1 Bank Liquidity Mismatch, Strategic Complementarity and Deposit Flows * Qi Chen Duke University, 100 Fuqua Drive, Durham, NC 27708, United States Phone: 919-660-7753 / Email: [email protected]Itay Goldstein Wharton School, 3620 Locust Walk, Philadelphia, PA 19104, United States Phone: 215-746-0499 / Email: [email protected]Zeqiong Huang Yale University, 165 Whitney Avenue, New Haven, CT 06511, United States Phone: 203-436-9426 / Email: [email protected]Rahul Vashishtha Duke University, 100 Fuqua Drive, Durham, NC 27708, United States Phone: 919-660-7755 / Email: [email protected]First Draft: May 2019 This Draft: July 2019 Preliminary and do not cite or circulate without permission Abstract: We find a significantly positive relation between bank liquidity mismatch and the sensitivity of deposit flows to bank performance. The result is driven by uninsured deposits, when banks experience poor performance, and for small and medium sized banks. Banks with more liquidity mismatch are more prone to failure, experience more deposit withdraws, and lending reduction during the Financial Crisis of 2008. Our results support the idea that liquidity creation by banks comes at the cost of fragility in banks’ financial structures that seed the potential for banking instability.
49
Embed
Bank Liquidity Mismatch, Strategic Complementarity and ... Flows and Bank [email protected] . First Draft: May 2019 . This Draft: July 2019 . ... Iyer and Puri
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Bank Liquidity Mismatch, Strategic Complementarity and Deposit Flows *
Qi Chen Duke University, 100 Fuqua Drive, Durham, NC 27708, United States
Yale University, 165 Whitney Avenue, New Haven, CT 06511, United States Phone: 203-436-9426 / Email: [email protected]
Rahul Vashishtha
Duke University, 100 Fuqua Drive, Durham, NC 27708, United States Phone: 919-660-7755 / Email: [email protected]
First Draft: May 2019 This Draft: July 2019
Preliminary and do not cite or circulate without permission
Abstract: We find a significantly positive relation between bank liquidity mismatch and the sensitivity of deposit flows to bank performance. The result is driven by uninsured deposits, when banks experience poor performance, and for small and medium sized banks. Banks with more liquidity mismatch are more prone to failure, experience more deposit withdraws, and lending reduction during the Financial Crisis of 2008. Our results support the idea that liquidity creation by banks comes at the cost of fragility in banks’ financial structures that seed the potential for banking instability.
One of the key questions in banking literature is whether or to what extent banks runs are
driven by poor fundamental or panic. Banks perform a critical role in the economy by transforming
liquidity. On one hand banks provide liquidity to borrowers by funding long-term, illiquid loans,
and on the other hand they take demand deposits and thus provide liquidity to depositors. While
providing liquidity to borrowers and depositors, banks create liquidity mismatch on their own
balance sheet and make them inherently fragile institutions.1 It creates strategic complementarities
in depositors’ payoff: when a depositor withdraws money, he/she leaves fewer liquid resources
remaining to pay the remaining depositors. Remaining depositors’ payoff would be further reduced
if the bank has to liquidate illiquid assets at a discount. Therefore, depositors wish to withdraw
money more when they expect others to do so. As illustrated in Diamond and Dybvig (1983), this
creates the possibility of pure panic based bank failures in which a large mass of depositors
withdraws money purely because they expect others to do so even though the bank has enough
assets to eventually pay everybody in the long run. Such panic runs are distinct from fundamental
runs (modelled in Chari and Jagannathan, 1988; Jacklin and Bhattacharya, 1988; Allen and Gale,
1998) in which depositors run on an insolvent bank lacking sufficient assets to pay all depositors.
In this study, we empirically examine whether and to what extent panic characterizes
depositor behavior. Separating between panic- and fundamental-based behaviors is critical from a
policy perspective. Several (and arguably quite costly) government initiatives – deposit insurance,
lender of last resort, suspension of convertibility – are predicated on the idea that panic
characterizes many of the bank runs. Given the importance, it is not surprising that we are not the
1 Hanson, Shleifer, Stein, and Vishny (2015) find that over the period 1896 to 2012, deposits have financed 80% of banks’ assets on average with a standard deviation of just 8%.
3
first to attempt to tackle this issue. Most prior works address this issue by examining the link
between financial crises and fundamentals.2 These works largely found that crises episodes are
correlated with poor fundamentals and have interpreted these findings as supportive of
fundamental-based runs and against panic-based runs. Goldstein and Pauzner (2005), however,
illustrate that this is a flawed conclusion. Using the global games approach, they link various run
equilibriums to the level of fundamentals. They show that even if crises can be linked to
fundamentals, it can still be the case that they would not have occurred without panic driven
coordination failure amongst depositors. This could occur if low fundamentals trigger depositor
panic, creating an indirect correlation between crises and fundamentals. In this scenario, panic
amplifies the response to fundamentals and generates crises in a level of fundamentals that (absent
coordination problems) could have supported a non-crisis outcome. Therefore, while the prior
empirical works highlight the importance of fundamentals in crises, they speak little about the role
of panic.
Motivated by the insights from Goldstein and Pauzner (2005), we empirically explore the
presence of panic in depositor behavior. Specifically, we examine whether depositors respond more
strongly to fundamentals when strategic complementarities are expected to be stronger. Such
evidence would be indicative of the presence of panic as it implies that depositors react more
strongly to the same level of fundamental news because of their beliefs about actions of other
depositors. This approach has also been used in prior studies to explore the importance of strategic
complementarities in non-banking institutions (Chen, Goldstein, and Jiang, 2010; Schmidt,
Timmerman, and Wermers, 2016; Goldstein, Jiang, and Ng, 2017).
2 Examples of studies include Gorton (1988), Demirguc-Kunt and Detragiahe (1998, 2002), Schumacher (2000), Martinez-Peria and Schmukler (2001), and Calomiris and Mason (2003).
4
Our main identification strategy is that depositors of banks that engage in more liquidity
transformation are expected to exhibit greater strategic complementarities and thus more incentive
for panic based running. A key element of our research design is to measure the extent of liquidity
mismatch between banks’ assets and liabilities that creates strategic complementarities. We use the
measure of liquidity transformation developed by Berger and Bouwman (2009) that captures the
extent to which banks employ short-term, liquid funding sources to invest in illiquid, long-term
assets. Their measure CatFat is a summary measure of mismatch in a bank’s liquidity, but not risk
or maturity. We provide a more detailed discussion of the measure in Section 2. The measure
exhibits significant cross-sectional variation across banks, which we exploit to examine whether
depositor response to fundamental shocks varies with the extent of liquidity transformation.
Using a large sample of U.S. commercial banks over the period 1993-2014, we uncover
large differences in depositor sensitivity to fundamental news across banks with different levels of
liquidity transformation. Compared to a bank at the 25th percentile of liquidity transformation, the
uninsured deposit flows of a bank at the 75th percentile are 40% more sensitive to a unit change in
return on equity (ROE). This result is consistent with strategic complementarities significantly
amplifying the response to fundamentals.
A key concern is that this result might reflect differences in persistence of earnings shock
instead of panic. If earnings shocks at banks with greater liquidity transformation are more
persistent, then the depositor response at these banks could be stronger because of the greater
magnitude of the fundamental shock and not because of the amplification caused by strategic
complementarities. Two analyses address this concern. First, we measure the persistence of
earnings shocks at different banks and find that controlling for it makes little difference to our
results. Second, we document that uninsured deposit flows exhibit a concave relation with ROE,
5
consistent with theory prediction that strategic complementarity is stronger (depositor’s incentive to
run is higher) when banks’ fundamental is weaker. In fact, our results are almost entirely driven by
below median ROE and we find no differences in depositor responses for above median shocks.
More importantly, such concavity is stronger when the bank engages in more liquidity
transformation, consistent with the channel works through strategic complementarity. Collectively,
these results suggest that there is a significant element of panic in the behavior of depositors not
protected by deposit insurance.
We also explore the behavior of insured depositors. Recent evidence suggests that banks
attempt to deal with the fragility of their uninsured depositor base by actively attracting insured
depositors in times of poor performance (Martin, Puri, and Ufier, 2018; Chen, Goldstein, Huang,
and Vashishtha, 2019). Consistent with these prior results, we find that insured deposit flows are
less sensitive to fundamental shocks at high liquidity transformation banks. The strategy appears
effective, as we do not find any differences in the sensitivity of total deposit flows (both uninsured
and insured) across banks with different levels of liquidity transformation. These findings not only
yield insights into how banks actively manage the fragility of their depositor base but also highlight
the efficacy of deposit insurance in mitigating panic.
Another concern is that the result might reflect banks’ risk; that is, banks with more liquidity
mismatch take on more risk and as a result, depositors are more sensitive to its performance and
respond more strongly. While risk transformation and liquidity transformation are related in some
cases, they do not move perfectly in tandem. 3 For a given level of risk, the amount of liquidity
transformation might vary considerably. We conduct two analyses to address this concern. First, we
3 For example, a bank can securitize loans into asset backed securities and improve its liquidity.
6
partition the sample into four groups based on liquidity creation on asset side, and within each
group test whether more liquidity creation on liability side makes uninsured deposits more sensitive
to banks’ performance. The idea is that holding the asset side constant, the result cannot be driven
by asset risk and thus is attributable to strategic complementarity from bank’s liquidity creation
from its liability side. We find that more liquid liability leads to a more sensitive uninsured deposit
base as long as the asset side is not too liquid, and the sensitivity is higher when the asset side is
more illiquid. Second, we also partition observations into four quartiles based on liability liquidity
creation. We find that uninsured deposits are more sensitive to performance when the banks’ assets
are less liquid, and even more so when the liability side is more liquid. These results lend support to
that the results are driven by strategic complementarity but not just asset side risk.
In our final set of analyses, we explore the implications of our above findings on the
presence of panic in depositor behavior on three aggregate banking outcomes: failure rates,
performance during crises, and overall profitability. We find that liquidity mismatch has strong
predictive ability for future bank failure. An interquartile increase in bank’s liquidity mismatch is
associated with a 6% increase in failure chance over the next 3 years. We use the financial crisis of
2008 as a laboratory to observe the performance and response of banks with different levels of
liquidity mismatch during a crisis episode. We find that during the crisis, banks with more liquidity
mismatch exhibit a greater erosion in their deposit base (despite offering higher rates), lower
growth in credit, and higher failure rates. Finally, we analyze profitability and find that greater
liquidity transformation is associated with higher profitability. Collectively, these results enhance
our understanding of the banking business model and highlight that liquidity transformation allows
banks to earn higher profits at the cost of enhanced fragility and failure risk.
7
Our paper relates to prior studies that explore the role of fundamentals and panic in bank
runs. Based on a thorough review of the earlier work, Goldstein (2013) concludes that while prior
studies document that runs are correlated with fundamentals, they speak little to the presence and
importance of panic. Among the more recent studies on this topic, Iyer and Puri (2012, 2016) and
Egan et al. (2017) are the most related. Iyer and Puri (2012, 2016) explore depositor responses in
their case study of one bank run in India that arguably was triggered by panic. Egan et al. (2017)
study a sample of the 16 largest US retail banks and find that uninsured deposit elasticity to banks
distress is sufficiently high to make banks very fragile. Unlike our study, however, they do not
separately identify the fragility that results from panic. Furthermore, to the best of our knowledge,
we are the first to provide large sample evidence of the presence of panic based running in
depositors in the commercial banking industry.
Our study is also related to papers that document presence of panic in other settings such as
where Δ𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝑈𝑈 is the deposit flows measured as the changes in bank i’s deposit balances over period
t scaled by the beginning of period assets; 𝑃𝑃𝐷𝐷𝑃𝑃𝑓𝑓𝑖𝑖,𝑖𝑖−1 is a measure of bank performance that
depositors observe at the end of quarter t-1.
Following Chen et al (2019), we measure the deposit flows over the two quarters following
the end of quarter t-1 for which bank performance is measured. This is because banks typically file
call reports with a delay of 30 days after the calendar quarter ending (Baderscher et al., 2017) and
because the literature on post-earnings announcement drift suggests that investors respond to
quarterly accounting reports with a delay of up to a quarter following the announcement (Bernard
and Thomas, 1989). We measure deposit flows as the change in deposits over the subsequent two
5See Chevalier and Ellison (1997), Chen, Goldstein, and Jiang (2010) and Goldstein, Jiang and Ng (2017).
11
quarters scaled by the beginning of period assets. We also follow Chen et al. (2019) and use return
on equity (ROE) as the primary measure of bank performance.6
The control variables include both bank and quarter fixed effects. In addition, we also include
time varying controls for bank characteristics that are shown to affect deposit flows in prior works
(e.g., Acharya and Mora, 2015). These control variables include (i) capital ratio defined as book value
of capital scaled by total assets (Capital Ratio), (ii) wholesale funding scaled by total assets
(Wholesale Funding), (iii) real estate loan share calculated as the amount of loans secured by real
estate divided by total loans (RealEstate_Loans), and (iv) the logarithm of asset size (Ln(Assets)), all
measured at the end of the quarter t-1. Finally, we control for lagged deposit rate which would also
be expected to affect the deposit flows (Deposit Rate). Ideally, we would like to control for rates
offered on uninsured and insured deposits when modelling these two categories of deposit flows.
However, Call reports do not separately report the interest expenses on insured and uninsured deposits.
We use the core deposit rate to proxy the rates offered on insured deposits and the rate on large time
deposit to proxy the rates on uninsured deposits. We believe this is a reasonable approximation
because core (large time) deposits are most likely to be insured (uninsured).7 We measure these rates
as the quarterly interest expense on the deposits divided by the average quarterly deposits over the
same period.
Figure 3 presents the semi-parametric plot of uninsured deposit flows and lagged ROE for
the full sample. It shows that uninsured deposit flow is positively related to past performance, and
more so when the performance is lower. The concavity indicates the withdrawal pressure from
6 Chen et al (2019) provide a detailed discussion on why ROE is a preferred measure of bank performance than the alternative measures such as ROA or non-performing loans. 7 Until March 31, 2011, core deposits were defined in the Uniform Bank Performance Report (UBPR) User Guide as the sum of demand deposits, all NOW and automatic transfer service (ATS) accounts, money market deposit accounts (MMDAs), other savings deposits, and time deposits under $100,000. As of March 31, 2011, the definition was revised to reflect the permanent increase to FDIC deposit insurance coverage from $100,000 to $250,000 and to exclude insured brokered deposits from core deposits.
12
depositors when bank performance is poor is higher than deposit inflows when bank performance is
good. Goldstein et al. (2017) find similar concave relation between fund flows and fund past
performance in open-end corporate bond funds. A casual inspection of Figure 1 suggests that the
concavity starts when the ROE is somewhere in the middle of 10-20%, higher than the average
level in our sample.
We formally test the non-linearity in the flow-performance sensitivity by estimating
where Δ𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝑈𝑈 is the uninsured deposit flows for bank i during period t, and 𝐶𝐶𝑀𝑀𝐶𝐶𝑓𝑓𝑀𝑀𝐶𝐶𝑖𝑖𝑖𝑖−1 is the
measure of banks’ liquidity mismatch.
To start with, to assess the average pattern, we first estimate a reduced form of Eqn. (2)
without including the interaction terms with D(ROE<Median). The result is presented in Column
(1) of Table 3. It shows that liquidity mismatch significantly increases the sensitivity of uninsured
deposit flows to bank performance. The coefficient estimate for ROE is 0.035 (t-stat = 4.932) and
that for the interaction term between ROE and CatFat is 0.103 (t-state = 5.327). These estimates
suggest that an interquartile increase in liquidity mismatch (of 0.22) would increase the sensitivity
of uninsured deposit flows to bank performance by 40%.8 In terms of deposit stability (volatility),
the estimates imply that for the same changes in the fundamental ROE, a bank at 75 percentile level
8This is estimated as (sensitivity at 75th percentile – sensitivity at 25th percentile) / (sensitivity at 25th percentile) = 0.103*(0.43-0.21)/ (0.03335+0.103*0.21)= 40%.
14
of liquidity mismatch would experience 96% (=1.4*1.4-1) higher volatility in its uninsured deposit
flows than a bank with 25 percentile level of liquidity mismatch.
Column (2) presents the results from estimating the full version of Eqn. (2). It shows that the
effect of liquidity mismatch is primarily driven when banks experience below median performance.
Specifically, the estimated coefficient for ROE*CatFat is -0.024 and insignificant from zero at
conventional levels, indicating that liquidity mismatch does not affect flow-performance sensitivity
for banks with above median performance. However, the coefficient estimate for
ROE*CatFat*D(ROE<Median) is 0.130 and significant at less than 5% (t-stat of 2.36), indicating
that for banks with below median ROE, liquidity mismatch significantly increases the sensitivity of
uninsured deposit flows to bank performance.
In Columns (3) to (4), we repeat the above analyses with insured deposit flows as the
dependent variable. Unlike uninsured deposits, insured deposits are fully protected in case of bank
failure, and depositors’ payoff is not affected by other depositors’ run behavior. This implies that
the strategic complementarity is not as important a concern for insured deposits, and we should not
expect to observe a positive relation between liquidity mismatch and the insured flows sensitivity to
performance. In addition, prior literature has shown that banks tend to attract insured deposits to
compensate for the outflows in uninsured deposits (Manju et al. 2018). When this is the case, we
may observe a negative coefficient for the interaction term between CatFat and ROE for insured
deposits.
Estimates in Column (3) show that the coefficient estimate for ROE is 0.094 (t-stat =
12.394). This finding is consistent with findings in prior literature that insured deposit flows are
sensitive to bank performance (e.g., Peria and Schumkler, 2001; Berger and Turk-Ariss, 2015).
However, the interactive term between ROE and CatFat is negative at -0.079 (t-state = -4.024),
15
indicating that banks liquidity mismatch is negatively related to the sensitivity of insured deposit
flows to bank performance, suggesting that more mismatched banks manage their insured deposits
to offset the loss of uninsured deposits.
Column (4) shows that the relationship between insured deposit flows and bank
performance is concave (as shown in Table 2), but the degree of concavity does not differ
significantly by banks’ liquidity mismatch. Specifically, the coefficient estimate for the
ROE*D(ROE<median)*CatFat is -0.066 and not statistically significant from zero at conventional
levels. These results are consistent with the interpretation that the effect of liquidity mismatch on
the sensitivity of uninsured depositor flows to bank performance is driven by the strategic
complementarities in the payoffs of uninsured depositors, and not by other uncontrolled bank
characteristics that affect both insured and uninsured deposits similarly.
Columns (5) and (6) use total deposit flows as the dependent variable to assess the net effect
of liquidity mismatch on the stability of banks’ deposit funding. The results shown there indicate
that the mediating effect of insured deposit flows largely offset the flows from uninsured deposits.
As a result, the liquidity mismatch does not have a significant effect on the sensitivity of total
deposits to bank performance. Additional analysis: controlling for persistence
One may be concerned that the difference in flow-performance sensitivity we find reflect
differences in the performance persistence across banks. If banks with higher liquidity mismatch
also have more persistent performance, then depositors will respond more strongly to each unit of
performance. To address this concern, we directly control for the effect of banks’ performance
persistence on flow-performance sensitivity by adding the interactive between a measure of bank
performance persistence and ROE. Specifically, we measure banks’ performance persistence using
16
the coefficient 𝛽𝛽1 estimated from the following AR(1) regression, estimated for each bank-quarter
using the bank’s observations over the previous 12 quarters:
𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖 = 𝛼𝛼0 + 𝛽𝛽1𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖−1 + 𝜀𝜀𝑖𝑖 (3)
Table 3, Panel B presents the results after controlling for persistence. They show that the
coefficients on our main variable of interest, 𝐶𝐶𝑀𝑀𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶𝑖𝑖−1 × 𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖−1, remain qualitatively positive
and statistically significant for uninsured deposits.9 In untabulated anaylsis, we also include
interactions of control variables and 𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖−1, and the results are robust. Thus our results are not
driven by potential correlation with these variables.
3.3 The impact of liquidity mismatch during financial crisis and in normal times
We have shown that more liquidity mismatch is associated with a larger deposit flow
performance sensitivity. However, it is not clear whether the impact is different during financial crisis
and normal times. In this section, we first split the sample into two: crisis period and non-crisis period
and examine if the results are similar in the two subsamples. Following Archarya and Mora (2016),
we look at the recent Global Financial Crisis and define it as from 2007Q3 to 2009Q2.
Table 3, Panel C shows how deposit flow performance sensitivity varies in and out of the
Financial Crisis. Column (1) examines the uninsured deposit flow during the Crisis. The coefficient
on ROE it-1 is 0.030, on CatFatit-1 × ROE it-1 is -0.055, and neither is statistically significant. However,
the coefficient on CatFatit-1 is -5.635, and is significant at 5% level. This suggests that during the
crisis, uninsured deposits were indiscriminate about bank’s recent performance; rather, given that the
banking system is in a crisis, they were concerned more about the systematic risk and withdrew more
9 The number of observations in Table 5 is lower than those in Table 3. This is because we estimate Eqn. (5) only for banks that did not experience more than 10% increase in annual assets during the previous 12 quarters to minimize the impact of bank mergers on performance persistence. In untabulated sensitivity analysis, we re-estimate our main specifications using the same sample as in Table 5 and find little change in our results quantitatively.
17
from banks with more liquidity mismatch. Column (4) shows that our main results are robust in non-
crisis period.
3.4 Liquidity creation on liability side and uninsured deposit sensitivity to performance
Another concern is that the result might reflect banks’ risk; that is, banks with more liquidity
mismatch take on more risk and as a result, depositors are more sensitive to its performance and
respond more strongly. While risk transformation and liquidity transformation are related in some
cases, they do not move perfectly in tandem. For a given level of risk, the amount of liquidity
transformation might vary considerably. We conduct two analyses to address this concern. First, we
focus on the liability side and examine whether more liquid liability makes uninsured deposits more
sensitive. This is to hold asset liquidity constant and focus on the strategic complementarity among
depositors, and the results would not be attributable to pure risk of the asset portfolio.
We partition the sample into four based on the level of asset side liquidity creation, and for
where 𝑌𝑌𝑖𝑖𝑖𝑖 is the outcome variables for bank i at time t. We examine three categories of outcomes:
deposit flows, deposit rates, and loan decisions. 𝐶𝐶𝑃𝑃𝑀𝑀𝐶𝐶𝑀𝑀𝐶𝐶𝑖𝑖 is an indicator variable for the financial
crisis period of 2007Q3 to 2009Q2.
Column (1) examines the behavior of uninsured deposit flows. We find that a significant
positive coefficient estimate for CatFatit-1 and a negative coefficient estimate for the interaction
term between CatFatit-1 and Crisis, both are highly statistically significant. This suggests that banks
with more liquidity mismatch had a larger growth in uninsured deposit during non-crisis period, but
during the crisis, they experience a larger decrease in uninsured deposits. Column (2) uses the
insured deposit flows as the dependent variable and shows that a positive coefficient on CatFatit-1 ×
24
Crisis at 3.414 (t-stat = 6.577), suggesting some degree of substitution between insured and
uninsured deposits. Estimates in Column (3), which examines total deposit flows, show that during
the crisis, banks with more liquidity mismatch still lose more deposits, as evidenced by the negative
coefficient on CatFatit-1 × Crisis of -2.550 (t-stat=-4.045).
Columns (4) to (7) examine how deposit rates vary with crisis. We study the rates on four
types of deposits: transaction, saving, small time deposits and large time deposits. The deposits in
the first three categories are more likely to be within FDIC insurance limit while the large time
deposits are more likely to be uninsured. For the average bank in our sample, 25% of its deposits
are in the form of transaction deposits, 30% in savings deposits that pays the average rate of 1.81%,
and 28% in small time deposit and 17% in large time deposits, both paying on average 3.6% deposit
rate. As expected, the deposit rates on transaction and saving deposits are much lower than those on
time deposits.
In constructing the CatFat measure, Berger and Bouwman (2009) classify both transaction
and saving deposits as liquid liabilities because they can be withdrawn without any penalty. All
time deposits regardless of maturity are classified semi-liquid liabilities as they can be withdrawn
with some penalty. A bank that finances long-term illiquid loan with more liquid liabilities creates
more liquidity for the economy and is more liquidity mismatched. Thus, we expect CatFat to be
positively correlated with the percentage of deposits in transactions and savings deposit and
negatively correlated with time deposits. This is indeed the case. The correlation coefficients with
both small and large time deposits are negative, at -0.32 and -0.02, respectively. Interestingly, while
CatFat is positively correlate with the sum of transaction and savings deposit (at 0.28), it is driven
by the positive correlation with the savings deposit. The correlation coefficient between CatFat and
25
savings deposit is 0.34 and that between CatFat and transaction deposit is -0.11. Therefore, it
appears that savings deposits play a significant role in banks’ ability to create liquidity.
Column (4) to (7) show that during the non-crisis period, there is no significant relation
between liquidity mismatch and deposit rates except the rates on savings deposit. However, the
coefficient estimates for the interaction term between CatFat and Crisis are all significantly positive
for all rates, indicating that during the crisis period, banks with more liquidity mismatch had to
offer higher rates to attract deposits.
We next examine how liquidity mismatch affects banks’ credit decisions during the crisis
period. Columns (8) shows the result from using growth in banks’ loan balances as the dependent
variable. It documents a negative coefficient for CatFat*Crisis, suggesting a negative effect of crisis
on banks’ loan growth for more liquidity mismatch banks. The effect however is not statistically
significant at conventional levels. Columns (9) and (10) use growth in in banks’ commitment and in
total credit (which is the sum of growth in loan and commitment) as the dependent variables. In
both columns, the coefficients for CatFat*Crisis are significantly negative, indicating that banks
subject to liquidity mismatch risk reduce their credit growth more during the crisis period. The
economic magnitude of the effect is significant. The average annualized growth in commitment is
0.96%, whereas a 0.22 increase in CatFat (corresponding to the change in CatFat from 25th
percentile to 75th percentile) would lower growth in commitment during the crisis period by 1.3%.
In untabulated analysis, we further break the crisis period into two subperiods: crisis 1,
defined as the period between 2007Q3 and 2008Q2 and crisis 2, defined as the period between
2008Q3 and 2009Q2. We continue to find significant negative coefficients for both CatFat*Crisis 1
and CatFat*Crisis 2 when the dependent variable is either growth in commitment and growth in
total credit. However, when we examine growth in loan, we find a significantly positive coefficient
26
for CatFat*Crisis 1 but a negative coefficient for CatFat*Crisis 2. These results suggest that
borrowers draw down their commitment from banks during the first stage of the crisis, consistent
with the findings from Ivashina and Scharfstein (2010) and Acharya and Mora (2015).
5. Additional analyses
5.1 Alternative liquidity mismatch measures
In this section, we examine whether our results are robust to two alternative measures of
liquidity mismatch. The first is CATNONFAT, which differs from CatFat is that it does not consider
the impact from banks’ off-balance sheet commitment. The second alternative measure is LMIRisk,
constructed based on the Liquidity Mismatch Index from Bai et al. (2018). 11 We follow Bai et al.
(2018) and its online appendix and construct LMI for a sample of commercial banks. Specifically,
𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖 for bank i at time t is computed as the net of the asset and liability liquidities: 𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖 =
∑ 𝜆𝜆𝑖𝑖,𝑎𝑎𝑘𝑘 𝑀𝑀𝑖𝑖,𝑘𝑘𝑖𝑖
𝑘𝑘 + ∑ 𝜆𝜆𝑖𝑖,𝑙𝑙𝑘𝑘′ 𝑙𝑙𝑖𝑖,𝑘𝑘′𝑖𝑖
𝑘𝑘 .where 𝜆𝜆𝑖𝑖,𝑎𝑎𝑘𝑘 is time varying asset liquidity factor that was backed out
from haircuts in repo market ; 𝜆𝜆𝑖𝑖,𝑙𝑙𝑘𝑘′ is the liability liquidity factor that calculated recursively using
the maturity and liquidity cost. Thus, LMI measures the mismatch between the market liquidity of a
bank’s assets and the funding liquidity of liabilities. LMI_Risk is calculated as max (𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖 −
𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖1𝜎𝜎, 0). It measures the exposure of a bank to a 1σ unfavorable change in both market and funding
liquidity conditions. It is based on a stress scenario under which all claimants on the firm are assumed
to act under the terms of their contract to extract the maximum liquidity possible, and the firm reacts
by maximizing the liquidity it can raise from its assets. LMI measure differs from Berger and
11 Bai et al. (2017)’s LMI index is constructed for bank holding companies. We take their parameters and calculate the LMI index at commercial bank level, using data from call report. A priori, it is not clear whether depositors make withdrawal decisions based on the health of the top bank holding company or of the subsidiary commercial bank alone. We use commercial bank level specification because the insured deposits data are not available from Y9-C reports filed by bank holding companies.
27
Bouwman (2009) in the weight assigned is dynamic and incorporate marked liquidity conditions,
while Berger and Bouwman (2009) use fixed weight.
Table 1, Panel B shows that LIM and LMI_Risk are highly correlated with CatFat, indicating
that they all reflect the degree of liquidity mismatch in a given bank. However, the LMI measure is
designed to capture market wide liquidity mismatch, and therefore its variation is primarily driven by
over time changes in market condition and less by cross-bank differences. This can be seen from
Figure 3 in the Appendix where we plot the summary statistics for LMI risk over time. Further
analyses (untabulated) show that quarterly fixed effects alone would explain 92% of the variations in
LMI where bank fixed effects would explain only 7%.
Table 9 Panel A show that our results are robust to using these two alternative measures.
Column 1 shows that the sensitivity of uninsured deposit flows to performance is significantly higher
in banks with more liquidity mismatch as measured by CatNonFat. The coefficient estimate for ROE*
CatNonFat is 0.198 (t-stat = 8.98). Column 2 shows that CatNonFat does not appear to affect the
sensitivity of insured deposit flows to bank performance. Column 4-6 use LMI_Risk as the liquidity
mismatch measure and find similar results.
5.2 Alternative performance measures
In our final set of robustness tests presented in Panel B of Table 9, we explore the sensitivity of
our results to two alternative performance measures: return on assets (ROA) and non-performing
loans (NPL). The results using these measures are qualitatively similar to those using ROE.
Specifically, Panel A shows that the sensitivity of uninsured deposit to ROA is increasing in liquidity
28
mismatch measured as CatFat. Panel B shows uninsured deposit flows is negatively associated with
banks’ non-performing loans, and more so for banks with more liquidity mismatch.
6. Conclusion
In this paper we examine the relation between liquidity mismatch on bank’s balance sheet
and depositors’ response to bank performance. We find that the sensitivity of deposit flows to bank
performance is much higher for banks with more liquidity mismatch on their balance sheet. The
result is driven by uninsured deposits, when banks experience poor performance, and for small and
medium sized banks. We also find a positive association between bank’s liquidity mismatch and
their future failure. Banks with more liquidity mismatch at the beginning of the financial crisis are
much more likely to fail during the financial crisis of 2008, regardless of their profitability. Banks
with more liquidity mismatch also experience more deposit withdraws and lending reduction during
the Financial Crisis of 2008. Our results are consistent with the theoretical prediction of Goldstein
and Pauzner (2005) where liquidity mismatch can generate strategic complementarities in
depositors’ payoff and induce panic-based runs due to coordination failure among depositors. While
they emphasize the importance of fundamental in bank runs, they also support the idea that liquidity
creation by banks comes at the cost of fragility in banks’ financial structures that seed the potential
for banking instability.
29
References
Acharya, V.V. and Mora, N., 2015. A crisis of banks as liquidity providers. Journal of Finance, 70(1), pp.1-43.
Badertscher, B.A., Burks, J.J. and Easton, P.D., 2018. The market reaction to bank regulatory reports.
Review of Accounting Studies, 23(2), pp.686-731. Bai, J. Krishnamurthy and Weymuller, C. 2018. Measuring liquidity mismatch in the banking sector.
Journal of Finance Benston, G.J. and Kaufman, G.G., 1997. FDICIA after five years. Journal of Economic
Perspectives, 11(3), pp.139-158. Berger, A.N. and Bouwman, C.H., 2009. Bank liquidity creation. Review of Financial Studies, 22(9),
pp.3779-3837. Berger, A.N. and Turk-Ariss, R., 2015. Do depositors discipline banks and did government actions
during the recent crisis reduce this discipline? An international perspective. Journal of Financial Services Research, 48(2), pp.103-126.
Bernard, V.L. and Thomas, J.K., 1989. Post-earnings-announcement drift: delayed price response or
risk premium? Journal of Accounting Research, pp.1-36. Brown, M., Guin, B. and Morkoetter, S., 2017. Deposit withdrawals from distressed commercial
banks. Unpublished working paper. University of St. Gallen. Calomiris, C.W. and Kahn, C.M., 1991. The role of demandable debt in structuring optimal banking
arrangements. American Economic Review, pp.497-513. Calomiris, C.W. and Mason, J.R., 1997. Contagion and bank failures during the Great Depression: The
June 1932 Chicago banking panic. American Economic Review, 87(5), pp.863-883. Calomiris, C.W. and Mason, J.R., 2003. Fundamentals, panics, and bank distress during the
depression. American Economic Review, 93, pp.1615-1647. Chevalier J. and Ellison E. 1997. Risk taking by mutual funds as a response to incentives. Journal of
Political Economy 105(6), pp. 1167-1200. Chen, Q., Goldstein, I. and Jiang, W., 2007. Price informativeness and investment sensitivity to stock
price. Review of Financial Studies, 20(3), pp.619-650. Chen, Q., Goldstein, I. Huang, Z.Q., and Vashishtha, R., 2019. Bank transparency and deposit
flows. Working paper, Duke, Wharton and Yale. Davenport, A.M. and McDill, K.M., 2006. The depositor behind the discipline: A micro-level case
study of Hamilton Bank. Journal of Financial Services Research, 30(1), pp.93-109.
30
Demirguc-Kunt, A. and Detragiache, E. 1998. The determinants of banking crisis: evidence from developed and developing countries. IMF Staff Papers 45, 81-109.
Demirguc-Kunt, A. and Detragiache, E. 1998. Does deposit insurance increase banking system
stability? An empirical investigation. Journal of Monetary Economics 49. Pp. 1373-1406. Diamond, D.W., 1984. Financial intermediation and delegated monitoring. Review of Economic
Studies, 51(3), pp.393-414. Diamond, D.W. and Dybvig, P.H., 1983. Bank runs, deposit insurance, and liquidity. Journal of
Political Economy, 91(3), pp.401-419. Diamond, D.W. and Rajan, R.G., 2001. Liquidity risk, liquidity creation, and financial fragility: A
theory of banking. Journal of Political Economy, 109(2), pp.287-327. Egan, M., Hortaçsu, A. and Matvos, G., 2017. Deposit competition and financial fragility: Evidence
from the U.S. banking sector. American Economic Review, 107(1), pp.169-216. Egan, M., Lewellen, S. and Sunderam, A., 2017. The cross section of bank value (No. w23291).
National Bureau of Economic Research. Federal Financial Institutions Examination Council, 2000. Uniform retail credit classification and
account measurement policy. Gatev, E. and Strahan, P.E., 2006. Banks' advantage in hedging liquidity risk: Theory and evidence
from the commercial paper market. Journal of Finance, 61(2), pp.867-892. Goldberg, L.G. and Hudgins, S.C., 1996. Response of uninsured depositors to impending S&L failures:
evidence of depositor discipline. Quarterly Review of Economics and Finance, 36(3), pp.311-325. Goldstein, I. and Pauzner, A., 2005. Demand–deposit contracts and the probability of bank
runs. Journal of Finance, 60(3), pp.1293-1327. Goldstein, I., 2013. Empirical literature on financial crisis: fundamentals vs. panic. pp.523-534. Goldstein, I., Jiang. H and Ng. D. 2017. Investor flows and fragility in corporate bond funds. Journal of
Financial Economics, 126, pp. 592-613. Gorton, G., 1988. Banking panics and business cycles. Oxford economic papers, 40(4), pp.751-781. Gorton, G. and Pennacchi, G., 1990. Financial intermediaries and liquidity creation. Journal of
Finance, 45(1), pp.49-71. Gorton, G. and Winton, A., 2003. Financial intermediation. Handbook of the Economics and Finance,
edited by G. Constantinides, M. Harris and R. Stulz, pp.431-552. Hanson, S.G., Shleifer, A., Stein, J.C. and Vishny, R.W., 2015. Banks as patient fixed-income
investors. Journal of Financial Economics, 117(3), pp.449-469.
31
Ivashina V. and Scharfstein, D. 2010. Bank lending during the financial crisis of 2008. Journal of Financial Economics 97, pp. 319-338.
Iyer, R. and Puri, M., 2012. Understanding bank runs: The importance of depositor-bank relationships
and networks. American Economic Review, 102(4), pp.1414-45. Iyer, R., Puri, M. and Ryan, N., 2016. A tale of two runs: Depositor responses to bank solvency
risk. Journal of Finance, 71(6), pp.2687-2726. Kashyap, A.K., Rajan, R. and Stein, J.C., 2002. Banks as liquidity providers: An explanation for the
coexistence of lending and deposit-taking. Journal of Finance, 57(1), pp.33-73. Martin, C., Puri, M. and Ufier, A., 2018. Deposit Inflows and Outflows in Failing Banks: The Role of
Deposit Insurance. NBER Working Paper Martinez Peria, M.S. and Schmukler, S.L., 2001. Do depositors punish banks for bad behavior? Market
discipline, deposit insurance, and banking crises. Journal of Finance, 56(3), pp.1029-1051. Morris, S and Shin. H. 1998. Unique equilibrium in a model of self-fulfilling currency attacks.
American Economic Review 88, pp. 587-597. Sage, A., 2007. Countrywide bank customers fret over deposits, Reuters. Saunders, A. and Wilson, B., 1996. Contagious bank runs: evidence from the 1929–1933
period. Journal of Financial Intermediation, 5(4), pp.409-423. Schmidt, L., Timmerman A. and Wermers, R. 2016. Runs on money market mutual funds. American
Economic Review 106, 2625-2657. Schumacher, L. 2000. Bank runs and currency run in a system without a safety net: Argentina and the
‘tequila’ shock. Journal of Monetary Economics 46, pp. 257-277. Vives, X. 2014. Strategic complementarities, fragility, and regulation. Review of Financial Studies,
27(12), pp. 3547-3592.
32
Figure 1: Over time changes in CatFat
Panel A: Full sample
Panel B: Subsample by bank size
33
Figure 2: Sensitivity of uninsured deposits to bank ROE for the full sample
Panel A: Full sample
Panel B: Subsamples by Catfat
-20
24
Uni
nsur
ed D
epos
it Fl
ow
-10 0 10 20 30ROE
-20
24
Uni
nsur
ed D
epos
it Fl
ow
-10 0 10 20 30ROE
Below Median Catfat Above Median Catfat
34
Table 1. Summary statistics
This table presents summary statistics for the main regression variables. These statistics are calculated over the regression sample. To avoid the impact of mergers and acquisitions, we exclude bank-quarter observations with quarterly asset growth greater than 10%. We also exclude bank quarters with total assets smaller than 100 million. See the Appendix for variable definitions.
Table 2. Sensitivity of deposit flows to bank performance
This table presents OLS estimates of Equation (1). The dependent variable is in Column (1) to (3) is respectively, the change in the uninsured, insured, and total deposits scaled by the beginning value of total assets. D(ROE<Median) is a dummy variable that equals 1 (0) when the bank’s ROEit-1 is below the sample median. All regressions include bank- and quarter-fixed effects. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
Table 3. Liquidity mismatch and sensitivity of deposit flows to bank performance
This table presents OLS estimates of Equations (2) and (3). The dependent variables are changes in the uninsured, insured, and total deposits scaled by the beginning value of total assets. D(ROEit-1<Median) is an indicator variable that equals 1(0) if the ROEit-1 is above sample median. All regressions include bank- and quarter-fixed effects. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
This table presents ordinary least-squares estimates of Equation (2) after controlling for the effect of performance persistence. We measure Persistence as the AR(1) coefficient from a time-series regression of banks’ ROE on its lagged value over the 12-quarters Columns (1) to (3) present the results for uninsured deposit flows, insured deposit flows and total deposit flows, with Persistence and its interaction with ROE included in the regressor. Column The Appendix contains detailed descriptions for the independent variables. All regressions include bank- and quarter-fixed effects. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
Table 4. Decompose Asset side and Liability Side Liquidity
This table presents ordinary least-squares estimates of Equation (3). Columns (1) to (4) present the results for uninsured deposit flows, with control variables and their interaction with ROE included. The Appendix contains detailed descriptions for the independent variables. All regressions include bank- and quarter-fixed effects. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
Panel A: Partition based on liquidity creation on aside side
Table 5. Liquidity mismatch and deposit flows by bank size
This table presents OLS estimates of a modified version of Equation (3). Small, Medium and Large are indicator variables based on bank asset size at the beginning of period t. I(.) is an indicator function that equals 1 (0) if the argument is true (false). All regressions include the control variables, and bank- and quarter-fixed effects. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
Controls included Yes Yes Yes Bank FE Yes Yes Yes Qtr FE Yes Yes Yes Observations 233,942 234,058 233,748 R-squared 0.338 0.364 0.219
43
Table 6. Liquidity mismatch and bank performance
This table explores the effect of liquidity mismatch on bank performance. The dependent variable is return on equity (ROEit+1) for period t+1. Outflowit is a dummy variable that equals 1 if the bank experienced a total deposit outflows equal or larger than 2.5% of its total assets in period t, and 0 otherwise. D(CatFatit>median) is a dummy variable for whether the bank’s CatFatit is above or below sample median. The Appendix contains detailed descriptions for the independent variables. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
This table shows the result of logit regression that explores the association between liquidity mismatch and bank failure. In Columns (1) to (3), the dependent variable is an indicator variable that the bank fails within the next 1, 2, and 3 years during the non-crisis period, respectively. In Column (4), the dependent variable is an indicator variable that the bank failed during the financial crisis period. The Appendix contains detailed descriptions for the independent variables. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
Table 8. Effects of liquidity mismatch during crisis
This table presents ordinary least-squares estimates of Equation (6). The dependent variable is changes in the balance of total loans in Columns (1), the changes in the balance of total commitments in Columns (2), the changes in the sum of loans and commitment in Column (3), and changes in the balances of liquid assets in Columns (4). All dependent variables are scaled by lagged total assets. The Appendix contains detailed descriptions for the independent variables. All regressions include bank- and quarter-fixed effects. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
Dependent Variable
Δ𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝑈𝑈 Δ𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝐼𝐼 Δ𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝑇𝑇𝑇𝑇𝑖𝑖𝑎𝑎𝑙𝑙 Trans
Panel A of this table explores the robustness of our main results to use of two alternative liquidity mismatch measures: CatNonFat in Columns (1) to (3) and LMIRisk in Column (4) to (6). Panel B explores the robustness to two alternative bank performance measures– ROA (return on assets) in Column 1-3 and Non-performing Loan (NPL) in Columns (4)-(6). The Appendix contains detailed descriptions for the independent variables. All regressions include bank- and quarter-fixed effects. T-statistics, reported in parentheses, are based on standard error estimates clustered at the bank level. Statistical significance (two-sided) at the 10%, 5%, and 1% level is denoted by *, **, and ***, respectively.
Panel A: Robustness to alternative liquidity mismatch measures
The preferred measure of Bank liquidity creation per unit of gross total assets, by Berger and Bouwman (2009) and downloaded from https://sites.google.com/a/tamu.edu/bouwman/data. Step 1: Classify all bank activities (asset, liability, and off-balance-sheet) as liquid, semi-liquid, or illiquid based on product category. Step 2: Assign weights to the activities classified in Step 1. Illiquid assets, liquid liabilities get ½, Liquid assets, illiquid liabilities and equities get -1/2. Certain loans and liabilities are classified as semi-liquid and get 0. Step 3: Combine bank activities as classified in Step 1 and as weighted in Step 2 to construct our liquidity creation measure.
CatNonFat The same as CatFat, but does not include off-balance sheet items.
LMI_Risk
𝐿𝐿𝑀𝑀𝐼𝐼_𝑅𝑅𝐼𝐼𝐶𝐶𝐵𝐵𝑖𝑖𝑖𝑖 = max (𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖 − 𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖1𝜎𝜎 , 0). The liquidity risk of a bank is the exposure of that bank to a 1σ unfavorable change in both market and funding liquidity conditions. 𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖 is constructed following Bai et al. (2018) and its online appendix for a sample of commercial banks. Specifically, 𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖 for bank i at time t is computed as the net of the asset and liability liquidities: 𝐿𝐿𝑀𝑀𝐼𝐼𝑖𝑖𝑖𝑖 = ∑ 𝜆𝜆𝑖𝑖,𝑎𝑎𝑘𝑘 𝑀𝑀𝑖𝑖,𝑘𝑘
𝑖𝑖𝑘𝑘 + ∑ 𝜆𝜆𝑖𝑖,𝑙𝑙𝑘𝑘′ 𝑙𝑙𝑖𝑖,𝑘𝑘′
𝑖𝑖𝑘𝑘 .where 𝜆𝜆𝑖𝑖,𝑎𝑎𝑘𝑘 is time varying asset
liquidity factor that was backed out from haircuts in repo market ; 𝜆𝜆𝑖𝑖,𝑙𝑙𝑘𝑘′ is the liability liquidity factor that calculated recursively using the maturity and liquidity cost. The parameters on liquidity factors are from Bai et al. (2018).
Commercial Loan Commercial and industrial loan (RCFD1766). Scaled accordingly. RealEstate_Loans Loans secured by real estate (RCFD1410). Scaled accordingly
ROE I,t-1 Annualized ROE (in %) in quarter t-1, calculated as net income (RIAD4300, adjust year-to-date reporting to within quarter) divided by beginning equity (RCFD3210).
StDev_ROE i,t-1 Standard deviation of ROE measured over 12 rolling quarters (from Quarter 𝐶𝐶 − 12 to 𝐶𝐶 −1).
Capital_Ratio Total equity (RCFD3210) divided by total assets (RCFD2170).
Wholesale_Funding
Wholesale funds are the sum of following: large-time deposits (RCON2604), deposits booked in foreign offices (RCFN2200), subordinated debt and debentures (RCFD3200), gross federal funds purchased and repos [RCFD2800, or (RCONB993+RCFDB995 from 2002q1)], other borrowed money (RCFD3190). Scaled by total assets.
RealEstate_Loan Loans secured by real estate (RCFD1410) scaled by total loans. Ln(Assets) Log of total assets (RCFD2170).
𝛥𝛥𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝐼𝐼
Annualized growth rate in insured deposits as a percentage of lagged assets in quarter 𝐶𝐶 and 𝐶𝐶 + 1. (in %): (𝐼𝐼𝐶𝐶𝐶𝐶𝑂𝑂𝑃𝑃𝐷𝐷𝑀𝑀 𝐷𝐷𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶𝐶𝐶𝑖𝑖,𝑖𝑖+1 − 𝐼𝐼𝐶𝐶𝐶𝐶𝑂𝑂𝑃𝑃𝐷𝐷𝑀𝑀 𝐷𝐷𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶𝐶𝐶𝑖𝑖,𝑖𝑖−1)/𝐴𝐴𝐶𝐶𝐶𝐶𝐷𝐷𝐶𝐶𝑖𝑖,𝑖𝑖−1 ∗ 200%. Insured deposits are accounts of $100,000 or less. After 2006Q2, it includes retirement accounts of $250,000 or less. From 2009Q3, reporting thresholds on non-retirement deposits increased from $100,000 to $250,000. Insured deposits: RCON2702 (before 2006Q2); RCONF049 + RCONF045 (from 2006Q2).
𝛥𝛥𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝑈𝑈
Annualized growth rate in uninsured deposits as a percentage of lagged assets (in %) in quarter 𝐶𝐶 and 𝐶𝐶 + 1. Uninsured deposit is calculated as deposits (RCFD2200) – insured deposits.
𝛥𝛥𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝑇𝑇𝑇𝑇𝑖𝑖𝑎𝑎𝑙𝑙 Sum of Δ𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝐼𝐼 and Δ𝐷𝐷𝐷𝐷𝑝𝑝𝑖𝑖𝑖𝑖𝑈𝑈
Transaction Deposit Ratei,t
Annualized average interest rate (in %) over the two quarters 𝐶𝐶, 𝐶𝐶 + 1 on core deposits: (𝑇𝑇𝑃𝑃𝑀𝑀𝐶𝐶𝐶𝐶𝑀𝑀𝑇𝑇𝐶𝐶𝑀𝑀𝐶𝐶𝐶𝐶 𝑀𝑀𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶 𝑀𝑀𝐶𝐶𝐶𝐶𝐷𝐷𝑃𝑃𝐷𝐷𝐶𝐶𝐶𝐶 𝐷𝐷𝑒𝑒𝑝𝑝𝐷𝐷𝐶𝐶𝐶𝐶𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1)/(𝐴𝐴𝐴𝐴𝐻𝐻.𝑇𝑇𝑃𝑃𝑀𝑀𝐶𝐶𝐶𝐶𝑀𝑀𝑇𝑇𝐶𝐶𝑀𝑀𝐶𝐶𝐶𝐶 𝑀𝑀𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶 𝑏𝑏𝑀𝑀𝑙𝑙𝑀𝑀𝐶𝐶𝑇𝑇𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1) ) ∗ 400%) . Average transaction deposits: RCON3485
Annualized average interest rate (in %) over the two quarters 𝐶𝐶, 𝐶𝐶 + 1 on savings deposits: (𝐶𝐶𝑀𝑀𝐴𝐴𝑀𝑀𝐶𝐶𝐻𝐻 𝑀𝑀𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶 𝑀𝑀𝐶𝐶𝐶𝐶𝐷𝐷𝑃𝑃𝐷𝐷𝐶𝐶𝐶𝐶 𝐷𝐷𝑒𝑒𝑝𝑝𝐷𝐷𝐶𝐶𝐶𝐶𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1)/(𝐴𝐴𝐴𝐴𝐻𝐻. 𝐶𝐶𝑀𝑀𝐴𝐴𝑀𝑀𝐶𝐶𝐻𝐻 𝑀𝑀𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶 𝑏𝑏𝑀𝑀𝑙𝑙𝑀𝑀𝐶𝐶𝑇𝑇𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1) ) ∗ 400%) . Saving: RCONB563 (RCON3486 + RCON3487 before 2001Q1)
Small Time Deposit Ratei,t
Annualized average interest rate (in %) over the two quarters 𝐶𝐶, 𝐶𝐶 + 1 on savings deposits: (𝐶𝐶𝑚𝑚𝑀𝑀𝑙𝑙𝑙𝑙 𝐶𝐶𝑀𝑀𝑚𝑚𝐷𝐷 𝑀𝑀𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶 𝑀𝑀𝐶𝐶𝐶𝐶𝐷𝐷𝑃𝑃𝐷𝐷𝐶𝐶𝐶𝐶 𝐷𝐷𝑒𝑒𝑝𝑝𝐷𝐷𝐶𝐶𝐶𝐶𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1)/(𝐴𝐴𝐴𝐴𝐻𝐻. 𝐶𝐶𝑚𝑚𝑀𝑀𝑙𝑙𝑙𝑙 𝐶𝐶𝑀𝑀𝑚𝑚𝐷𝐷 𝑏𝑏𝑀𝑀𝑙𝑙𝑀𝑀𝐶𝐶𝑇𝑇𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1) ) ∗ 400%) . Average mall time deposits: RCONA529 (RCON3469 before 1997Q1).
Large Time Deposit Ratei,t
Annualized average interest rate (in %) over the two quarters 𝐶𝐶, 𝐶𝐶 + 1 on savings deposits: (𝑙𝑙𝑀𝑀𝑃𝑃𝐻𝐻𝐷𝐷𝐶𝐶 𝐶𝐶𝑀𝑀𝑚𝑚𝐷𝐷𝑀𝑀𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶 𝑀𝑀𝐶𝐶𝐶𝐶𝐷𝐷𝑃𝑃𝐷𝐷𝐶𝐶𝐶𝐶 𝐷𝐷𝑒𝑒𝑝𝑝𝐷𝐷𝐶𝐶𝐶𝐶𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1)/(𝐴𝐴𝐴𝐴𝐻𝐻. 𝑙𝑙𝑀𝑀𝑃𝑃𝐻𝐻𝐷𝐷 𝐶𝐶𝑀𝑀𝑚𝑚𝐷𝐷 𝑀𝑀𝐷𝐷𝑝𝑝𝐶𝐶𝐶𝐶𝑀𝑀𝐶𝐶 𝑏𝑏𝑀𝑀𝑙𝑙𝑀𝑀𝐶𝐶𝑇𝑇𝐷𝐷 𝑀𝑀𝐶𝐶 𝑄𝑄𝐶𝐶𝑃𝑃 𝐶𝐶 𝑀𝑀𝐶𝐶𝑀𝑀 𝐶𝐶 + 1) ) ∗ 400%) .
Core deposit Ratei,t Core deposits include transaction, saving, and small time deposits, and core deposit rate is the average interest rate paid on the three.
Failure in n year An indicator variable that equals 1 if the bank fails in the next n years, measured as 1 if the bank is on the FDIC failed banks list on https://www.fdic.gov/bank/individual/failed/
StDev_ROE i,t-1 Standard deviation of ROE measured over 12 rolling quarters (from Quarter 𝐶𝐶 − 12 to 𝐶𝐶 −1).
NPL i,t-1 The percentage of non-performing loan (RCFD1403+RCFD1407) to total loan.
ROA i,t-1 Annualized ROA (in %) in quarter t-1, calculated as net income (RIAD4300, adjust year-to-date reporting to within quarter) divided by beginning assets.
Persistence it-1
We measure banks’ performance persistence using the coefficient β1 estimated from the following AR(1) regression, estimated for each bank-quarter using the bank’s observations over the previous 12 quarters: 𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖 = 𝛼𝛼0 + β1𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖−1 + 𝜖𝜖