as ; a larger value 𝜎𝑅𝑂𝐴
indicates lower overall bank risk; means of ROA and Equity/GTA as well as the standard deviation of ROA are computed over the previous 12 quarters (t 11 to t) Average of NII of all other banks A dummy equal to 1 for a period from 2007:Q3-2009:Q2, and 0 otherwise. A dummy equal to 1 for a period from 2007:Q3-2008:Q2, and 0 otherwise. A dummy equal to 1 for a period from 2008:Q3-2009:Q2, and 0 otherwise. Bank fixed effects Time fixed effects, represented by dummies for each quarter of the sample period.
Jo
NII_AVG CRISIS CRISIS_1 CRISIS_2 BFE TFE
Non-interest incomes over the net operating incomes In each quarter, I rank NII variable into quartiles and create a variable called NII_QUARTILE, which takes value ranging from 1 (low) to 4 (high) Loans over gross total assets Including trading revenues, interest income from trading assets and the realized gains or losses from the held-to-maturity and available-for-sale securities 𝐷𝐼𝑉_1 = 1 − [(𝑁𝐼𝐼)2 + (𝑁𝐼𝑀)2 ] The natural logarithm of gross total assets
ur na
CAPITAL EARNINGS GROWTH SD(EARNINGS) NPL ZSCORE
of
Control variables NII NII_QUARTILE
ro
LC_ON
-p
Deviation LC LC_OFF
A bank’s total bank liquidity creation measure including on- and off-balance sheet activities normalized by the gross total assets (GTA) of a bank. For a more detailed definition, see Berger and Bouwman (2009). LCi,t – Average LCi,t of the industry A bank’s bank liquidity creation measure including only off-balance sheet activities normalized by the total asset size of a bank. For a more detailed definition, see to Berger and Bouwman (2009). A bank’s bank liquidity creation measure including only on-balance sheet activities normalized by the total asset size of a bank. For a more detailed definition, see to Berger and Bouwman (2009).
re
Dependent variables LC
lP
Variables
31
Table 2. Summary Statistics
Panel B: Correlation SIZE
1.000 0.3728*** 0.0762*** 0.1514*** 0.0163*** -0.0106** -0.0615*** 0.1586***
1.000 0.0550*** 0.1142*** 0.0436*** 0.0617*** 0.0461*** 0.1086***
ro
CAP
lP
NII
ur na
LC 1.000 -0.1055*** 0.1380*** -0.1915*** 0.1299*** 0.1214*** -0.1157*** -0.0200*** -0.0924***
1.000 0.3575*** -0.0280*** '-0.1311*** 0.2277*** 0.0334***
Min (0.027) 0.000 12.089 0.019 (0.020) (0.085) 0.465 0.001
Max 0.899 0.814 19.109 0.220 0.051 0.229 0.118 197.907 0.047
EARNINGS
GROWTH
NPL
ZSCORE
SD(EARNINGS)
1.000 0.1515*** -0.3612*** 0.2575*** -0.1770***
1.000 -0.2236*** 0.1078*** -0.1157***
1.000 -0.3868*** 0.3532***
1.000 -0.4209***
1.000
Jo
LC NII SIZE CAP EARNINGS GROWTH NPL ZSCORE SD(EARNINGS)
Std. Dev. 0.174 0.131 1.237 0.028 0.009 0.042 0.022 37.935 0.007
-p
Mean 0.456 0.235 13.934 0.090 0.015 0.016 0.018 43.508 0.005
re
Obs 34,941 34,309 34,941 34,941 34,941 34,941 34,941 34,941 34,941
LC NII SIZE CAP EARNINGS GROWTH NPL ZSCORE SD(EARNINGS)
of
This table reports summary statistics and correlation for the main sample of U.S. commercial banks used in the analysis. The sample period is from 2001:Q1 to 2015:Q4. All financial variables are winsorized at 1% and 99% levels. Panel A: Summary statistics
32
Table 3. Baseline Multivariate Analysis
EARNINGS GROWTH NPL ZSCORE
BFE TFE Obs Adj R2 # Banks
-0.108*** (0.025) -0.029** (0.013) -0.264** (0.124) 2.096*** (0.190) 0.002 (0.014) -0.669*** (0.119) 0.000 (0.000) -0.479 (0.303) 0.890*** (0.186) Yes Yes 32,600 0.239 1,767
Jo
Constant
ur na
SD(EARNINGS)
-0.158*** (0.061) 0.021 (0.029) 0.015 (0.284) 3.012*** (0.476) 0.005 (0.027) -0.362 (0.315) -0.000 (0.000) 0.471 (0.509) 0.180 (0.426) Yes Yes 8,469 0.211 210
-p
CAPITAL
-0.206*** (0.036) 0.033*** (0.003) -2.174*** (0.152) 4.779*** (0.660) 2.792*** (0.244) 0.847*** (0.307) -0.001*** (0.000) 1.822** (0.875) 0.100** (0.044) Yes Yes 1,829 0.260
re
SIZE
-0.106*** (0.024) -0.026* (0.013) -0.274** (0.122) 2.086*** (0.189) 0.003 (0.015) -0.664*** (0.119) 0.000 (0.000) -0.483 (0.304) 0.850*** (0.184) Yes Yes 32,946 0.235 1,767
lP
NII
ro
of
This table reports regression estimates of the relation between NII and LC. The sample period is from 2001:Q1 to 2015:Q4. The main independent variable is NII. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics. Baseline model Average analysis Balanced data Excluded M&A Newey-West Fama-MacBeth Two-way Clustering (1) (2) (3) (4) (5) (6) (7)
33
-0.273*** (0.015) 0.028*** (0.001) -1.626*** (0.062) 4.473*** (0.188) 0.312*** (0.025) -0.060 (0.084) -0.000*** (0.000) -0.507* (0.263) 0.211*** (0.020) Yes Yes 32,946 0.146
-0.263*** (0.018) 0.025*** (0.001) -1.350*** (0.116) 4.198*** (0.626) 0.160** (0.065) -0.450** (0.169) -0.001*** (0.000) -0.103 (0.431) 0.212*** (0.016) Yes Yes 32,946 0.159
-0.273*** (0.042) 0.028*** (0.004) -1.626*** (0.179) 4.473*** (0.554) 0.312*** (0.055) -0.060 (0.199) -0.000*** (0.000) -0.507 (0.626) 0.211*** (0.057) Yes Yes 32,946 0.146
Table 4. Quantile regression
GROWTH NPL SD(EARNINGS) ZSCORE Constant
Jo
Observations Pseudo R2
34
0.6 (6) -0.240*** (0.008) 0.035*** (0.001) -1.533*** (0.051) 4.081*** (0.136) 0.271*** (0.029) -1.089*** (0.046) -1.182*** (0.134) -0.000*** (0.000) 0.192*** (0.010) 34,309 0.088
ro
0.5 (5) -0.242*** (0.009) 0.035*** (0.001) -1.547*** (0.059) 3.858*** (0.164) 0.276*** (0.031) -1.005*** (0.070) -1.216*** (0.158) -0.001*** (0.000) 0.156*** (0.012) 34,309 0.083
-p
EARNINGS
0.4 (4) -0.247*** (0.012) 0.035*** (0.001) -1.637*** (0.053) 3.853*** (0.166) 0.248*** (0.034) -0.898*** (0.070) -1.449*** (0.188) -0.001*** (0.000) 0.132*** (0.015) 34,309 0.08
re
CAP
0.3 (3) -0.246*** (0.011) 0.034*** (0.001) -1.627*** (0.058) 3.303*** (0.218) 0.258*** (0.031) -0.685*** (0.087) -2.123*** (0.224) -0.001*** (0.000) 0.107*** (0.018) 34,309 0.078
lP
SIZE
0.2 (2) -0.249*** (0.013) 0.033*** (0.002) -1.663*** (0.049) 3.309*** (0.260) 0.237*** (0.036) -0.418*** (0.071) -1.985*** (0.179) -0.001*** (0.000) 0.065*** (0.020) 34,309 0.080
ur na
NII
Quantile 0.1 (1) -0.306*** (0.019) 0.027*** (0.003) -1.798*** (0.057) 3.812*** (0.315) 0.183*** (0.050) 0.033 (0.099) -1.149*** (0.343) -0.000*** (0.000) 0.064* (0.036) 34,309 0.083
of
This table reports regression estimates of the relation between NII and LC using quantile regressions. The sample period is from 2001:Q1 to 2015:Q4. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Numbers in parentheses are t-statistics.
0.7 (7) -0.243*** (0.009) 0.034*** (0.001) -1.448*** (0.047) 4.177*** (0.118) 0.275*** (0.043) -1.169*** (0.058) -0.940*** (0.177) -0.000*** (0.000) 0.234*** (0.012) 34,309 0.094
0.8 (8) -0.242*** (0.008) 0.032*** (0.001) -1.380*** (0.041) 4.458*** (0.146) 0.311*** (0.037) -1.280*** (0.055) -0.275 (0.267) -0.000*** (0.000) 0.294*** (0.011) 34,309 0.099
0.9 (9) -0.250*** (0.014) 0.030*** (0.001) -1.280*** (0.063) 4.893*** (0.147) 0.332*** (0.040) -1.612*** (0.065) 1.167*** (0.334) -0.001*** (0.000) 0.370*** (0.018) 34,309 0.103
Table 5. The effect of diversification on liquidity creation for different size range This table reports regression estimates of the relation between NII and LC. The sample period is from 2001:Q1 to 2015:Q4. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics. Small Medium Large (1) (2) (3)
EARNINGS GROWTH NPL ZSCORE SD(EARNINGS) Constant
Jo
ur na
lP
re
BFE TFE Obs Adj R2 # Banks
35
-0.148*** (0.056) -0.022 (0.055) 1.285*** (0.454) 1.894*** (0.424) 0.043 (0.032) -0.250 (0.370) 0.000 (0.000) -0.299 (0.650) 0.773 (0.916) Yes Yes 3,511 0.274 154
of
CAPITAL
-0.079** (0.038) 0.011 (0.020) 0.040 (0.193) 1.928*** (0.303) -0.018 (0.023) -0.440** (0.210) -0.000 (0.000) -0.164 (0.433) 0.332 (0.285) Yes Yes 11,103 0.224 530
ro
SIZE
-0.109*** (0.029) -0.073*** (0.016) -0.865*** (0.154) 2.065*** (0.228) -0.005 (0.019) -0.973*** (0.121) 0.000*** (0.000) -0.937* (0.519) 1.476*** (0.209) Yes Yes 18,332 0.261 1,367
-p
NII
Table 6. The effects of the crisis
CAPITAL EARNINGS GROWTH NPL ZSCORE SD(EARNINGS) Constant
SMALL (6)
MEDIUM (7)
LARGE (8)
CRISIS 2 ALL (9)
SMALL (10)
MEDIUM (11)
LARGE (12)
-0.112*** (0.024) 0.037** (0.019) -0.035*** (0.008) -0.025* (0.013) -0.270** (0.122) 2.082*** (0.189) 0.002 (0.015) -0.666*** (0.119) 0.000 (0.000) -0.487 (0.304) 0.844*** (0.183) Yes Yes 32,946 0.236 1,767
-0.114*** (0.030) 0.034 (0.026) -0.015* (0.008) -0.072*** (0.015) -0.864*** (0.153) 2.066*** (0.227) -0.006 (0.019) -0.974*** (0.121) 0.000*** (0.000) -0.940* (0.519) 1.470*** (0.208) Yes Yes 18,332 0.261 1,367
-0.084** (0.038) 0.039 (0.026) -0.036*** (0.014) 0.012 (0.020) 0.044 (0.194) 1.923*** (0.301) -0.019 (0.023) -0.446** (0.209) -0.000 (0.000) -0.163 (0.433) 0.319 (0.284) Yes Yes 11,103 0.225 530
-0.149*** (0.056) 0.009 (0.031) -0.010 (0.021) -0.022 (0.055) 1.288*** (0.456) 1.888*** (0.422) 0.043 (0.032) -0.249 (0.370) 0.000 (0.000) -0.304 (0.651) 0.773 (0.916) Yes Yes 3,511 0.274 154
-0.106*** (0.024) 0.010 (0.019) 0.006 (0.007) -0.026* (0.013) -0.273** (0.122) 2.086*** (0.189) 0.003 (0.015) -0.665*** (0.119) 0.000 (0.000) -0.484 (0.304) 0.848*** (0.184) Yes Yes 32,946 0.235 1,767
-0.109*** (0.030) 0.001 (0.032) 0.008 (0.008) -0.073*** (0.015) -0.865*** (0.153) 2.065*** (0.228) -0.005 (0.019) -0.973*** (0.121) 0.000*** (0.000) -0.937* (0.519) 1.476*** (0.208) Yes Yes 18,332 0.261 1,367
-0.077** (0.038) -0.020 (0.027) 0.021* (0.011) 0.011 (0.020) 0.038 (0.193) 1.930*** (0.302) -0.017 (0.023) -0.437** (0.210) -0.000 (0.000) -0.161 (0.433) 0.336 (0.285) Yes Yes 11,103 0.224 530
-0.148*** (0.056) 0.001 (0.034) -0.007 (0.021) -0.022 (0.055) 1.285*** (0.456) 1.894*** (0.424) 0.043 (0.032) -0.250 (0.370) 0.000 (0.000) -0.300 (0.650) 0.773 (0.916) Yes Yes 3,511 0.274 154
-0.111*** (0.024) 0.055*** (0.018) -0.039*** (0.008) -0.026* (0.013) -0.274** (0.122) 2.084*** (0.189) 0.003 (0.015) -0.663*** (0.119) 0.000 (0.000) -0.480 (0.304) 0.847*** (0.183) Yes Yes 32,946 0.236 1,767
-0.114*** (0.030) 0.055** (0.023) -0.019** (0.008) -0.073*** (0.016) -0.867*** (0.154) 2.067*** (0.227) -0.006 (0.019) -0.972*** (0.121) 0.000*** (0.000) -0.937* (0.522) 1.474*** (0.209) Yes Yes 18,332 0.262 1,367
-0.086** (0.038) 0.084*** (0.027) -0.047*** (0.014) 0.012 (0.020) 0.041 (0.193) 1.930*** (0.299) -0.019 (0.023) -0.438** (0.209) -0.000 (0.000) -0.148 (0.433) 0.320 (0.284) Yes Yes 11,103 0.226 530
-0.149*** (0.056) 0.015 (0.039) -0.009 (0.023) -0.022 (0.055) 1.286*** (0.455) 1.886*** (0.422) 0.043 (0.032) -0.248 (0.371) 0.000 (0.000) -0.298 (0.651) 0.773 (0.916) Yes Yes 3,511 0.274 154
ro
CRISIS 1 ALL (5)
Jo
BFE TFE Observations Adj R2 # Banks
LARGE (4)
-p
SIZE
MEDIUM (3)
re
CRISIS
SMALL (2)
lP
NII*CRISIS
CRISIS ALL (1)
ur na
NII
of
This table reports regression estimates of the relation between NII and LC during the crisis time. The sample period is from 2001:Q1 to 2015:Q4. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics.
36
Table 7. Alternative Measures of Liquidity creation and Diversification This table reports regression estimates of the relation between NII and LC using alternative measures of liquidity creation (Panel A), and alternative measures of diversification (Panel B). Panel C reports the net effects of diversification on liquidity creation. The sample period is from 2001:Q1 to 2015:Q4. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics. Panel A: Alternative measures of liquidity creation
NPL ZSCORE SD(EARNINGS) Constant
Jo
ur na
BFE TFE Observations Adj R2 # Banks
-0.009 (0.011) -0.003 (0.004) -0.024 (0.037) 0.536*** (0.058) 0.021*** (0.006) -0.293*** (0.035) 0.000*** (0.000) -0.342*** (0.113) 0.142** (0.058) Yes Yes 32,946 0.365 1,767
of
GROWTH
-0.099*** (0.016) -0.023** (0.010) -0.248** (0.101) 1.540*** (0.155) -0.021* (0.011) -0.351*** (0.094) -0.000 (0.000) -0.085 (0.226) 0.710*** (0.135) Yes Yes 32,946 0.127 1,767
ro
EARNINGS
-0.106*** (0.024) -0.026* (0.013) -0.274** (0.122) 2.086*** (0.189) 0.003 (0.015) -0.664*** (0.119) 0.000 (0.000) -0.483 (0.304) 0.429** (0.184) Yes Yes 32,946 0.131 1,767
-p
CAPITAL
LC_OFF (3)
re
SIZE
LC_ON (2)
lP
NII
Deviation LC (1)
37
Panel B: Alternative measures of diversification
LOAN (3)
-0.006*** (0.001)
-0.065*** (0.011)
0.917*** (0.022)
HHI
EARNINGS GROWTH NPL
lP
ZSCORE SD(EARNINGS)
HHI and NON (5)
-0.053** (0.021) -0.024* (0.013) -0.256** (0.124) 1.913*** (0.184) -0.002 (0.016) -0.713*** (0.124) 0.000 (0.000) -0.516* (0.304) 0.821*** (0.187) Yes Yes 32,946 0.230 1,767
-0.140*** (0.048) 0.088** (0.045) -0.028** (0.013) -0.335*** (0.119) 1.757*** (0.179) -0.002 (0.014) -0.751*** (0.102) 0.014 (0.258) 0.000 (0.000) 0.850*** (0.177) Yes Yes 34,309 0.239 1,828
-0.014 (0.009) -0.209** (0.084) 0.724*** (0.122) 0.107*** (0.013) -0.318*** (0.085) 0.000 (0.000) -0.378* (0.215) 0.079 (0.122) Yes Yes 33,560 0.546 1,767
Jo
ur na
Constant BFE TFE Observations Adj R2 # Banks
-0.022* (0.013) -0.229* (0.122) 1.933*** (0.182) -0.008 (0.016) -0.642*** (0.119) 0.000 (0.000) -0.576* (0.300) 0.777*** (0.177) Yes Yes 33,559 0.240 1,767
-p
CAPITAL
-0.025* (0.013) -0.260** (0.123) 1.936*** (0.183) -0.001 (0.016) -0.703*** (0.123) 0.000 (0.000) -0.505* (0.304) 0.834*** (0.186) Yes Yes 32,946 0.232 1,767
re
SIZE
HHI (4)
of
TRADING (2)
ro
NII
NII_QUARTILE (1)
38
25th -0.140*** 0.122** -0.018
50th -0.140*** 0.101** -0.039*
75th -0.140*** 0.076** -0.064***
90th -0.140*** 0.043** -0.097***
Jo
ur na
lP
re
-p
Direct effect Indirect effect Net effect
ro
Non-interest income 10th -0.140*** 0.140** 0.000
of
Panel C: Effects of Changes in Noninterest Income on Bank Liquidity Creation This table reports estimation of first derivative of LC on NII , based on regression results reported in Table 7, Panel B, Model (5) and evaluated at different values of the noninterest share based on percentile ranks (10th, 25th, 50th, 75th, and 90th percentile). Direct effect is estimated impact of a 1% increase in the noninterest income share. Indirect effect is estimated impact of a change in revenue diversification from a 1% increase in the noninterest income share. Net effect sums the direct and indirect effects. Robust standard errors are in parentheses. ***, **, * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
39
Table 8. Endogeneity concerns
SIZE CAPITAL EARNINGS GROWTH NPL ZSCORE
NII_AVG IMR Constant
0.913*** (0.225) Yes Yes 6,654 0.300 1,382
0.004* (0.002) 0.819*** (0.191) Yes Yes 31,106 0.241 1,712
ur na
SD(EARNINGS)
-0.119*** (0.027) -0.023* (0.014) -0.275** (0.125) 2.127*** (0.192) -0.000 (0.015) -0.650*** (0.121) 0.000 (0.000) -0.506* (0.305)
Jo
BFE TFE Observations Adj R2 # Banks Underidentification test Kleibergen-Paap Wald F-Stat Weak identification test Cragg-Donald Wald F-Stat Weak instrument robust inference Anderson-Rubin Wald test
-p
-0.090** (0.042) -0.029* (0.017) -0.279* (0.166) 1.896*** (0.278) -0.041 (0.029) -0.836*** (0.130) 0.000 (0.000) -0.584 (0.431)
0.152*** (0.033) 0.762 (1.123) 2.238 (2.975) -0.670** (0.331) -12.568*** (1.368) 0.005*** (0.001) -24.182*** (3.919) 9.599*** (1.690)
re
NII
Heckman 2nd step (3)
lP
(1)
Heckman 1rst step (2)
ro
PSM
of
Panel A: The table reports regression estimates of the relation between NII and LC. The sample period is from 2001:Q1 to 2015:Q4. Models (1) reports results from one-to-one matching PSM. Models (2)-(3) present estimations of Heckman selection model. Models (4)-(5) present estimations of IV estimations. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics.
-9.080*** (1.616) Yes Yes 33,622 1771
IV 1rst step (4)
0.038*** (0.003) 0.261 (0.165) 0.507 (0.387) -0.027 (0.029) 0.169 (0.186) -0.001*** (0.000) 3.643*** (0.723) 0.097*** (0.028)
-0.394*** (0.049) Yes Yes 29,592 0.196 1688
IV 2nd step (5) -0.621*** (0.184) 0.042*** (0.009) -1.288*** (0.182) 3.592*** (0.432) 0.186*** (0.034) 0.214 (0.204) -0.001*** (0.000) 1.013 (0.882)
0.089 (0.089) Yes Yes 28,184 0.126 1638
427.719*** 511.577*** 399.99***
40
Panel B: The system-GMM
of
System GMM (1)
ro
LC NII
-p
SIZE CAPITAL
re
EARNINGS GROWTH
lP
NPL ZSCORE
Constant
ur na
SD(EARNINGS)
Jo
Observations AR (1) (p-value) AR (2) (p-value) Hansen test (p-value)
-0.235*** (0.018) -0.366*** (0.017) 0.187*** (0.017) 2.415*** (0.137) 4.904*** (0.202) -0.286*** (0.015) -9.085*** (0.212) -0.003*** (0.000) 13.501*** (0.272) -2.059*** (0.236) 28,962 0.000 0.144 0.576
41
PII:
S0275-5319(19)30643-9
DOI:
https://doi.org/10.1016/j.ribaf.2020.101205
Reference:
RIBAF 101205
To appear in:
Research in International Business and Finance
Received Date:
4 June 2019
Revised Date:
14 February 2020
Accepted Date:
20 February 2020
Please cite this article as: Viet Tran D, Bank business models and liquidity creation, Research in International Business and Finance (2020), doi: https://doi.org/10.1016/j.ribaf.2020.101205
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier.
Bank business models and liquidity creation Dung Viet Tran1
1
of
Banking University Ho Chi Minh City, Ho Chi Minh city 700000, Vietnam. Mail: [email protected]. The author thanks the editor (John W. Goodell), two anonymous reviewers for their constructive comments and suggestions. The author gratefully acknowledges financial support of the National Foundation for Science and Technology Development of Vietnam (NAFOSTED).
Bank business model and liquidity creation
Expansion hypothesis Bank
re
business
-p
Diversification-Liquidity
model
ro
Graphical abstract
Liquidity creation
lP
Diversification-Liquidity
Abstract
ur na
Contraction hypothesis
Using a large panel of US bank holding companies from 2001 to 2015, this study investigates the
Jo
association between functional diversification and bank liquidity creation. I document evidence of lower liquidity creation for higher diversification. The effect of moving into nontraditional activities on liquidity creation is more apparent with large banks and less pronounced with small banks. The impact of diversification on liquidity creation is less significant during the late stage of crisis and is more clearly observed in small and medium-sized banks. Low liquidity creation banks, leveraged by a higher share of non-interest income, are more likely to further decrease
1
their liquidity creation. The study is of interest to regulators and policymakers who are concerned about bank business models.
Keywords: Bank liquidity creation; business models; diversification; crisis JEL Classification Codes: G21; G28; G34; G38
1. Introduction
of
Liquidity creation is a principal reason for the existence of banks. As liquidity providers, banks play an important role for the macroeconomy and the financial system. Banks provide necessary
ro
liquidity to informationally opaque borrowers without capital market opportunities (Levine and Zervos 1998) and supply liquid funds and payment services to households, which is the main
-p
driver for the functioning of the economy (Kashyap, Rajan, and Stein 2002). In other words, banks simultaneously satisfy the demand for liquidity by savers and the demand for longer-term
re
financing commitments by firms. Banks also provide loan commitments and other off-balance sheet guarantees that allow customers to plan their investments and expenditures, knowing that
and Stein 2002).
lP
the required funds are forthcoming when needed (Holmström and Tirole 1998; Kashyap, Rajan,
Deregulation, which started during the 1970s, allows banks to operate with fewer
ur na
constraints and to expand into highly volatile and complex non-bank activities, such as trading and market making (Tran et al. 2019), which, in turn, adds ever greater complexity to banks’ balance sheets (Financial Crisis Inquiry Commission 2011), increasing problems of information asymmetry in diversified banks. Most of the literature on bank diversification focuses on the relation between the bank business models and bank risk taking and performance (e.g. Stiroh
Jo
2004; Stiroh and Rumble 2006; DeYoung and Torna 2013; Tran et al. 2019). Generally, these studies point out that banks can benefit from diversification through a reduction of idiosyncratic risk and total risk, but this comes with the cost of heightened agency problems, leading to dispersed managerial resources and reduced stability. With the failure of a large number of banks and subsequent economic recessions, many claim the casino-style gambling on Wall Street was one of the main contributors to the last financial crisis. The banking industry is a particularly important sector in our economies, serving as a channel whose disruption could lead to adverse 2
fluctuations in the real economy. A sound banking system is a primary objective of regulators and policymakers (Tran et al. 2019). Given the importance of banks for the real economy, the lack of research on the activity strategies determining bank liquidity creation is surprising and potentially significant. This study is one of the first empirical investigations of whether liquidity creation differs across banks with different activity strategies. It provides empirical evidence on differing theoretical perspectives concerning the impact of functional diversification on bank liquidity
of
creation. On the one hand, under the framework of portfolio theory, the combination of different activities is generally believed to reduce the variance of returns, since they are not perfectly
correlated. This results in a coinsurance effect, diversification gains, and a more stable revenue
ro
stream (DeYoung and Roland 2001; Tran 2019), reducing the total risk and probability of failure of diversified banks (Brewer 1989; Saunders and Cornett 2008). Furthermore, in the absence of
-p
agency conflicts between banks and borrowers, diversified banks can retrieve borrower information and reuse it profitably for nontraditional banking activities, such as securities
re
underwriting (Diamond 1984; Yasuda 2005; Bharath et al. 2007). In turn, the information retrieved from non-interest income–generating activities also helps diversified banks make their
lP
loan-making decisions, resulting in better credit risk management. Diamond's (1984) seminal model suggests that banks that move toward nontraditional banking activities experience a more stable credit supply under aggregate shocks. When assuming the liquidity creation role, banks
ur na
are subject to liquidity risk, since they have to meet depositors’ demand for liquidity while providing funding to borrowers (Diamond and Rajan 2000, 2001). The more liquidity banks create, the greater the probability and severity of losses associated with the disposal of illiquid assets needed to meet customers’ liquidity demands (Berger and Bouwman 2009). Therefore, I suggest that diversified banks should create more liquidity than other banks, since they are
Jo
financially stronger in terms of meeting the demand from depositors and providing funds to borrowers. I call this hypothesis the diversification–liquidity expansion hypothesis. On the other hand, diversification raises concerns of intensified agency problems
associated with bank size, as well as banks’ opaqueness and complexity, leading to discretionary decisions to undertake value-decreasing investments (Berger and Ofek 1995). Managers derive private benefits from diversification that exceed their private costs (Denis, Denis, and Sarin 1997). They can adopt and maintain a strategy of inefficient diversification, even if it can 3
negatively affect shareholder value. Furthermore, more diversified activities do not translate into risk reduction if there is lack of expertise in the newly adopted business (Jiménez and Saurina 2004). Diversification can then spread out managerial resources, leading to higher costs to coordinate corporate policies and, in turn, inducing failure to meet the liquidity demands of both depositors and borrowers and negatively affecting liquidity creation. I call this hypothesis the diversification–liquidity contraction hypothesis. In this paper, I examine the interplay of diversification and bank liquidity creation for a
of
large sample of US bank holding companies (BHCs) from 2000:Q1 to 2015:Q4. Using the richest and most complete database related to banking institutions that provides the details of thousands of banks operating in hundreds of local markets, I can analyze the data at the highest
ro
frequency possible. I use the preferred measure of Berger and Bouwman (2009), catfat, as the main proxy for bank liquidity creation (LC), which is also used in numerous other studies
-p
(Distinguin, Roulet, and Tarazi 2013; Horvath, Seidler, and Weill 2016; Berger and Bouwman 2017; Jiang, Levine, and Lin 2017). In robustness checks, I also use different measures and still
re
find similar findings. I use the share of non-interest income (NII) as the proxy for diversification. A large share of non-interest income means greater diversification into nontraditional activities.
lP
Controlling for the effects of different bank characteristics and bank and time fixed effects, the empirical analysis provides consistent evidence of lower liquidity creation for diversified banks. The evidence is economically significant. For example, the baseline model
ur na
indicates that a one standard deviation increase in NII will lead to a decrease of LC of 0.014. Given LC’s mean of 0.445 and standard deviation of 0.186, this decrease is equivalent to 3.15% of the average bank’s assets. The evidence provides support for the diversification–liquidity contraction hypothesis, which suggests that agency problems derived from the complexity of diversified banks lead banks to fail to meet their clients’ liquidity demands.
Jo
To ensure the robustness of the findings, I run the baseline model with a single cross-
sectional regression (average analysis), with a balanced panel (excluding banks that only existed during part of the sample period), excluding merger and acquisition (M&A) banks (i.e., banks with large changes in assets over a certain periods, where I use a threshold of a 20% change in assets over a quarter), and with alternative econometric models, such as Newey–West, Fama– MacBeth, and two-way clustering by bank and time. The results of these specifications lend support to the diversification–liquidity contraction hypothesis. 4
I focus on the distribution of liquidity creation and use quantile regression to examine how the impact of activity strategies varies across the distribution of liquidity creation. Rather than relying on a single description of the central behavior of the sample, the quantile approach explores a range of conditional quantile functions, which, in turn, allows potential forms of conditional heterogeneity to be explored. The evidence indicates that diversification not only affects the conditional average of liquidity creation, but also influences the dispersion of liquidity creation. Low liquidity creation banks, leveraged by a higher share of non-interest income, are
of
more likely to decrease their liquidity creation. In other words, the impact on diversification appears to be more profound for low liquidity creation banks.
Since banks of different size create liquidity in different ways (Berger and Bouwman
ro
2009), I run the baseline model for banks with a range of different sizes: small banks, with assets of up to $1 billion; medium-sized banks, with assets between $1 billion and $5 billion; and large
-p
banks, with assets of over $5 billion. I document variations of the effect of diversification on bank liquidity creation across bank sizes. Large banks that engage in non-interest income–
re
generating activities, reduce their liquidity creation the most, followed by small diversified banks. Medium-sized banks reduce their liquidity creation the least when increasing their share
lP
of non-interest income. Potential explanations are as follows. Larger banks are more likely to gain from nontraditional activities for a long time and to then push their diversification level to saturation, no longer gaining benefit from marginal increases in diversification but potentially
ur na
incurring greater agency problems. This implies a large reduction in liquidity creation by large banks. Even though they have more to gain from diversification, small banks can lack experience in new and more complex nontraditional activities, thereby reducing their operating stability and liquidity creation.
I investigate whether diversification affects liquidity creation differently during financial
Jo
crises. Diversified banks create more (less negative) liquidity than other banks during crises than under normal circumstances. The evidence highlights the bright side of diversification, which could make banks less risky during times of turmoil, as documented by DeYoung and Torna (2013) and Tran et al. (2019), giving rise to more liquidity creation. When performing analyses across bank sizes, I find no differences in the effect of diversification on liquidity creation across bank sizes during times of crisis.
5
Since differences could exist in deposit flows between the early and late stages of the global financial crisis that affected banks’ ability to create liquidity, I split the crisis period into two subperiods—CRISIS 1, from 2007Q3 to 2008Q2, and CRISIS 2, from 2008Q3 to 2009Q2— and rerun the analysis separately for each stage of the crisis. I find that the bright side of diversification is mostly driven by the second stage of the crisis (CRISIS 2). Small and mediumsized banks that diversify into nontraditional banking activities create more liquidity during the late stage of the crisis, and this effect seems to be stronger for medium-sized banks. I find no
of
differences in the effect of diversification on liquidity creation at large banks during times of crisis.
In additional robustness tests, I use alternative measures of liquidity creation (deviating a
ro
bank’s liquidity creation from the industry average, on- and off-balance sheet components of liquidity creation). I also use alternative measures of diversification, such as the share of trading
-p
income, the ratio of loans over total assets, and an adjusted Herfindahl–Hirschman index. The results remain similar.
re
I also assess the net effect of moving toward nontraditional activities on bank liquidity creation by evaluating the 10th, 25th, 50th, 75th, 90th percentiles of the share of non-interest
lP
income. I find that banks benefit from diversification, but these gains are quickly offset by adverse effects from reliance on riskier assets. Banks with a relatively small share of non-interest income do not seem to change their liquidity creation when relying more on non-interest
ur na
income–generating activities. However, with greater reliance on non-interest income–generating activities, banks reduce their liquidity creation with an increase in the share of non-interest income.
Next, I address endogeneity concerns, since the decision to diversify is a deliberate decision by bank managers. They could choose to diversify when they receive more benefits than
Jo
the cost of diversification (Campa and Kedia 2002). I begin with propensity score matching (PSM). Next, I use a Heckman two-step model and an instrumental variable (IV) approach to examine whether the private information of the decision to diversify is correlated with the propensity to create liquidity. In all specifications, the findings remain quantitatively similar to the main results. This study contributes to the literature in several ways. First, to the best of my knowledge, this study is the first investigation of the impact of functional diversification on 6
liquidity creation within the US banking industry. In a concurrent study, Hou et al. (2018) document how an increase in the degree of bank diversification between traditional bank activities generating net interest income and nontraditional bank activities generating noninterest income reduces the liquidity creation of Chinese banks. Unlike the US banking system, the banking sector in China, as well as regulations regarding market access and the range of products, remains under strict government supervision. Bank’s net interest margin earnings are the main component of their profits. This study’s focus on one of the most regulated banking
of
industries, that in the United States, where banks have been allowed to extensively expand their activities into nontraditional activities for decades, and its comprehensive sample of long
duration strengthen its results, increase their robustness, and reduce sample selection bias. The
ro
study contributes to the liquidity creation literature by providing evidence of the dark side of
diversification for bank liquidity creation during normal circumstances, as well as a bright side
-p
of diversification during times of crisis. Furthermore, this study documents how the impact of diversification on bank liquidity creation varies by bank size.
re
Second, this study provides evidence of the effects of diversification over the entire range of liquidity creation distribution. The traditional inference approach (i.e., ordinary least squares,
lP
or OLS) reflects the mean behavior of the sample under the assumption of homogeneity in the relation between diversification and liquidity creation. However, with a sample as heterogeneous as this study’s, this approach could be a poor method for examining the relation between
ur na
diversification and liquidity creation across the entire industry. Rather than relying on a single description of the central behavior of the sample, the quantile approach explores a range of conditional quantile functions, which, in turn, allows potential forms of conditional heterogeneity to be explored (Tran, Hassan, and Houston 2019). I document that activity strategies not only affect the conditional average level of liquidity creation, but also influence the distribution of
Jo
liquidity creation.
This paper is organized as follows. The next section describes the data and variables.
Section 3 reports the main results and those of alternative tests. Section 4 provides the results of additional tests. Section 5 concludes the study.
7
2. Theoretical framework and hypothesis development Banks assume two central roles in our economy—creating liquidity and transforming risk— which are often jointly referred to as the qualitative asset transformation of banks (Bhattacharya and Thakor 1993). Despite a vast literature dedicated to the risk transformation role of banks (i.e., issuing riskless deposits to finance riskier loans), few empirical studies have focused on the liquidity creation role of banks. Banks create liquidity on balance sheets by financing illiquid loans, for example, business loans with relatively liquid liabilities, such as transaction deposits (Bryant 1980; Diamond and Dybvig 1983). They provide borrowers the necessary funds to
of
invest, while delivering deposits to assume liquidity and payment services to the public. In other words, the liquidity creation role of banks simultaneously satisfies the demand for liquidity from
ro
savers and the demand for longer-term financing commitments from firms. Banks also create liquidity off the balance sheets by providing loan commitments and other off-balance sheet
-p
guarantees that allow firms to develop and modify long-run investment strategies efficiently (Holmström and Tirole 1998; Kashyap, Rajan, and Stein 2002). Loan commitments allow
re
customers to plan their investments and expenditures, knowing that the required funds are forthcoming when needed (Berger and Bouwman 2017).
lP
Until recently, liquidity creation was mostly relegated to a theoretical concept and was not often used in empirical studies. Since Berger and Bouwman (2009) filled an important gap in bank research by providing a comprehensive measure of liquidity creation, a body of literature
ur na
has emerged on the association between bank liquidity creation and other bank financial decisions, such as equity capital (Horváth, Seidler, and Weill 2013), risk taking (Andreou, Philip, and Robejsek 2016), and other financial instruments and phenomena, such as monetary policy (Berger and Bouwman 2017), economic output (Berger and Sedunov 2017), government intervention (Berger et al. 2016), and competition (Horvath, Seidler, and Weill 2016; Jiang,
Jo
Levine, and Lin 2017). However, the present study is one of the first to investigate the impacts of the bank business models on liquidity creation. The bank business models, involving the integration of traditional and nontraditional
banking activities, has also been thoroughly examined. Most of the studies focus on the association between activity diversification and bank riskiness (Stiroh 2004; Stiroh and Rumble 2006; Berger, Hasan, and Zhou 2010; De Jonghe, Diepstraten, and Schepens 2015; Tran et al. 2019). 8
Within the framework of modern portfolio theory, the combination of different activities is generally believed to reduce the variance of returns, since they are not perfectly correlated, reducing the probability of failure (Brewer 1989; Saunders and Cornett 2008). Focused banks might have to reject positive net present value investment opportunities if external finance is costly (Kashyap and Stein 2000; Campello 2002) or if they are experiencing low income levels (Houston, James, and Marcus 1997). Diversified banks can reduce this problem, since diversification helps banks decrease volatility and the correlation between funding and
of
investment opportunities. By moving away from traditional intermediation activities, banks earn less interest income and, at the same time, earn less interest and take on less credit risk (Saunders and Walter 1994). Diversification could reduce the correlation of loan returns and, subsequently,
ro
idiosyncratic credit risk (Acharya, Hasan, and Saunders 2006), which decreases the risk of large liquidity outflows and the risk of early liquidation (Diamond and Dybvig 1983).
-p
Additionally, assuming the absence of agency conflicts between banks and borrowers, Diamond (1984) argues that diversified banks can enhance credibility in their loan making
re
decisions and in their borrowers’ monitoring by overcoming the information asymmetry between depositors and borrowers. Diversified banks can retrieve borrower information and reuse it
lP
profitably for nontraditional banking activities such as securities underwriting (Diamond 1984; Yasuda 2005; Bharath et al. 2007). In turn, the information retrieved from non-interest income– generating activities also helps diversified banks in their loan making decisions, resulting in
ur na
better credit risk management. Diamond (1984) suggests that diversified banks have a more stable credit supply under aggregate shocks, which could, in turn, lead to lower volatility of the cash flow from loan portfolios. In brief, these arguments suggest that banks engaging in nontraditional activities reduce the need for self-insurance, benefit from economies of scale and scope (Santomero and Eckles 2000), increase profitability (Sanya and Wolfe 2011), achieve
Jo
capital savings (Shim 2013), and experience a reduction in idiosyncratic risk exposure (Kwast 1989), total risk (Tran 2019), and the probability of failure (DeYoung and Torna 2013). When assuming the liquidity creation role, banks are subject to liquidity risk, since they
must meet depositors’ demand for liquidity when providing funding to borrowers (Diamond and Rajan 2000, 2001). In other words, the more liquidity banks create, the greater the probability and the severity of the losses associated with having the disposal of illiquid assets to meet their customers’ liquidity demands (Berger and Bouwman 2009). Kashyap, Rajan, and Stein (2002) 9
argue banks can still create liquidity as long as an imperfect correlation exists between drawdowns on demandable deposits and credit lines. Diversification can make simultaneous drawdowns less likely, allowing banks to create more liquidity. Therefore, I suggest that diversified banks should create more liquidity than other banks, since they are financially better able to meet the demand from depositors and to provide funds to borrowers. I call this hypothesis the diversification–liquidity expansion hypothesis. H1a (diversification-liquidity expansion hypothesis): Banks engaging in nontraditional
of
activities create more liquidity than other banks. On the other hand, diversification raises concerns of intensified agency problems, since it can increase a bank’s size, as well as its opaqueness and complexity, leading to discretionary
ro
decisions to undertake value-decreasing investments (Berger and Ofek 1995; Krishnaswami, Spindt, and Subramaniam 1999). The information asymmetry could prevent shareholders from
-p
rigorously monitoring managers, leading to increased agency problems. Managers derive private benefits from diversification that exceeds their private costs (Denis, Denis, and Sarin 1997).
re
They can adopt and maintain a strategy of inefficient diversification, even if it could negatively affect shareholder value (i.e., the risk shifting problem and the underinvestment problem), since
lP
managing a diversified bank can increase power and prestige (Jensen 1986), entrenchment (Shleifer and Vishny 1989), and compensation arrangements (Jensen and Murphy 1990), as well as reduce the risk of managers’ undiversified personal portfolios (Amihud and Lev 1981).
ur na
Furthermore, more diversified activities do not translate into risk reduction if there is a lack of expertise in the newly adopted business (Jiménez and Saurina 2004). Due to the moral hazard related to fixed-rate deposit insurance or the risk-seeking behaviors of bank management (Litan 1985), banks could support risky investments by using federally insured deposits. The seminal theoretical model of Shleifer and Vishny (2010) suggests banks that move toward nontraditional
Jo
banking activities tend to expand their balance sheets and transmit fluctuations from the security markets to the real economy; the volatility of sentiment thus turns into the volatility of real activity through the banking sector. The authors emphasize that such profit maximization practices could induce system instability. Diversification could then spread out managerial resources, leading to higher costs of coordinating corporate policies, which, in turn, induces failure to meet the liquidity demands of
10
both depositors and borrowers and negatively affects liquidity creation. I call this hypothesis the diversification–liquidity contraction hypothesis. H1b (diversification–liquidity contraction hypothesis): Banks engaging in nontraditional activities create less liquidity than other banks. The banking literature tends to presume that diversification and size go hand in hand. However, larger banks that are usually better diversified than smaller banks are not less risky. Studies document that large banks benefit from diversification advantages to pursue riskier
of
activities (Demsetz and Strahan 1997). Nevertheless, large banks are typically exposed to more stringent monitoring from regulators and other stakeholders. Therefore, I propose the following hypothesis.
ro
H2: The effect of diversification on bank liquidity creation varies with bank size.
Another interesting question relates to the association between diversification and
-p
liquidity creation during times of crisis. The financial crisis of 2007–2009 led to the failure of a large number of banks, forced recapitalizations, distressed mergers with healthier banks, and
re
significant government intervention around the world. In this context, it is of particular interest to investigate whether this extraordinary situation led to a reassessment of the role of bank
lP
diversification. On the one hand, diversification potentially enables banks to better resist shocks in one area of their business activity. This suggests that diversified banks have incentives to increase liquidity creation during crises to take advantage of market share. On the other hand,
ur na
banks that engage in more nontraditional banking activities could be the most clearly exposed, due to commercial banks expanding into investment banking activities (Elsas, Hackethal, and Holzhäuser 2010). It therefore follows that diversified banks will curtail their intermediation activity to protect financial stability. The following hypotheses will therefore be tested. H3a: The effect of diversification on bank liquidity creation is mitigated during a crisis.
Jo
H3b: The effect of diversification on bank liquidity creation is stronger during a crisis.
3. Data and variables 3.1 Sample banks
We obtain from the Federal Reserve System quarterly FR Y-9C regulatory reports filed by BHCs with assets of $150 million and over. The raw data cover the period from 2001:Q1 to 2015:Q4. Bank–quarter observations with missing or incomplete financial data for the accounting variables 11
in the main regression model are removed. Following Berger and Bouwman (2013), I replace all observations with a ratio of total equity to total assets of less than 1% by 1% to avoid distortions in ratios involving equity. I also exclude observations with negative or nonexistent outstanding loans or deposits. The data set contains 32,946 observations for 1,767 BHCs. All financial ratios are winsorized at the 1% level at the top and bottom of their distribution to mitigate the effects of outliers.
of
3.2 Measures of bank liquidity creation I use the liquidity creation measure of Berger and Bouwman (2009). Since banks are able to
securitize or sell loans, they prefer to classify loans by category (cat) rather than by maturity
ro
(mat). Additionally, since banks create liquidity both on and off the balance sheet, they prefer including off-balance sheet items (fat) to excluding them (nonfat). I therefore use their preferred
-p
measure, the catfat measure, which includes both on- and off-balance sheet items and classifies items by category. In robustness tests, I use alternative measures and still obtain similar
re
evidence. The data on liquidity creation were retrieved from Christa Bouwman’s website.1 The catfat and other liquidity measures are normalized by gross total assets, since they measure the
lP
dollar amount of liquidity creation. 3.3 Activity strategies and control variables
ur na
Following prior literature (e.g., Stiroh and Rumble, 2006; De Jonghe, Diepstraten, and Schepens, 2015; Tran et al., 2019), I capture the activity strategies of banks using bank income structure, which is measured as the ratio of non-interest income to net operating income (NII). In robustness tests, I use alternative measures and still obtain similar evidence. To mitigate potential omitted variable bias, I control for various bank-specific variables,
Jo
including bank size (SIZE), capital ratio (CAPITAL), bank performance (EARNINGS), asset growth (GROWTH), different indicators of bank risk such as earnings volatility (SD(EARNINGS)), loan portfolio quality (NPL), and bank risk (ZSCORE). See Table 1 for the variable definitions and Table 2 for a descriptive summary.
See https://sites.google.com/a/tamu.edu/bouwman/data. See Berger and Bouwman (2009) for the measure’s explanation. 1
12
4. Do activity strategies affect bank liquidity creation? 4.1 Main findings In this section, I conduct multivariate analysis to formally investigate the magnitude of activity strategies on bank liquidity creation, after controlling for the other control variables. Specifically, the empirical specification I estimate is 𝑌𝑖𝑡 = 𝛼𝑖 + 𝑁𝐼𝐼𝑖𝑡−1 + 𝑍𝑖𝑡−1 + 𝜇𝑖 + 𝜃𝑡 + 𝜀𝑖𝑡
(1)
of
where 𝑌𝑖𝑡 is the measure of liquidity created by bank i at time t. Following Berger and Bouwman (2009), I use their preferred measure, catfat, as a proxy of bank liquidity creation that includes
ro
both on- and off-balance sheet items. As argued above, this measure classifies loans by category (cat) instead of maturity (mat) and gauges how much liquidity banks create off the balance sheet
-p
(fat). Let LC denote the catfat measure. The variable of interest is NII, which is the ratio of noninterest income to net operating income, and 𝑍𝑖𝑡 is the vector of control variables described
re
above.
In all specifications, I lag bank factors by one period to mitigate simultaneity and
lP
endogeneity problems. I include bank fixed effects, 𝜇𝑖 , motivated by the fact that differences in bank liquidity creation are partly related to unobservable but time-invariant characteristics of banks, such as corporate culture and bank management. The estimation of a separate intercept
ur na
(𝛼𝑖 ) for each bank before fitting the slope coefficients controls for unobservable, time-invariant sources of bank heterogeneity, allowing a focus on the variation in liquidity creation at the level of individual banks over time. I also include time fixed effects, 𝜃𝑡 , to control for time effects that can affect the liquidity creation of banks. The term 𝜀𝑖𝑡 is the error term. Since LC is likely to be correlated within a bank over time, standard errors used to assess significance are corrected for
Jo
heteroscedasticity and bank-level clustering.2 The main results from the multivariate analysis are shown in Table 3. Model (1) reports
the baseline model. The estimated constant indicates that the expected average LC value of
2
To avoid the problem of spurious regressions, the model should include all variables that are stationary. I run unit root tests, not tabulated here. If I can reject the null hypothesis of a unit root, the findings from these unit root tests will allow an investigation into the association between diversification and liquidity creation. If not, the estimated parameters could be biased, due to the spurious regression problem. I employ a panel-based unit root test, a Fishertype test (Choi 2001), to test the null of the unit root of all the variables. The untabulated results strongly suggest that all the variables are stationary. I thank an anonymous referee for suggesting this complementary test.
13
banks is 0.850 and statistically significant. The coefficient of the main variable of interest, NII, is negative and statistically significant at the 1% level, suggesting that moving toward non-interest income–generating activities reduces liquidity creation. The economic magnitude of this effect is also significant. A one standard deviation increase of NII, holding all others equal, results in a decrease in LC of 0.014 (i.e. the coefficient on NII, -0.102, times the standard deviation of NII, 0.13). With LC having a mean of 0.445 and a standard deviation of 0.186, this decrease is equivalent to 3.15% of the average bank’s assets. This evidence is consistent with the
of
diversification–liquidity contraction hypothesis (H1b), showing that liquidity creation is decreased in diversified banks. The evidence is also in line with the findings of Hou et al. (2018) for the Chinese banking system.
ro
In Model (2), I run the baseline model with a single cross-sectional regression (average analysis) to address the potential error dependence problem. By performing this time series mean
-p
regression (one observation per bank), I eliminate the problem of serially correlated errors. This estimation maintains heterogeneity across banks but does not exploit the time series variation in
re
the observations (Tran, Hassan, and Houston 2019). The findings are comparable to the earlier results.
lP
For Model (3), I report the results for a balanced panel (excluding banks that only existed during part of the sample period). This exclusion mitigates the effects of M&A activities and bank defaults on the investigation, although at the price of overrepresenting “successful” banks.
ur na
In Model (4), I exclude M&A banks (banks with a large change in assets over a period, where I use a threshold of 20% of assets changing over a quarter). The results show the coefficients of NII remain negative and statistically significant for all specifications. Next, I use alternative econometric methods, such as Newey–West (Model (5)), Fama– MacBeth (Model (6)), and two-way clustering by bank and time, to allow for correlations among
Jo
different banks in the same quarter and across quarters for the same bank (Model (7)). The results are still similar. In untabulated tests, instead of including all three risk measures (NPL, SD(EARNINGS),
ZSCORE) at one time, I rerun the baseline model by alternatively including these measures individually. I also use a lag of four periods instead of one period, including the state dummy, to take into account characteristics of the local environment. I also exclude the top largest banks
14
(top 10th percentile of bank assets, too big to fail banks). Again, I observe that diversified banks still experience lower liquidity creation than other banks. Regardless of the control variables (Model (1)), I observe that large and well capitalized banks are more likely to create less liquidity.3 I documented that highly profitable banks are more likely to create liquidity than other banks. The evidence also shows that banks with a high proportion of nonperforming loans experience lower liquidity creation than other banks. I find no evidence of growth and other risk proxies.
of
In brief, the findings support the diversification–liquidity contraction hypothesis (H1b), in that diversification will make banks more complex, heighten moral hazard problems, and disperse managerial resources and operating stability, leading to failure to satisfy clients’
ro
liquidity demands. The negative effects of dispersing managerial resources and strengthening moral hazard due to diversification dominate the bright side of the risk mitigation and economies
-p
of scale and scope from diversification.
re
4.2 Quantile regression
The main purpose of this study is to investigate the relation between banks’ business models and
lP
their liquidity creation. Regulators and policymakers seem to be more interested in bank behaviors at the tails of the distribution of liquidity creation, due to the link between liquidity creation and real economic output (Berger and Sedunov 2017).
ur na
Table 4 shows the results of quantile regression—a generalization of median regression analysis to other quantiles—to assess whether the interplay between diversification and liquidity creation differ across quantiles of liquidity creation. The traditional inference approach (i.e., OLS) used in earlier sections describes the average behavior of the sample, with the assumption of the homogeneity of the effects of diversification on bank liquidity creation. However, when
Jo
heterogeneity in the sample is important, the traditional OLS approach might not be ideal. Rather than relying on a single description of the central behavior of the sample, the quantile approach
3
In an untabulated test, I use alternative measures of bank size. First, since size is largely an outcome of bank decision making and highly correlated with other independent and dependent variables, I decompose bank size with respect to all the other independent variables into two components: an organic growth component that is measured by the fitted value and a historical size component that equals the residual. Orthogonalizing size allows the pure effects of size to be derived (De Jonghe 2010). Second, I also check for a nonlinear relation between earnings management and size by including size–decile fixed effects to control for unobserved heterogeneity across banks in different size categories, as suggested in Ellul and Yerramilli (2013). I obtain similar results.
15
explores a range of conditional quantile functions, which, in turn, allows potential forms of conditional heterogeneity to be explored and an understanding of the behavior of banks across the distribution of liquidity creation (Tran, Hassan, and Houston 2019). Furthermore, the quantile regression approach avoids the restrictive assumption that the error terms are identically distributed at different locations of bank liquidity creation (Klomp and Haan 2012). I report the relations between NII and LC across quantiles. The coefficients on NII in Models (1) to (9) demonstrate that the impact of diversification on bank liquidity creation is
of
indeed uniform in sign (negative), suggesting that NII decreases banks’ liquidity creation in banks of all levels of liquidity creation. I plot the coefficients for NII across the LC distribution quantiles in Figure (1) and observe that these coefficients do not vary significantly across the
ro
distribution of liquidity creation, except for the 10th quantile of LC. This finding implies that, for banks with low liquidity creation (10th quantile of LC), moving toward non-interest income–
-p
generating activities would decrease their liquidity creation.
Overall, the empirical findings indicate activity strategies not only affect the conditional
re
average liquidity creation but also influence the liquidity creation distribution.
lP
4.3 Net effect of diversification on liquidity creation for large, medium-sized, and small banks Table 5 contains the regression results for different size ranges, that is, small banks, with assets of up to $1 billion; medium-sized banks, with assets between $1 billion and $5 billion; and large
ur na
banks, with assets of over $5 billion. The results for all three models, Models (1) to (3), indicate that diversification is associated with lower liquidity creation across all size classes, but with different magnitudes. The magnitude of the coefficient on NII is -0.109, significant at 1%, for the sample of small banks, suggesting that a one percentage point higher level of diversification will decrease a small bank’s liquidity creation by 0.1% of its assets. This effect seems to be mitigated
Jo
in the sample of medium-sized banks, where the coefficient on NII is -0.079, significant at 5%. For the sample of large banks, a one percentage point higher level of diversification will decrease a large bank’s liquidity creation by 0.15% of its assets, which appears to be a substantial effect. Potential explanations are as follows. Larger banks are more likely to gain from nontraditional activities for a long time and to then push their diversification levels to saturation, no longer gaining benefits from marginal increases in diversification but potentially incurring greater agency problems. This suggests a large reduction in the liquidity creation of large banks. Despite 16
having more to gain from diversification, small banks can lack experience in new and more complex nontraditional activities, thereby reducing their operating stability and liquidity creation. In summary, I document a variation of the effect of diversification on bank liquidity creation across bank sizes. Large banks that engage in non-interest income–generating activities reduce their liquidity creation the most, followed by small diversified banks. Medium-sized banks reduce their liquidity creation the least when increasing their share of non-interest income.
of
The evidence supports H2. 4.4 Effects of diversification on liquidity creation during banking crises
ro
In this section, I investigate whether the association between NII and LC changes during crisis periods. This study starts in 2001:Q1 and includes the last crisis, from 2007:Q3 to 2009:Q2,
-p
following Acharya and Mora (2015). I rerun the baseline model by adding the crisis dummy and its interaction with NII (NII*CRISIS). The focus is the coefficient of the interaction term, which
The results are reported in Table 6.
re
measures how the crisis affects the association between diversification and liquidity creation.
lP
Model (1) in Table 6 shows that, during times of crisis, banks create less liquidity than in normal times, as indicated by the negative and statistically significant coefficient on CRISIS. The interaction term NII*CRISIS is positive and statistically significant, indicating that
ur na
diversified banks created more (less negative) liquidity than other banks during the crisis than during normal circumstances. The results support H3a, that is, banks that move toward nontraditional activities create more liquidity during crisis times. The evidence points out the bright side of diversification, which can induce banks to take on less risk during times of turmoil, as documented recently by DeYoung and Torna (2013) and Tran et al. (2019), giving rise to
Jo
greater liquidity creation.
However, differences in deposit flows could exist between the early and late stages of the
crisis. Indeed, there is deposit funding pressure in the first phase of the crisis, starting August 9, 2007, due to the freezing of asset-backed commercial paper markets. This reflects investors’ perception of the greater risk of bank deposits relative to other instruments offering similar liquidity and payment services (Acharya and Mora 2015). The situation changed when the government explicitly backed the depository system through an increase in deposit insurance to 17
$250,000, among other measures. These factors could have also affected banks’ ability to create liquidity. I therefore divide the crisis period into two subperiods: CRISIS 1, from 2007Q3 to 2008Q2, and CRISIS 2, from 2008Q3 to 2009Q2. I rerun the analysis separately for each stage of the crisis. The results in Models (5) and (9) show that the results reported in Model (1) are mostly driven by the second stage of the crisis (CRISIS 2). I further investigate the effects of the crisis on the impacts of diversification on liquidity creation across bank size classes. The coefficients on NII in Models (2) to (4) are negative and
of
statistically significant, indicating that, during normal times, diversification will lead to lower liquidity creation across bank size classes. The coefficients on NII*CRISIS are positive but not
creation across bank size classes during the crisis period.
ro
statistically significant, implying no difference in the effects of diversification on liquidity
When the crisis is split into two subperiods, I find no evidence during the early stage of
-p
the crisis (CRISIS 1), but there is evidence during the late stage of the crisis (CRISIS 2). In summary, the results shown in Table 6 show evidence of the bright side of
re
diversification during a crisis; that is, diversified banks create more liquidity during crisis times. This effect can be clearly observed among medium-sized and small banks during the late stage of
5. Robustness checks
lP
the crisis.
ur na
5.1 Alternative measures of liquidity creation
In Table 7, Panel A, I rerun the baseline model with alternative measures of bank liquidity creation. In Model (1), inspired by Foos, Norden, and Weber (2010) and Tran, Hassan, and Reza (2019), I use the deviation of LC of bank i at time t from the industry average at time t as a measure of bank liquidity creation.
Jo
In Models (2) and (3), I divide the LC measure into on- and off-balance sheet
components (LC_ON and LC_OFF, respectively) to investigate the effect of diversification on different types of liquidity creation activities. The results show that the negative association between diversification and liquidity creation is mainly driven by the on-balance sheet component of liquidity creation.
18
5.2 Alternative measures of diversification In Table 7, Panel B, I rerun the baseline model with alternative measures of bank diversification. In Model (1), I first rank the NII variable by quartiles and create the variable NII_QUARTILE, which takes a value ranging from one (low) to four (high). This approach generates greater variation in the distribution of the extent of diversification. I next focus on the most controversial type of non-interest income, that is, the trading income in Model (2). I compute trading income following Gandhi, Kiefer, and Plazzi (2016),
of
who suggest totaling trading revenue, interest income from trading assets, and gains or losses realized from held-to-maturity and available-for-sale securities. Using this measure mitigates
ro
concerns of non-interest income including income derived from traditional activities. I use the ratio of loans over total assets as the proxy for a bank’s functional
-p
diversification (LOAN) in Model (3). In contrast to other measures, this measure is interpreted as an inverse diversification measure, since the higher value of loan ratio means that banks focus more on traditional banking activities.
re
The results shown in Models (1) to (3) in Panel B of Table 7 confirm earlier findings, suggesting that a move toward nontraditional activities will make banks create less liquidity.
lP
Following Stiroh and Rumble (2006), in Model (4), I use an adjusted Herfindahl– Hirschman index to measure diversification (HHI), which accounts for variations in the breakdown of net operating income (NOI) into the two main categories of net interest income
ur na
(NIM) and non-interest income (NII). The findings in Model (4) highlight that an increase of diversification leads to a reduction in bank liquidity creation: 𝐻𝐻𝐼 = 1 − [(𝑁𝐼𝑀)2 + (𝑁𝐼𝐼)2 ]
(2)
Jo
For each appropriate value of HHI, there are two possible values of NII, each reflecting
different specific underlying activities of banks (with greater reliance on either NIM or NII) and each potentially having different effects on LC. Then, in Model (5), I simultaneously include HHI and NII to account for the potential impacts of the heterogeneity of activities on LC. I find a positive coefficient for HHI and a negative coefficient for NII, both statistically significant. Economically, holding all other variables constant, an increase of HHI from zero to its mean value (i.e., 0.325) enhances LC by about 0.029 (= 0.088*0.325, i.e., from 0.456 to 0.485). Banks 19
experience a decrease in LC of about 0.033 (= -0.140*0.234), from 0.456 to 0.423, when moving from pure lending activities toward an activity generating $0.234 of NII per $1 of NOI. It is worth noting that, since HHI is a quadratic function of NII, it is clear that the effects of a variation in NII can be disentangled into a direct exposure effect (i.e., the coefficient on NII) and an indirect exposure effect (i.e., the coefficient on HHI times the first derivatives of HHI on NII). I then calculate the net effect by evaluating at the 10th, 25th, 50th, 75th, and 90th percentiles of NII the direct, indirect, and net effects on changes in NII on LC. The results are
of
shown in Panel C of Table 7. The net effect, which combines both effects, offers interesting evidence. For banks with a heavily concentrated interest income (at the 10th and 20th percentiles of NII), the diversification gains do not differentiate from the negative direct effects of an
ro
increase in NII, suggesting that the direct and indirect exposure effects on LC come close to
canceling each other out. The net effect is not statistically different from zero. However, the net
-p
effect becomes statistically negative with greater reliance on NII as the indirect positive effects progressively decrease. The net effect becomes significantly negative after the 50th percentile of
re
NII. At the 90th percentile of NII, one standard deviation in NII (i.e. 0.132) induces to a jump of -0.013 in LC, which corresponds to 2.8% of the mean of LC.
lP
In summary, the results with alternative measures of diversification confirm earlier findings, suggesting that a move toward nontraditional activities would make banks create less liquidity. I also find that banks benefit from diversification, but these gains are quickly offset by
ur na
the adverse effects from reliance on riskier assets. Banks with a relatively small share of noninterest income do not seem to change their liquidity creation when relying more on non-interest income–generating activities. However, with greater reliance on non-interest income–generating activities, banks reduce their liquidity creation with an increase in the share of non-interest
Jo
income.
5.3 Endogeneity concerns This investigation could be subject to reverse causality or endogeneity among the variables. The specification in the baseline model is based on the assumption that a bank’s decision to diversify is exogenous. However, diversification is not random, but a deliberate decision made by bank managers; hence, the same bank-level characteristics that drive the decision to diversify could also affect the liquidity creation of banks. Failure to control for factors that drive banks to 20
diversify leads to biased econometric results that inappropriately attribute the diversification costs and benefits to diversification per se, rather than to the underlying traits that lead banks to diversify (Campa and Kedia 2002). I therefore the OLS estimation with different approaches: the Heckman selection model, the IV approach, and PSM. These procedures should control for any potential selection bias in the above estimation. The results are tabulated in Table 8. First, I employ the PSM method developed by Rosenbaum and Rubin (1983) and
of
extended by Heckman et al. (1997). To conduct PSM, I separate the full sample into two groups: diversified (treated) and focused (untreated) banks. Following Laeven and Levine (2007), I
classify banks into two separate groups: (i) Banks with a share of net interest income between
ro
10% and 90% are classified as diversified, whereas (ii) banks with a share of net interest income either below 10% or above 90% are classified as focused (or specialized). I measure the
-p
propensity of undergoing treatment (i.e., the probability of diversification) by using a logit model for both treated and untreated samples. The dependent variable in this logit model is a binary
re
variable that equals one if the bank is classified as diversified, and zero otherwise. The logit model is similar to that in the first stage of the Heckman selection model described below. I
lP
match each diversified bank with one focused bank that shares similar characteristics, as reflected in their propensity scores. I retain only untreated observations whose propensity scores fall inside the interval defined for the treated group. I impose a tolerance level of 0.5% on the
ur na
maximum propensity score distance allowed (caliper) to minimize the risk of bad matches. I use one-to-one matching without replacement, which requires each focused bank to be used exactly once. The result in Model (1) confirms the earlier finding, that is, banks create less liquidity when engaging in nontraditional activities. The matching estimator presented above mitigates selection bias. However, there could
Jo
be unobservable factors that explain decisions to diversify. I use Heckman’s two-step approach to eliminate bias due to unobservable variables. I first model the selection of diversification by using the logit selection model and then obtain the inverse Mills ratio (IMR), the omitted variable in Equation (1). Following Laeven and Levine (2007), I use the average non-interest income of other banks as an IV. I then estimate the logit diversification choice model and calculate IMR. The IMR measure is the conditional expectation of the model selection error term, given banks’ observable characteristics and decision to diversify. In the second stage, I re21
estimate Equation (1) by including IMR as an additional control variable to correct for potential self-selection bias. Models (2) and (3) in Table 8 present the maximum likelihood estimates of the logit diversification choice and baseline model augmented by IMR. Consistent with the core findings, I still document a negative and significant coefficient of NII. I continue the inquiry into the endogeneity of the diversification decision with an IV estimation. As above, the instrument is the average non-interest income of other banks. I report the first- and second-stage IV regression results in Models (4) and (5) of Table 8, respectively. The Kleibergen–Paap underidentification and Cragg–Donald weak identification test statistics
of
further indicate that the selected IV is relevant.4 The results of the second stage also support the earlier finding. I observe that the coefficient in the IV estimation is much larger than in the OLS
ro
estimate (-0.621 vs. -0.106, both significant at 1%), which is consistent with concerns regarding reverse causality and, hence, with the need to use an IV approach to identify the impact of
-p
diversification on bank liquidity creation (Tran, Hassan, and Houston 2019). The OLS estimation could yield coefficient estimates of the impact of NII on LC that are biased toward zero, whereas
re
the IV estimation yields a more accurate (and larger) impact of NII on LC. It is worth noting that, in Models (2) and (4), the IV is positively and significantly associated with the decision to
lP
diversify. This confirms the relevance of the IV.
I end the examination with an autoregressive distributed lag model using system generalized method of moments (system-GMM) estimators to address the persistence of the
ur na
endogeneity bias.5 The regression also includes time fixed effects, following Roodman (2009). The results are reported in Table 8, Panel B. I first discuss the diagnostic tests. The Hansen test (chi-squared test) of overidentification checks the validity of the instruments (i.e., exogeneity). The p-value of the Hansen test suggests that the null hypotheses of overidentification cannot be rejected, that is, the instruments are valid. The results of tests for autocorrelation in the residuals
Jo
AR(1) and AR(2) are also reported (the null hypothesis being no serial correlation of order 1 or 2 in the first-difference residuals), and the residuals show no evidence of second-order autocorrelation. I still find a negative and statistically significant coefficient on NII, the variable of interest, consistent with the previous findings.
4 5
I did not perform the overidentification test, since my model is only identified. I am grateful to an anonymous referee for this suggestion.
22
In brief, whether estimating a Heckman selection model, employing an IV approach, using PSM, or using system-GMM, the findings remain unchanged: diversification per se lowers bank liquidity creation. 6. Conclusions This study investigates the impacts of functional diversification on banks’ liquidity creation, using a large sample of US banks from 2001:Q1 to 2015:Q4. The basic regressions suggest that banks with an increased share of non-interest income decrease their liquidity creation, which is
of
consistent with the diversification–liquidity contraction channel. Banks benefit from
diversification, but these gains are quickly offset by adverse effects from reliance on riskier
ro
assets. Banks with a relatively small share of non-interest income do not seem to change their liquidity creation when relying more on non-interest income–generating activities. However,
-p
with greater reliance on non-interest income–generating activities, banks reduce their liquidity creation with increases in the share of non-interest income. The empirical findings indicate
re
activity strategies not only affect the conditional average of liquidity creation, but also influence the distribution of liquidity creation. When performing analysis across bank sizes, I observe that large banks that engage in non-interest income–generating activities reduce their liquidity
lP
creation the most, followed by small diversified banks. Medium-sized banks reduce their liquidity creation the least when they increase their share of non-interest income. Finally, I
ur na
document evidence of the bright side of diversification during a crisis. Diversified banks create more liquidity during times of crisis. This effect can be clearly observed among medium-sized and small banks during the late stage of crisis. From a policy perspective, I believe that this study is of interest to regulators and policymakers. After the last financial crisis, various initiatives, such as the Volcker Rule in the
Jo
United States, Vickers Initiatives in the United Kingdom, the Liikanen Report in the European Union, and, recently, the call for the 21st Century Glass–Steagall Act propose narrow banking policies that aim to diminish banks’ ability to diversify across product lines (Tran et al. 2019). The evidence shows that, during the normal times, diversified banks, especially large ones, create less liquidity. However, during crises, there could be a bright side, allowing medium-sized and small banks to expand into non-interest–generating activities, since banks could become less risky during times of turmoil, leading to a rise in liquidity creation and thereby contributing to the recovery of the real economy. Taken together, this study suggests that ring-fencing initiatives 23
could help mitigate the negative effects of diversification on liquidity creation during normal
Jo
ur na
lP
re
-p
ro
of
times, but at the price of lowering the benefits of diversification in times of crisis.
24
References Acharya, Viral V., and Nada Mora, 2015, A Crisis of Banks as Liquidity Providers, The Journal of Finance 70, 1–43. Acharya, Viral V., Iftekhar Hasan, and Anthony Saunders, 2006, Should Banks Be Diversified? Evidence from Individual Bank Loan Portfolios, The Journal of Business 79, 1355–1412. Amihud, Yakov, and Baruch Lev, 1981, Risk reduction as a managerial motive for conglomerate mergers, The bell journal of economics, 605–617.
of
Andreou, Panayiotis C., Dennis Philip, and Peter Robejsek, 2016, Bank Liquidity Creation and Risk-Taking: Does Managerial Ability Matter?, Journal of Business Finance & Accounting 43, 226–259.
ro
Berger, Allen N., and Christa H. S. Bouwman, 2013, How does capital affect bank performance during financial crises?, Journal of Financial Economics 109, 146–176.
-p
Berger, Allen N., and Christa H. S. Bouwman, 2017, Bank liquidity creation, monetary policy, and financial crises, Journal of Financial Stability 30, 139–155.
re
Berger, Allen N., Christa H. S. Bouwman, Thomas Kick, and Klaus Schaeck, 2016, Bank liquidity creation following regulatory interventions and capital support, Journal of Financial Intermediation 26, 115–141.
lP
Berger, Allen N., and Christa HS Bouwman, 2009, Bank liquidity creation, Review of Financial Studies 22, 3779–3837.
ur na
Berger, Allen N., Iftekhar Hasan, and Mingming Zhou, 2010, The effects of focus versus diversification on bank performance: Evidence from Chinese banks, Journal of Banking & Finance 34. Performance Measurement in the Financial Services Sector, 1417–1435. Berger, Allen N., and John Sedunov, 2017, Bank liquidity creation and real economic output, Journal of Banking & Finance 81, 1–19.
Jo
Berger, Philip G., and Eli Ofek, 1995, Diversification’s effect on firm value, Journal of financial economics 37, 39–65. Bharath, Sreedhar, Sandeep Dahiya, Anthony Saunders, and Anand Srinivasan, 2007, So what do I get? The bank’s view of lending relationships, Journal of Financial Economics 85. The economics of conflicts of interest financial institutions, 368–419. Bhattacharya, Sudipto, and Anjan V. Thakor, 1993, Contemporary banking theory, Journal of financial Intermediation 3, 2–50. Brewer, Elijah, 1989, Relationship between bank holding company risk and nonbank activity, Journal of Economics and Business 41, 337–353. 25
Bryant, John, 1980, A model of reserves, bank runs, and deposit insurance, Journal of Banking & Finance 4, 335–344. Campa, Jose Manuel, and Simi Kedia, 2002, Explaining the Diversification Discount, The Journal of Finance 57, 1731–1762. Campello, Murillo, 2002, Internal Capital Markets in Financial Conglomerates: Evidence from Small Bank Responses to Monetary Policy, The Journal of Finance 57, 2773–2805. Choi, In, 2001, Unit root tests for panel data, Journal of International Money and Finance 20, 249–272.
ro
of
De Jonghe, Olivier, 2010, Back to the basics in banking? A micro-analysis of banking system stability, Journal of Financial Intermediation 19. Risk Transfer Mechanisms and Financial Stability, 387–417.
-p
De Jonghe, Olivier, Maaike Diepstraten, and Glenn Schepens, 2015, Banks’ size, scope and systemic risk: What role for conflicts of interest?, Journal of Banking & Finance 61, Supplement 1. Global Trends in Banking, Regulations, and Financial Markets, S3–S13.
re
Demsetz, Rebecca S., and Philip E. Strahan, 1997, Diversification, size, and risk at bank holding companies, Journal of money, credit, and banking, 300–313.
lP
Denis, David J., Diane K. Denis, and Atulya Sarin, 1997, Agency Problems, Equity Ownership, and Corporate Diversification, Journal of Finance 52, 135–160. DeYoung, Robert, and Karin P. Roland, 2001, Product Mix and Earnings Volatility at Commercial Banks: Evidence from a Degree of Total Leverage Model, Journal of Financial Intermediation 10, 54–84.
ur na
DeYoung, Robert, and Gökhan Torna, 2013, Nontraditional banking activities and bank failures during the financial crisis, Journal of Financial Intermediation 22, 397–421. Diamond, Douglas W., 1984, Financial intermediation and delegated monitoring, The Review of Economic Studies 51, 393–414.
Jo
Diamond, Douglas W., and Philip H. Dybvig, 1983, Bank runs, deposit insurance, and liquidity, The journal of political economy, 401–419. Diamond, Douglas W., and Raghuram G. Rajan, 2000, A theory of bank capital, The Journal of Finance 55, 2431–2465. Diamond, Douglas W., and Raghuram G. Rajan, 2001, Liquidity Risk, Liquidity Creation, and Financial Fragility: A Theory of Banking, Journal of Political Economy 109, 287. Distinguin, Isabelle, Caroline Roulet, and Amine Tarazi, 2013, Bank regulatory capital and liquidity: Evidence from US and European publicly traded banks, Journal of Banking & Finance 37, 3295–3317. 26
Ellul, Andrew, and Vijay Yerramilli, 2013, Stronger Risk Controls, Lower Risk: Evidence from U.S. Bank Holding Companies, The Journal of Finance 68, 1757–1803. Elsas, Ralf, Andreas Hackethal, and Markus Holzhäuser, 2010, The anatomy of bank diversification, Journal of Banking & Finance 34, 1274–1287. Financial Crisis Inquiry Commission, 2011, The Financial Crisis Inquiry Report (Cosimo, Inc.). Foos, Daniel, Lars Norden, and Martin Weber, 2010, Loan growth and riskiness of banks, Journal of Banking & Finance 34. International Financial Integration, 2929–2940.
of
Gandhi, Priyank, Patrick Christian Kiefer, and Alberto Plazzi, 2016, A False Sense of Security: Why U.S. Banks Diversify and Does it Help? SSRN Scholarly Paper, Social Science Research Network, Rochester, NY.
ro
Heckman, James J., Hidehiko Ichimura, and Petra E. Todd, 1997, Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme, The review of economic studies 64, 605–654.
-p
Holmström, Bengt, and Jean Tirole, 1998, Private and public supply of liquidity, Journal of political Economy 106, 1–40.
re
Horváth, Roman, Jakub Seidler, and Laurent Weill, 2013, Bank Capital and Liquidity Creation: Granger-Causality Evidence, Journal of Financial Services Research 45, 341–361.
lP
Horvath, Roman, Jakub Seidler, and Laurent Weill, 2016, How bank competition influences liquidity creation, Economic Modelling 52. Special Issue on Recent Developments in DecisionMaking, Monetary Policy and Financial Markets, 155–161.
ur na
Hou, Xiaohui, Shuo Li, Wanli Li, and Qing Wang, 2018, Bank diversification and liquidity creation: Panel Granger-causality evidence from China, Economic Modelling 71, 87–98. Houston, Joel, Christopher James, and David Marcus, 1997, Capital market frictions and the role of internal capital markets in banking, Journal of Financial Economics 46, 135–164. Jensen, Michael C., 1986, Agency costs of free cash flow, corporate finance, and takeovers, The American economic review, 323–329.
Jo
Jensen, Michael C., and Kevin J. Murphy, 1990, Performance pay and top-management incentives, Journal of political economy, 225–264. Jiang, Liangliang, Ross Levine, and Chen Lin, 2017, Competition and Bank Liquidity Creation. SSRN Scholarly Paper, Social Science Research Network, Rochester, NY. Jiménez, Gabriel, and Jesus Saurina, 2004, Collateral, type of lender and relationship banking as determinants of credit risk, Journal of banking & Finance 28, 2191–2212.
27
Kashyap, Anil K., Raghuram Rajan, and Jeremy C. Stein, 2002, Banks as Liquidity Providers: An Explanation for the Coexistence of Lending and Deposit-taking, The Journal of Finance 57, 33–73. Kashyap, Anil K, and Jeremy C. Stein, 2000, What Do a Million Observations on Banks Say about the Transmission of Monetary Policy?, The American Economic Review 90, 407–428. Klomp, Jeroen, and Jakob de Haan, 2012, Banking risk and regulation: Does one size fit all?, Journal of Banking & Finance 36. Systemic risk, Basel III, global financial stability and regulation, 3197–3212.
of
Krishnaswami, Sudha, Paul A Spindt, and Venkat Subramaniam, 1999, Information asymmetry, monitoring, and the placement structure of corporate debt, Journal of Financial Economics 51, 407–434.
ro
Kwast, Myron L., 1989, The impact of underwriting and dealing on bank returns and risks, Journal of Banking & Finance 13, 101–125.
-p
Laeven, Luc, and Ross Levine, 2007, Is there a diversification discount in financial conglomerates?, Journal of Financial Economics 85, 331–367.
re
Levine, Ross, and Sara Zervos, 1998, Stock Markets, Banks, and Economic Growth, The American Economic Review 88, 537–558.
lP
Litan, Robert, 1985, Evaluating and Controlling the Risks of Financial Product Deregulation, Yale Journal on Regulation 3. Roodman, David, 2009, How to do Xtabond2: An Introduction to Difference and System GMM in Stata, The Stata Journal 9, 86–136.
ur na
Rosenbaum, Paul R., and Donald B. Rubin, 1983, The central role of the propensity score in observational studies for causal effects, Biometrika 70, 41–55. Santomero, Anthony M., and David L. Eckles, 2000, The Determinants of Success in the New Financial Services Environment: Now that Firms Can Do Everything, What Should They Do and Why Should Regulators Care? SSRN Scholarly Paper, Social Science Research Network, Rochester, NY.
Jo
Sanya, Sarah, and Simon Wolfe, 2011, Can Banks in Emerging Economies Benefit from Revenue Diversification?, Journal of Financial Services Research 40, 79–101. Saunders, Anthony, and Marcia Millon Cornett, 2008, Financial Institutions Management: A Risk Management Approach (McGraw-Hill Education). Saunders, Anthony, and Ingo Walter, 1994, Universal Banking in the United States: What Could We Gain? What Could We Lose? (Oxford University Press).
28
Shim, Jeungbo, 2013, Bank capital buffer and portfolio risk: The influence of business cycle and revenue diversification, Journal of Banking & Finance 37, 761–772. Shleifer, Andrei, and Robert W. Vishny, 1989, Management entrenchment: The case of manager-specific investments, Journal of financial economics 25, 123–139. Shleifer, Andrei, and Robert W. Vishny, 2010, Unstable banking, Journal of Financial Economics 97. The 2007-8 financial crisis: Lessons from corporate finance, 306–318. Stiroh, Kevin J., 2004, Diversification in Banking: Is Noninterest Income the Answer?, Journal of Money, Credit and Banking 36, 853–882.
of
Stiroh, Kevin J., and Adrienne Rumble, 2006, The dark side of diversification: The case of US financial holding companies, Journal of Banking & Finance 30, 2131–2161.
ro
Tran, Dung V., Isabelle Girerd-Potin, Pascal Louvet, and Kabir M. Hassan, 2017, Activity strategies, agency problems and bank risk. SSRN Scholarly Paper, Rochester, NY.
-p
Tran, Dung Viet, M.K Hassan, and Reza Houston, 2019, How does listing status affect bank risk? The effects of crisis, market discipline and regulatory pressure on listed and unlisted BHCs, The North American Journal of Economics and Finance 49, 85–103.
Jo
ur na
lP
re
Yasuda, Ayako, 2005, Do Bank Relationships Affect the Firm’s Underwriter Choice in the Corporate-Bond Underwriting Market?, The Journal of Finance 60, 1259–1292.
29
.1
.2
.3
.4
.5
.6
.7
Jo
ur na
lP
re
Quantile
-p
-0.35
ro
-0.30
of
NII
-0.25
-0.20
Figure 1. Plots of coefficients for non-interest income across quantiles based on quantile regression results in Table (4)
30
.8
.9
Table 1. Variables Definitions This table presents definitions of all main variables used in the analysis. Definitions
LOAN TRADING DIV SIZE
Book value of equity over gross total assets Income before taxes, provisions recognized in income over gross total assets
Growth rate of gross total assets
Standard deviation of EARNINGS over the previous 12 quarters
Nonperforming assets over the quarter, scaled by total loans 𝐶𝐴𝑃+𝜇𝑅𝑂𝐴 A bank measure of financial risk calculated as ; a larger value 𝜎𝑅𝑂𝐴
indicates lower overall bank risk; means of ROA and Equity/GTA as well as the standard deviation of ROA are computed over the previous 12 quarters (t 11 to t) Average of NII of all other banks A dummy equal to 1 for a period from 2007:Q3-2009:Q2, and 0 otherwise. A dummy equal to 1 for a period from 2007:Q3-2008:Q2, and 0 otherwise. A dummy equal to 1 for a period from 2008:Q3-2009:Q2, and 0 otherwise. Bank fixed effects Time fixed effects, represented by dummies for each quarter of the sample period.
Jo
NII_AVG CRISIS CRISIS_1 CRISIS_2 BFE TFE
Non-interest incomes over the net operating incomes In each quarter, I rank NII variable into quartiles and create a variable called NII_QUARTILE, which takes value ranging from 1 (low) to 4 (high) Loans over gross total assets Including trading revenues, interest income from trading assets and the realized gains or losses from the held-to-maturity and available-for-sale securities 𝐷𝐼𝑉_1 = 1 − [(𝑁𝐼𝐼)2 + (𝑁𝐼𝑀)2 ] The natural logarithm of gross total assets
ur na
CAPITAL EARNINGS GROWTH SD(EARNINGS) NPL ZSCORE
of
Control variables NII NII_QUARTILE
ro
LC_ON
-p
Deviation LC LC_OFF
A bank’s total bank liquidity creation measure including on- and off-balance sheet activities normalized by the gross total assets (GTA) of a bank. For a more detailed definition, see Berger and Bouwman (2009). LCi,t – Average LCi,t of the industry A bank’s bank liquidity creation measure including only off-balance sheet activities normalized by the total asset size of a bank. For a more detailed definition, see to Berger and Bouwman (2009). A bank’s bank liquidity creation measure including only on-balance sheet activities normalized by the total asset size of a bank. For a more detailed definition, see to Berger and Bouwman (2009).
re
Dependent variables LC
lP
Variables
31
Table 2. Summary Statistics
Panel B: Correlation SIZE
1.000 0.3728*** 0.0762*** 0.1514*** 0.0163*** -0.0106** -0.0615*** 0.1586***
1.000 0.0550*** 0.1142*** 0.0436*** 0.0617*** 0.0461*** 0.1086***
ro
CAP
lP
NII
ur na
LC 1.000 -0.1055*** 0.1380*** -0.1915*** 0.1299*** 0.1214*** -0.1157*** -0.0200*** -0.0924***
1.000 0.3575*** -0.0280*** '-0.1311*** 0.2277*** 0.0334***
Min (0.027) 0.000 12.089 0.019 (0.020) (0.085) 0.465 0.001
Max 0.899 0.814 19.109 0.220 0.051 0.229 0.118 197.907 0.047
EARNINGS
GROWTH
NPL
ZSCORE
SD(EARNINGS)
1.000 0.1515*** -0.3612*** 0.2575*** -0.1770***
1.000 -0.2236*** 0.1078*** -0.1157***
1.000 -0.3868*** 0.3532***
1.000 -0.4209***
1.000
Jo
LC NII SIZE CAP EARNINGS GROWTH NPL ZSCORE SD(EARNINGS)
Std. Dev. 0.174 0.131 1.237 0.028 0.009 0.042 0.022 37.935 0.007
-p
Mean 0.456 0.235 13.934 0.090 0.015 0.016 0.018 43.508 0.005
re
Obs 34,941 34,309 34,941 34,941 34,941 34,941 34,941 34,941 34,941
LC NII SIZE CAP EARNINGS GROWTH NPL ZSCORE SD(EARNINGS)
of
This table reports summary statistics and correlation for the main sample of U.S. commercial banks used in the analysis. The sample period is from 2001:Q1 to 2015:Q4. All financial variables are winsorized at 1% and 99% levels. Panel A: Summary statistics
32
Table 3. Baseline Multivariate Analysis
EARNINGS GROWTH NPL ZSCORE
BFE TFE Obs Adj R2 # Banks
-0.108*** (0.025) -0.029** (0.013) -0.264** (0.124) 2.096*** (0.190) 0.002 (0.014) -0.669*** (0.119) 0.000 (0.000) -0.479 (0.303) 0.890*** (0.186) Yes Yes 32,600 0.239 1,767
Jo
Constant
ur na
SD(EARNINGS)
-0.158*** (0.061) 0.021 (0.029) 0.015 (0.284) 3.012*** (0.476) 0.005 (0.027) -0.362 (0.315) -0.000 (0.000) 0.471 (0.509) 0.180 (0.426) Yes Yes 8,469 0.211 210
-p
CAPITAL
-0.206*** (0.036) 0.033*** (0.003) -2.174*** (0.152) 4.779*** (0.660) 2.792*** (0.244) 0.847*** (0.307) -0.001*** (0.000) 1.822** (0.875) 0.100** (0.044) Yes Yes 1,829 0.260
re
SIZE
-0.106*** (0.024) -0.026* (0.013) -0.274** (0.122) 2.086*** (0.189) 0.003 (0.015) -0.664*** (0.119) 0.000 (0.000) -0.483 (0.304) 0.850*** (0.184) Yes Yes 32,946 0.235 1,767
lP
NII
ro
of
This table reports regression estimates of the relation between NII and LC. The sample period is from 2001:Q1 to 2015:Q4. The main independent variable is NII. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics. Baseline model Average analysis Balanced data Excluded M&A Newey-West Fama-MacBeth Two-way Clustering (1) (2) (3) (4) (5) (6) (7)
33
-0.273*** (0.015) 0.028*** (0.001) -1.626*** (0.062) 4.473*** (0.188) 0.312*** (0.025) -0.060 (0.084) -0.000*** (0.000) -0.507* (0.263) 0.211*** (0.020) Yes Yes 32,946 0.146
-0.263*** (0.018) 0.025*** (0.001) -1.350*** (0.116) 4.198*** (0.626) 0.160** (0.065) -0.450** (0.169) -0.001*** (0.000) -0.103 (0.431) 0.212*** (0.016) Yes Yes 32,946 0.159
-0.273*** (0.042) 0.028*** (0.004) -1.626*** (0.179) 4.473*** (0.554) 0.312*** (0.055) -0.060 (0.199) -0.000*** (0.000) -0.507 (0.626) 0.211*** (0.057) Yes Yes 32,946 0.146
Table 4. Quantile regression
GROWTH NPL SD(EARNINGS) ZSCORE Constant
Jo
Observations Pseudo R2
34
0.6 (6) -0.240*** (0.008) 0.035*** (0.001) -1.533*** (0.051) 4.081*** (0.136) 0.271*** (0.029) -1.089*** (0.046) -1.182*** (0.134) -0.000*** (0.000) 0.192*** (0.010) 34,309 0.088
ro
0.5 (5) -0.242*** (0.009) 0.035*** (0.001) -1.547*** (0.059) 3.858*** (0.164) 0.276*** (0.031) -1.005*** (0.070) -1.216*** (0.158) -0.001*** (0.000) 0.156*** (0.012) 34,309 0.083
-p
EARNINGS
0.4 (4) -0.247*** (0.012) 0.035*** (0.001) -1.637*** (0.053) 3.853*** (0.166) 0.248*** (0.034) -0.898*** (0.070) -1.449*** (0.188) -0.001*** (0.000) 0.132*** (0.015) 34,309 0.08
re
CAP
0.3 (3) -0.246*** (0.011) 0.034*** (0.001) -1.627*** (0.058) 3.303*** (0.218) 0.258*** (0.031) -0.685*** (0.087) -2.123*** (0.224) -0.001*** (0.000) 0.107*** (0.018) 34,309 0.078
lP
SIZE
0.2 (2) -0.249*** (0.013) 0.033*** (0.002) -1.663*** (0.049) 3.309*** (0.260) 0.237*** (0.036) -0.418*** (0.071) -1.985*** (0.179) -0.001*** (0.000) 0.065*** (0.020) 34,309 0.080
ur na
NII
Quantile 0.1 (1) -0.306*** (0.019) 0.027*** (0.003) -1.798*** (0.057) 3.812*** (0.315) 0.183*** (0.050) 0.033 (0.099) -1.149*** (0.343) -0.000*** (0.000) 0.064* (0.036) 34,309 0.083
of
This table reports regression estimates of the relation between NII and LC using quantile regressions. The sample period is from 2001:Q1 to 2015:Q4. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Numbers in parentheses are t-statistics.
0.7 (7) -0.243*** (0.009) 0.034*** (0.001) -1.448*** (0.047) 4.177*** (0.118) 0.275*** (0.043) -1.169*** (0.058) -0.940*** (0.177) -0.000*** (0.000) 0.234*** (0.012) 34,309 0.094
0.8 (8) -0.242*** (0.008) 0.032*** (0.001) -1.380*** (0.041) 4.458*** (0.146) 0.311*** (0.037) -1.280*** (0.055) -0.275 (0.267) -0.000*** (0.000) 0.294*** (0.011) 34,309 0.099
0.9 (9) -0.250*** (0.014) 0.030*** (0.001) -1.280*** (0.063) 4.893*** (0.147) 0.332*** (0.040) -1.612*** (0.065) 1.167*** (0.334) -0.001*** (0.000) 0.370*** (0.018) 34,309 0.103
Table 5. The effect of diversification on liquidity creation for different size range This table reports regression estimates of the relation between NII and LC. The sample period is from 2001:Q1 to 2015:Q4. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics. Small Medium Large (1) (2) (3)
EARNINGS GROWTH NPL ZSCORE SD(EARNINGS) Constant
Jo
ur na
lP
re
BFE TFE Obs Adj R2 # Banks
35
-0.148*** (0.056) -0.022 (0.055) 1.285*** (0.454) 1.894*** (0.424) 0.043 (0.032) -0.250 (0.370) 0.000 (0.000) -0.299 (0.650) 0.773 (0.916) Yes Yes 3,511 0.274 154
of
CAPITAL
-0.079** (0.038) 0.011 (0.020) 0.040 (0.193) 1.928*** (0.303) -0.018 (0.023) -0.440** (0.210) -0.000 (0.000) -0.164 (0.433) 0.332 (0.285) Yes Yes 11,103 0.224 530
ro
SIZE
-0.109*** (0.029) -0.073*** (0.016) -0.865*** (0.154) 2.065*** (0.228) -0.005 (0.019) -0.973*** (0.121) 0.000*** (0.000) -0.937* (0.519) 1.476*** (0.209) Yes Yes 18,332 0.261 1,367
-p
NII
Table 6. The effects of the crisis
CAPITAL EARNINGS GROWTH NPL ZSCORE SD(EARNINGS) Constant
SMALL (6)
MEDIUM (7)
LARGE (8)
CRISIS 2 ALL (9)
SMALL (10)
MEDIUM (11)
LARGE (12)
-0.112*** (0.024) 0.037** (0.019) -0.035*** (0.008) -0.025* (0.013) -0.270** (0.122) 2.082*** (0.189) 0.002 (0.015) -0.666*** (0.119) 0.000 (0.000) -0.487 (0.304) 0.844*** (0.183) Yes Yes 32,946 0.236 1,767
-0.114*** (0.030) 0.034 (0.026) -0.015* (0.008) -0.072*** (0.015) -0.864*** (0.153) 2.066*** (0.227) -0.006 (0.019) -0.974*** (0.121) 0.000*** (0.000) -0.940* (0.519) 1.470*** (0.208) Yes Yes 18,332 0.261 1,367
-0.084** (0.038) 0.039 (0.026) -0.036*** (0.014) 0.012 (0.020) 0.044 (0.194) 1.923*** (0.301) -0.019 (0.023) -0.446** (0.209) -0.000 (0.000) -0.163 (0.433) 0.319 (0.284) Yes Yes 11,103 0.225 530
-0.149*** (0.056) 0.009 (0.031) -0.010 (0.021) -0.022 (0.055) 1.288*** (0.456) 1.888*** (0.422) 0.043 (0.032) -0.249 (0.370) 0.000 (0.000) -0.304 (0.651) 0.773 (0.916) Yes Yes 3,511 0.274 154
-0.106*** (0.024) 0.010 (0.019) 0.006 (0.007) -0.026* (0.013) -0.273** (0.122) 2.086*** (0.189) 0.003 (0.015) -0.665*** (0.119) 0.000 (0.000) -0.484 (0.304) 0.848*** (0.184) Yes Yes 32,946 0.235 1,767
-0.109*** (0.030) 0.001 (0.032) 0.008 (0.008) -0.073*** (0.015) -0.865*** (0.153) 2.065*** (0.228) -0.005 (0.019) -0.973*** (0.121) 0.000*** (0.000) -0.937* (0.519) 1.476*** (0.208) Yes Yes 18,332 0.261 1,367
-0.077** (0.038) -0.020 (0.027) 0.021* (0.011) 0.011 (0.020) 0.038 (0.193) 1.930*** (0.302) -0.017 (0.023) -0.437** (0.210) -0.000 (0.000) -0.161 (0.433) 0.336 (0.285) Yes Yes 11,103 0.224 530
-0.148*** (0.056) 0.001 (0.034) -0.007 (0.021) -0.022 (0.055) 1.285*** (0.456) 1.894*** (0.424) 0.043 (0.032) -0.250 (0.370) 0.000 (0.000) -0.300 (0.650) 0.773 (0.916) Yes Yes 3,511 0.274 154
-0.111*** (0.024) 0.055*** (0.018) -0.039*** (0.008) -0.026* (0.013) -0.274** (0.122) 2.084*** (0.189) 0.003 (0.015) -0.663*** (0.119) 0.000 (0.000) -0.480 (0.304) 0.847*** (0.183) Yes Yes 32,946 0.236 1,767
-0.114*** (0.030) 0.055** (0.023) -0.019** (0.008) -0.073*** (0.016) -0.867*** (0.154) 2.067*** (0.227) -0.006 (0.019) -0.972*** (0.121) 0.000*** (0.000) -0.937* (0.522) 1.474*** (0.209) Yes Yes 18,332 0.262 1,367
-0.086** (0.038) 0.084*** (0.027) -0.047*** (0.014) 0.012 (0.020) 0.041 (0.193) 1.930*** (0.299) -0.019 (0.023) -0.438** (0.209) -0.000 (0.000) -0.148 (0.433) 0.320 (0.284) Yes Yes 11,103 0.226 530
-0.149*** (0.056) 0.015 (0.039) -0.009 (0.023) -0.022 (0.055) 1.286*** (0.455) 1.886*** (0.422) 0.043 (0.032) -0.248 (0.371) 0.000 (0.000) -0.298 (0.651) 0.773 (0.916) Yes Yes 3,511 0.274 154
ro
CRISIS 1 ALL (5)
Jo
BFE TFE Observations Adj R2 # Banks
LARGE (4)
-p
SIZE
MEDIUM (3)
re
CRISIS
SMALL (2)
lP
NII*CRISIS
CRISIS ALL (1)
ur na
NII
of
This table reports regression estimates of the relation between NII and LC during the crisis time. The sample period is from 2001:Q1 to 2015:Q4. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics.
36
Table 7. Alternative Measures of Liquidity creation and Diversification This table reports regression estimates of the relation between NII and LC using alternative measures of liquidity creation (Panel A), and alternative measures of diversification (Panel B). Panel C reports the net effects of diversification on liquidity creation. The sample period is from 2001:Q1 to 2015:Q4. All regressions include time (quarter) fixed effects, and bank fixed effects. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics. Panel A: Alternative measures of liquidity creation
NPL ZSCORE SD(EARNINGS) Constant
Jo
ur na
BFE TFE Observations Adj R2 # Banks
-0.009 (0.011) -0.003 (0.004) -0.024 (0.037) 0.536*** (0.058) 0.021*** (0.006) -0.293*** (0.035) 0.000*** (0.000) -0.342*** (0.113) 0.142** (0.058) Yes Yes 32,946 0.365 1,767
of
GROWTH
-0.099*** (0.016) -0.023** (0.010) -0.248** (0.101) 1.540*** (0.155) -0.021* (0.011) -0.351*** (0.094) -0.000 (0.000) -0.085 (0.226) 0.710*** (0.135) Yes Yes 32,946 0.127 1,767
ro
EARNINGS
-0.106*** (0.024) -0.026* (0.013) -0.274** (0.122) 2.086*** (0.189) 0.003 (0.015) -0.664*** (0.119) 0.000 (0.000) -0.483 (0.304) 0.429** (0.184) Yes Yes 32,946 0.131 1,767
-p
CAPITAL
LC_OFF (3)
re
SIZE
LC_ON (2)
lP
NII
Deviation LC (1)
37
Panel B: Alternative measures of diversification
LOAN (3)
-0.006*** (0.001)
-0.065*** (0.011)
0.917*** (0.022)
HHI
EARNINGS GROWTH NPL
lP
ZSCORE SD(EARNINGS)
HHI and NON (5)
-0.053** (0.021) -0.024* (0.013) -0.256** (0.124) 1.913*** (0.184) -0.002 (0.016) -0.713*** (0.124) 0.000 (0.000) -0.516* (0.304) 0.821*** (0.187) Yes Yes 32,946 0.230 1,767
-0.140*** (0.048) 0.088** (0.045) -0.028** (0.013) -0.335*** (0.119) 1.757*** (0.179) -0.002 (0.014) -0.751*** (0.102) 0.014 (0.258) 0.000 (0.000) 0.850*** (0.177) Yes Yes 34,309 0.239 1,828
-0.014 (0.009) -0.209** (0.084) 0.724*** (0.122) 0.107*** (0.013) -0.318*** (0.085) 0.000 (0.000) -0.378* (0.215) 0.079 (0.122) Yes Yes 33,560 0.546 1,767
Jo
ur na
Constant BFE TFE Observations Adj R2 # Banks
-0.022* (0.013) -0.229* (0.122) 1.933*** (0.182) -0.008 (0.016) -0.642*** (0.119) 0.000 (0.000) -0.576* (0.300) 0.777*** (0.177) Yes Yes 33,559 0.240 1,767
-p
CAPITAL
-0.025* (0.013) -0.260** (0.123) 1.936*** (0.183) -0.001 (0.016) -0.703*** (0.123) 0.000 (0.000) -0.505* (0.304) 0.834*** (0.186) Yes Yes 32,946 0.232 1,767
re
SIZE
HHI (4)
of
TRADING (2)
ro
NII
NII_QUARTILE (1)
38
25th -0.140*** 0.122** -0.018
50th -0.140*** 0.101** -0.039*
75th -0.140*** 0.076** -0.064***
90th -0.140*** 0.043** -0.097***
Jo
ur na
lP
re
-p
Direct effect Indirect effect Net effect
ro
Non-interest income 10th -0.140*** 0.140** 0.000
of
Panel C: Effects of Changes in Noninterest Income on Bank Liquidity Creation This table reports estimation of first derivative of LC on NII , based on regression results reported in Table 7, Panel B, Model (5) and evaluated at different values of the noninterest share based on percentile ranks (10th, 25th, 50th, 75th, and 90th percentile). Direct effect is estimated impact of a 1% increase in the noninterest income share. Indirect effect is estimated impact of a change in revenue diversification from a 1% increase in the noninterest income share. Net effect sums the direct and indirect effects. Robust standard errors are in parentheses. ***, **, * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
39
Table 8. Endogeneity concerns
SIZE CAPITAL EARNINGS GROWTH NPL ZSCORE
NII_AVG IMR Constant
0.913*** (0.225) Yes Yes 6,654 0.300 1,382
0.004* (0.002) 0.819*** (0.191) Yes Yes 31,106 0.241 1,712
ur na
SD(EARNINGS)
-0.119*** (0.027) -0.023* (0.014) -0.275** (0.125) 2.127*** (0.192) -0.000 (0.015) -0.650*** (0.121) 0.000 (0.000) -0.506* (0.305)
Jo
BFE TFE Observations Adj R2 # Banks Underidentification test Kleibergen-Paap Wald F-Stat Weak identification test Cragg-Donald Wald F-Stat Weak instrument robust inference Anderson-Rubin Wald test
-p
-0.090** (0.042) -0.029* (0.017) -0.279* (0.166) 1.896*** (0.278) -0.041 (0.029) -0.836*** (0.130) 0.000 (0.000) -0.584 (0.431)
0.152*** (0.033) 0.762 (1.123) 2.238 (2.975) -0.670** (0.331) -12.568*** (1.368) 0.005*** (0.001) -24.182*** (3.919) 9.599*** (1.690)
re
NII
Heckman 2nd step (3)
lP
(1)
Heckman 1rst step (2)
ro
PSM
of
Panel A: The table reports regression estimates of the relation between NII and LC. The sample period is from 2001:Q1 to 2015:Q4. Models (1) reports results from one-to-one matching PSM. Models (2)-(3) present estimations of Heckman selection model. Models (4)-(5) present estimations of IV estimations. All financial variables are winsorized at the 1% and 99% levels. ***, **, * indicate significance at the 1%, 5%, and 10% level respectively. Standard errors are clustered at the bank level. Numbers in parentheses are t-statistics.
-9.080*** (1.616) Yes Yes 33,622 1771
IV 1rst step (4)
0.038*** (0.003) 0.261 (0.165) 0.507 (0.387) -0.027 (0.029) 0.169 (0.186) -0.001*** (0.000) 3.643*** (0.723) 0.097*** (0.028)
-0.394*** (0.049) Yes Yes 29,592 0.196 1688
IV 2nd step (5) -0.621*** (0.184) 0.042*** (0.009) -1.288*** (0.182) 3.592*** (0.432) 0.186*** (0.034) 0.214 (0.204) -0.001*** (0.000) 1.013 (0.882)
0.089 (0.089) Yes Yes 28,184 0.126 1638
427.719*** 511.577*** 399.99***
40
Panel B: The system-GMM
of
System GMM (1)
ro
LC NII
-p
SIZE CAPITAL
re
EARNINGS GROWTH
lP
NPL ZSCORE
Constant
ur na
SD(EARNINGS)
Jo
Observations AR (1) (p-value) AR (2) (p-value) Hansen test (p-value)
-0.235*** (0.018) -0.366*** (0.017) 0.187*** (0.017) 2.415*** (0.137) 4.904*** (0.202) -0.286*** (0.015) -9.085*** (0.212) -0.003*** (0.000) 13.501*** (0.272) -2.059*** (0.236) 28,962 0.000 0.144 0.576
41