Received: 30 November 2017 Revised: 6 July 2020 Accepted: 23 July 2020

DOI: 10.1111/jbfa.12484

ARTICLE

Bank accounting conservatism and bank loan

quality

Joohyung Ha

School of Management, University of San

Francisco, 2130 Fulton Street, San Francisco, Abstract
California, USA
This study examines the effect of conditional conservatism
Correspondence in loan loss accounting—recognizing bad news in a timely
Joohyung Ha, School of Management, Univer- fashion and deferring the recognition of gains—on a bank’s
sity of San Francisco, 2130 Fulton Street, San
Francisco, California, USA own monitoring effort as reflected in its loan portfolio qual-
Email: jha6@[Link] ity using a sample of publicly traded bank holding companies
in the United States over the period 1994–2015. Sample
banks adopting conservative loan loss accounting practices
have lower nonperforming loans (NPLs) and lower net
charge-offs (NCOs), consistent with conservatism enhanc-
ing monitoring over lending decisions by improving financial
transparency. The paper further finds that the effect of
conservatism on loan quality is more pronounced for banks
with high ex ante information asymmetry or distress risk and
during low and high lending growth cycles, when loan quality
is likely to deteriorate. Collectively, the findings suggest
that conservative reporting reduces banks’ risk-taking in
lending. These findings should be of interest to regulators
and policymakers who debate how to incentivize banks to
use the discretion inherent in their loan loss provisioning in
ways that are more informative and less opportunistic.

KEYWORDS
banks, conditional conservatism, discretion, information asymme-
try, loan loss accounting, loan loss provisioning, loan quality, moni-
toring effort, risk-taking, transparency

J E L C L A S S I F I C AT I O N
G21, M41

J Bus Fin Acc. 2020;1–35. [Link]/journal/jbfa © 2020 John Wiley & Sons Ltd 1
2 HA

1 INTRODUCTION

This study examines the relation between asymmetric timeliness in the recognition of gains and losses (conditional
conservatism) and loan quality for a sample of public banks in the United States.1 While there is some evidence
indicating that timely recognition of expected loan losses serves as a monitoring mechanism (Beatty & Liao, 2011;
Bushman & Williams, 2012, 2015), only a few studies investigate the implications of asymmetric timeliness of losses,
otherwise known as conditional conservatism, for a bank’s risk taking or its own monitoring effort. This study attempts
to fill this gap by taking a more direct approach that focuses on banks’ primary operational decisions, that is, decisions
about making loans. This study builds on the premise that the monitoring benefit of conservatism in reducing banks’
risk taking translates into better loan portfolio quality. The study provides direct evidence that banks’ conditional con-
servatism in loan loss provision is positively associated with their loan portfolio quality, and that the effect of conser-
vatism on loan quality is more pronounced when managers have a higher incentive to shift risk and during economic
cycles when the quality of loans is likely to be of low quality. By establishing a formal connection between conservatism
and loan quality, this research complements and extends the literature on the benefits of bank reporting transparency.
An emerging stream of research views timely recognition of expected loan losses as desirable for curtailing exces-
sive risk taking by banks (Bushman & Williams, 2012, 2015). While GAAP traditionally requires an ‘incurred loss’
methodology for recognizing credit losses that delays recognition until it is probable a loss has been incurred, the
complexity of loan portfolios allows substantial discretion in provisioning (Dugan, 2009).2 Banks’ delayed provision-
ing for expected loan losses can mask the true risk of loan portfolios and thus deteriorate bank transparency. Because
the provisioning is made against expected loan losses and bank capital is set aside to buffer against unexpected loan
losses, less timely provisioning may obscure the true ability of a bank’s capital to provide a buffer against unexpected
losses by forcing the expected loan losses to be absorbed by future bank capital. Consequently, most banking litera-
ture focuses on how banks exercise discretion in loan loss provisioning to understand the implications of timely loan
loss recognition in banks. For example, less timely recognition of loan losses reduces banks’ willingness to lend, which
increases procyclicality during crisis periods (Beatty & Liao, 2011), increases risk shifting and illiquidity in the economy
(Bushman & Williams, 2012), increases equity price crash risks (Andreou, Cooper, Louca, & Philip, 2017), increases
lending corruption (Akins, Dou, & Ng, 2017), exposes banks to systematic downside risk (Bushman & Williams, 2015),
and contributes to a severe financial crisis such as the US savings and loan crisis in the late 1980s and the Japanese
banking crisis in the 1990s (McDonough, Panaretou, & Shakespeare, 2020).
Notably, while most research focuses on the timeliness of recognition of loan losses, less attention is given to the
asymmetric timeliness of loss recognition versus gains recognition or conditional conservatism.3 Asymmetric timeli-
ness is a different aspect of financial reporting than timeliness in that asymmetric timeliness focuses on the relative
timing of recognition of bad news versus good news, while timeliness does not distinguish good news timeliness from
bad news timeliness. Asymmetric timeliness is uniquely designed to provide contracting benefits by counteracting

1 Beaver and Ryan (2005) use the term ‘conditional conservatism’ to refer to asymmetric timely loss versus gain recognition, to distinguish it from uncondi-

tional conservatism, which is the predetermined understatement of the book value of net assets. This paper focuses solely on conditional conservatism and,
for the sake of brevity, often uses ‘conservatism’ instead of ‘conditional conservatism’.
2 InJune 2016, the FASB issued ASU No. 2016-13, Financial Instruments—Credit Losses (Topic 326), which requires banks to immediately record the full
amount of credit losses that are expected in their loan portfolios, providing investors with better information about those losses on a timelier basis. GAAP’s
traditional ‘incurred loss’ methodology has been criticized, as the recognition of loan losses is delayed until it is probable that a loss has been incurred based
on past events and conditions existing at the date of the financial statements, which magnifies the procyclicality of loan making. The new credit loss standard
allows managers to incorporate estimates of potential future events, and therefore affords managers considerably more discretion than does the incurred
loss model.
3 This study focuses on conditional conservatism rather than unconditional conservatism. Unconditional conservatism is news independent in the sense that

it describes the accounting process determined at the inception of assets and liabilities that yields expected unrecorded goodwill, and thus is likely to be a
response to regulatory or tax incentives in order to minimize punitive or reputational damage. Conditional conservatism as described by Basu (1997) results
in earnings that capture difficult-to-verify economic losses more quickly than gains, thus generating a downward bias in the value of net assets. It often arises
in response to contracting demands such as compensation or debt contracts. This study assumes that managers will commit to this level of conservatism to
counteract the potential costs imposed by contracting parties, leading to more efficient loan making decisions. Therefore, conditional conservatism is more
relevant for the setting examined in our research questions.
HA 3

the managerial tendency to quickly disclose good news rather than bad news. The fact that losses must be reported
sooner than gains effectively disciplines managers to engage in optimal investments and triggers early abandonment
of poorly performing projects. Prior studies also find evidence suggesting that asymmetric timely loan loss recogni-
tion facilitates the efficient allocation of capital in the form of improved investment decisions by reducing information
asymmetry and enhancing market disciplining of firms’ risk-taking decisions. While many recent studies support this
prediction (e.g., Balakrishnan, Watts, & Zuo, 2016; García Lara, Garcia Osma, & Penalva, 2016), relatively little atten-
tion has been paid to the benefits of conservatism in the banking industry4 .
Studying the implications of conservatism in the bank setting is important because conservatism is particularly
desirable for banks, for several reasons. First, debt contracting is the primary explanation of why conservatism
arises in financial reporting (Watts, 2003). Since banks are highly leveraged institutions, they are expected to exhibit
higher levels of conditional conservatism due to contracting demands, litigation costs, and regulators’ preferences
(Armstrong, Guay, & Weber, 2010; Watts, 2003). Second, under more conservative accounting, managers report
economic losses in a timelier manner, which counteracts the managerial desire to report only good news and hide bad
news and, consequently, facilitates informational transparency. Because banks have greater information asymmetry
and a complex structure, conservatism can be particularly helpful for banks by requiring the timely disclosure of bad
news (Leventis, Dimitropoulos, & Owusu-Ansah, 2013; Levine, 2004). Central bankers consistently prefer to have
banks exercise conservative accounting practices during economic upturns (Leventis et al., 2013; Turner, 2010). Given
that banks are expected to exhibit high levels of conservatism and that they are also in a position to gain substantial
benefits from conservative reporting, it is important to examine the implications of bank reporting conservatism for
banks’ risk taking or their own monitoring efforts.
In addition to investigating the overall effect of conservatism on loan quality, I introduce two conditional hypothe-
ses. The first hypothesis is that greater conservatism is likely to be more desirable in mitigating risk-taking incentives
for banks when they are inclined to take more risks. Thus, drawing on prior research showing that incentives for risk
shifting are higher for firms with high information asymmetry or distress risk, this study further examines whether
the improvement in loan portfolio quality is more sensitive to conservatism when these banks have high information
asymmetry or experience greater distress. Second, this study considers the role of credit cycles, as prior studies doc-
ument that loan quality is highly sensitive to credit cycles. More specifically, the second hypothesis is that the impact
of conservatism on loan quality is more pronounced for times when the average loan quality deteriorates, that is, dur-
ing times of extreme lending growth cycles. In low lending growth cycles, banks face higher adverse selection costs
because there is a higher likelihood that unknown borrowers have been rejected by other lenders. In high lending
growth cycles, as the perceived benefits of monitoring diminish, banks may be tempted to reduce their monitoring and
screening activities. Furthermore, as bank managers have an incentive to cater to the high demand through excessive
lending (Berger & Udell, 2004; Foos, Norden, & Weber, 2010), the quality of loan portfolios will deteriorate during high
lending growth cycles.
To capture the degree of conditional accounting conservatism among banks, I use Khan and Watts’s (2009) C-Score
measure of timeliness of loss recognition for the main tests as well as the decomposition of the C-Score into the C-
Score associated with loan loss provision and the C-Score associated with earnings before provisions. Additionally, I
develop another measure of conservatism, loan loss provision asymmetry, by modifying the measures of timely recog-
nition of expected loan losses employed by Beatty and Liao (2011), who estimate an equation that regresses loan loss
provisions on lagged (previous two quarters), contemporaneous, and future (one quarter ahead) changes in nonper-
forming loans (NPLs) along with capital and earnings before provisions. In the spirit of Basu’s (1997) piecewise linear
regression model, I similarly modify Beatty and Liao’s (2011) model by decomposing the changes in NPLs into increases
in NPLs ( = bad news) and decreases in NPLs ( = good news). Then I take the average of the coefficients on increases in

4 García Lara, Garcia Osma and Penalva (2016) provide evidence that conservatism reduces underinvestment by facilitating external financing and that it
reduces overinvestment by improving stakeholders’ ability to discipline managers. Balakrishnan, Watts and Zuo (2016) further complement this by using a
setting where firms suffer severe funding problems (during a financial crisis) and find that accounting conservatism reduces underinvestment.
4 HA

NPLs in the past, present and future to capture the extent to which loss provision reflects the asymmetric timeliness
of bad news relative to good news for NPLs.
The main results can be summarized as follows: First, banks with greater conservatism (i.e., banks that recognize
loan losses in a timelier fashion) are likely to have higher quality loan portfolios because of good monitoring of loan-
making decisions. Second, a positive association between conservatism and loan quality is more pronounced for banks
with high information asymmetry or high financial distress. Finally, the positive impact of conservatism on loan portfo-
lio quality is more pronounced during those parts of economic cycles when both the degree of information asymmetry
and uncertainty about the quality of banks’ loan portfolios increase, namely, during recessionary periods (low credit
cycles) as well as expansionary periods (high credit cycles).
I conduct a number of additional tests to examine the robustness of the primary findings. First, I find that the effect
of conservatism on the quality of a loan portfolio holds across all bank size groups. Second, I rerun all the main regres-
sion models with first differences, where the change in loan quality is regressed on the change (from t to t+1) in con-
servatism and in the control variables to mitigate the effect of bank-specific characteristics that are relatively constant
over time. The results are robust to this specification. Third, the results continue to hold with alternative measures of
monitoring quality such as Z-Scores and salary expenses used as a dependent variable. Lastly, the results hold for both
weak and strong corporate governance subsamples.
This study provides direct evidence of how conservatism affects banks’ lending portfolio quality, which reflects
the banks’ own monitoring efforts. While prior studies (Bushman & Williams, 2012, 2015) examine the link between
conservatism and the risk profiles of banks, they mainly focus on systemic economy-wide risks or the market’s percep-
tion as captured by stock behaviors using measures such as stock market illiquidity risk or value-at-risk (Bushman &
Williams, 2015), which is an indirect indicator for banks’ own risk taking. Furthermore, given that loan performance
is closely linked to financial stability, this study contributes to the discussion of financial stability in light of the recent
financial crisis by highlighting the role of banks’ financial reporting quality.
The remainder of the paper is organized as follows. Section 2 presents the background and hypothesis develop-
ment. Section 3 introduces the research methodology. Section 4 introduces the data and descriptive statistics. Sec-
tion 5 covers the empirical analyses of bank monitoring efforts and financial reporting quality. Section 6 details some
additional analyses, and Section 7 concludes the paper.

2 BACKGROUND AND HYPOTHESES

The banking literature posits that informational transparency plays a fundamental role in enhancing market discipline
and prudential bank regulation.5 Market discipline refers to a process by which market participants monitor and dis-
cipline excessive risk taking by banks. For example, bank creditors can exert market discipline by withdrawing their
funds or demanding higher interest rates from riskier banks. For publicly traded banks, equity holders can exert some
discipline by replacing management or by rationing their capital or demanding higher premiums on their investments.
The key to effective market discipline is that banks should provide information regarding their risk exposure and per-
formance reliably and in a timely manner (Stephanou, 2010).
The role of reporting transparency is important, especially since the lack of transparency in financial institutions
played a major role in the 2007–2009 financial crisis by impairing bank supervisors’ and market participants’ under-
standing of bank risks (Acharya & Richardson, 2009). Because banks are more opaque than other industries due to the
inherent complexity of the business and the nature of the underlying assets (Morgan, 2002), transparent information
is essential for investors to monitor the banking industry. If banks choose to withhold information from investors or

5 Banks face two types of information asymmetry. One is information asymmetry between borrowers and depositors, and the other is the asymmetry between

insiders (e.g., managers, loan officers) and outsiders (e.g., capital providers, regulators). This study focuses on the impact of financial reporting quality on the
latter type of information asymmetry, between insiders and outsiders.
HA 5

disclose information that is not credible, they are opaque (Bushman, 2016). Additionally, banks are highly leveraged,
and thus they may exhibit greater conservatism because conservatism tends to enhance debt contracting efficiency
(Watts, 2003). More conservatism can reduce investor uncertainty about banks’ intrinsic value, strengthen market dis-
cipline over risk-taking behavior, and provide disincentives for banks to suppress negative information that can engen-
der capital inadequacy concerns. In this sense, conservatism is an important aspect of bank transparency because it
reduces the managerial temptation to hide bad news by imposing timely recognition of expected loan losses.
Prior studies find evidence suggesting that conservatism improves financial transparency. For example, Gebhardt
and Novotny-Farkas (2011) find that after the mandatory adoption of International Financial Reporting Standards
(IFRS), which required application of the incurred loss approach, banks in the European Union (EU) recognized the
losses in their loan portfolios on a less timely basis. The authors interpret this as an unintended consequence of IFRS, in
particular the IAS 39 impairment rules, because the application of a stricter incurred loss approach reduces bank trans-
parency by limiting management’s ability to communicate private information regarding future credit losses. Andreou
et al. (2017) find that banks that have less conservatism are more likely to experience significant drops in their equity
values. Akins et al. (2017) find evidence that conservatism constrains lending corruption, as measured by a survey
response of borrowers. Manganaris, Beccalli, and Dimitropoulos (2017) find that there was increased conservatism
after the financial crisis among European listed banks, which they interpret as evidence of banks’ attempts to increase
transparency to alleviate the adverse consequences of opaqueness.
Among the multiple benefits of conservatism, the most relevant to this study is conservatism’s ability to reduce
managerial risk shifting behavior. A risk shifting problem is a phenomenon where shareholders of a firm with outstand-
ing risky debt may benefit from increasing the firm’s risk (Jensen & Meckling, 1976). Risk shifting is a more severe prob-
lem in banks than in other industrial firms because (1) debt financing represents the predominant source of external
funding for banks (Freixas & Rochet, 1997), (2) it is relatively easy for banks to alter financial risks without this being
immediately noticed by creditors (Myers & Rajan, 1998), and (3) there are implicit and explicit public guarantees for
debts (Bhattacharya & Thakor, 1993).
Conservatism can prevent banks from taking excessive risk because it helps in the design of effective incentive con-
tracts for managers that assess and reward managerial performance more effectively in the presence of asymmetric
information and payoffs (Armstrong et al., 2010; Bushman & Smith, 2001). Riskier investments would subject firms
to greater potential losses or gains. Under conservative accounting, losses need to be reflected in earnings sooner,
while gains are reported gradually over time. Thus, riskier investments can result in losses being reported for earn-
ings in the short term without a corresponding increase in the probability of reporting larger gains in the short term.
This implies that conservatism can decrease managerial incentives to make risky investments. Consistent with this,
Kravet (2014) finds that under more conservative accounting, managers make less risky acquisitions, and firms with
accounting-based debt covenants drive this association. Similarly, Brockman, Ma, and Ye (2015) find that a conserva-
tive accounting policy dampens CEO incentives to undertake negative NPV risky projects when managerial compen-
sation risk is high. In a similar vein, in banks, managers are deterred from extending loans to riskier borrowers under
conservative accounting. Bushman and Williams (2012) evaluate the impact of conservatism on the risk shifting behav-
ior of banks and find that forward-looking provisions designed to reflect expected loan losses in a timelier manner are
associated with increased discipline. Building on the idea that conditional conservatism reduces the risk of potential
investment distortion and manager expropriation, Li (2015) argues that shareholders are likely to require a lower risk
premium from firms that commit to a conservative financial reporting system in exchange. Consistently, Li (2015) finds
a negative association between conditional accounting conservatism and the cost of capital in an international context.
Because conservatism improves investment efficiency by limiting managerial risk-taking incentives, I expect that
conservatism can promote more prudent loan portfolio selection and deter managers from continuing or increas-
ing loans to default-prone borrowers, resulting in lower default risks in the bank’s loan portfolio. Without conser-
vatism, managers of banks with deteriorating loans have an incentive to postpone revealing this to the market by
increasing their loan volume, which generates profitable upfront fees and boosts earnings even though these loans
will ultimately result in negative net income. When conservatism is required, future anticipated losses need to be fully
6 HA

reflected in loan loss provisioning, which reduces the profit and capital levels on which managers’ compensation is
based.
Based on the preceding discussion, I hypothesize that, ceteris paribus, conservatism disciplines banks’ risk taking
with respect to lending practices, leading to higher quality loan portfolios. My first hypothesis is as follows:

H1: More conservative banks have higher quality loan portfolios.

Next, I derive several auxiliary hypotheses to shed light on the channel through which conservatism affects loan
quality. The relative importance of conservatism in monitoring investment decisions likely varies with firms’ ex ante
risk shifting incentives. I consider two settings where ex ante risk shifting incentives are likely to be high: when banks
have greater inherent information asymmetry and when banks face greater distress risk.
The greater the information asymmetry between banks and outside capital providers, the greater the uncertainty
about the value of a borrower’s assets. In turn, the true value of the borrower as well as the true value of loan
portfolios is largely unknown, which results in greater agency costs for bank stakeholders. Timely recognition of
loan losses can help to directly inform investors or depositors about a bank’s true value (i.e., the value of its loan
portfolios) because the book value of a bank that recognizes losses on a timelier basis provides a lower-bound
estimate for the bank’s orderly liquidation value. Information asymmetry exacerbates risk-taking problems because
greater information asymmetry hinders market monitoring and monitoring by boards of directors and regulators,
implying a greater potential for moral hazard and adverse selection problems (Jensen & Meckling, 1976). Therefore,
managers in firms with greater information asymmetry have greater latitude for indulging their own preferences,
such as consuming excessive perks or taking excessive risks. Information asymmetry can also exacerbate the financing
frictions imposed on banks seeking to raise capital in response to negative balance sheet shocks (Beatty & Liao, 2011;
Bushman & Williams, 2015), which can adversely affect the incentives of bank managers and lead them to make
inefficient investment decisions (Goldstein & Sapra, 2014).
Prior studies suggest that the demand for conservatism arises from information asymmetry and that firms can
benefit more from conditional conservatism when they have greater information asymmetry. For example, LaFond
and Watts (2008) find that information asymmetry between shareholders and managers leads to more conservatism
and that implementation of accounting conservatism reduces information asymmetry. Consistent with this, Nichols,
Wahlen, and Wieland (2009) find that conservatism is more pronounced in public banks than in private banks because
the demand for conservatism increases with greater information asymmetry. Similarly, Martin, Thomas, and Wieland
(2016) find that conditional conservatism decreases (increases) as firms are added to (deleted from) the Standard
and Poor’s (S&P) 500 index, showing that firms choose conditional conservatism to mitigate information asymmetry.
Francis and Martin (2010) document that the positive association between timely loss recognition and acquisition
profitability is more pronounced for firms with higher ex ante information asymmetry. Balakrishnan et al. (2016) and
García Lara et al. (2016) both find that the ability of conservatism to reduce investment inefficiency is greater for
firms that have greater information asymmetry. Conservatism is therefore likely to play an even more prominent
role in reducing managers’ incentives to make unprofitable and overly risky investment decisions or delay project
abandonment decisions. These arguments lead to the next hypothesis:

H2A: A positive association between conservatism and loan quality is more pronounced for banks with high informa-
tion asymmetry.

Jensen and Meckling (1976) offer a risk shifting hypothesis in which managers of financially distressed firms max-
imize the limited liability option of shareholders by investing in risky projects that offer improbably high pay-offs at
the expense of debtholders. Distressed banks are more likely to violate regulatory capital requirements and perform
poorly. Thus, banks could be motivated to use the discretion inherent in loan loss provisioning to avoid violation of
regulatory capital constraints. Following Bushman and Williams (2012), who find that the benefits of forward-looking
HA 7

provisioning in reducing risk shifting problems are more pronounced in banks with low balance sheet capital or a low
return on equity (ROE), I hypothesize that the effects of conservatism on loan portfolio quality are more pronounced
for highly distressed banks by using similar proxies to measure banks’ distress risk:

H2B: A positive association between conservatism and loan quality is more pronounced for banks with high financial
distress.

Prior studies document that loan quality is highly sensitive to credit cycles. Thus, I examine whether the impact of
conservatism on loan quality will be more pronounced for times when the average loan quality deteriorates, that is,
during times of extreme lending growth cycles. For example, during low lending growth cycles (credit crunches), banks
face higher adverse selection costs because there is a higher likelihood that unknown borrowers have been rejected by
other lenders (Dell’Ariccia & Marquez, 2006). During high lending growth cycles that coincide with economic booms,
the default probability for the average loan decreases as firms have a higher probability of having positive net present
value projects to pursue. In turn, during high lending growth cycles, banks may be tempted to reduce their monitoring
and screening activities because the perceived benefits of monitoring efforts are diminished. Furthermore, as bank
managers have an incentive to cater to this high demand through excessive lending (Berger & Udell, 2004; Foos et al.,
2010), the quality of loan portfolios will deteriorate. Consistent with this, Andreou et al. (2017) find that conservatism
reduces the risk of abrupt stock price crashes, particularly during low and high lending growth cycles6 .
As a result, during both low and high lending growth cycles, loan quality is likely to deteriorate, and conservatism
can play a greater role in improving loan quality as there is more room for improvement. In turn, conservatism and
future loan quality can be more positively associated during low and high lending growth cycles than during moderate
lending growth cycles. Based on the above reasoning, I propose the following hypothesis:

H3: The impact of conservatism on loan portfolio quality is more pronounced during low and high lending growth
cycles than during moderate lending growth cycles.

3 RESEARCH METHODOLOGY

3.1 Conservatism proxies

3.1.1 Khan and Watts’s (2009) firm-level conservatism measure (C-Score)

Following Beatty and Liao (2011), I use (with modification) Khan and Watts’s (2009) bank-specific conditional conser-
vatism measure, which is based on Basu’s (1997) differential timeliness of earnings measure. Given the premise that
market returns capture information about both past and future loan losses, timelier provisioning practices would be
reflected in a greater association between negative returns and current net income. Following prior research, I remove
observations with a price per share of less than $1 and with a negative book value of equity. I also require at least five
annual observations to run each regression:

NIit = 𝛽0 + 𝛽1 Dit + Rit (𝜇1 + 𝜇2 SIZEit + 𝜇3 MBit + 𝜇4 LEVit )

∗
+ Dit Rit (𝜆1t + 𝜆2t SIZEit + 𝜆3t MBit + 𝜆4t LEVit )
( ∗ ∗
+ 𝛿1t SIZEit + 𝛿2t MBit + 𝛿3t LEVit + 𝛿4t Dit SIZEit + 𝛿5t Dit MBit 𝛿6t Dit LEVit ) + 𝜀t , (1)

6 Duringmoderate lending growth cycles, where loan volumes are ordinary and screening and monitoring of loan portfolios are normally conducted, any
incremental monitoring benefits from timely loss recognition would be insignificant.
8 HA

where the subscripts i and t respectively represent the firm and fiscal year-quarter; NI is net income (COMPUSTAT
‘ni’) divided by the lagged market value of equity (COMPUSTAT ‘csho’ * share price at the fiscal quarter end); D is an
indicator variable that equals one for negative returns and zero otherwise; R is quarterly returns compounded from
monthly returns beginning the second month after the fiscal quarter starts to ensure the market response to the prior
quarterly earnings is excluded; SIZE is the natural log of the market value of equity (COMPUSTAT ‘csho’ * share price
at the fiscal quarter end) at the end of the fiscal year; MB is the market value of equity divided by the book value
of equity (COMPUSTAT ‘ceq’); and LEV is long-term debt (COMPUSTAT ‘dltt’) divided by the market value of equity
(COMPUSTAT ‘csho’ * share price at the fiscal quarter end).
After equation (1) is estimated, the C-Score is constructed using the estimated coefficients as follows:

C−Scoreit = 𝛽3 = 𝜆1t + 𝜆2t SIZEt + 𝜆3t MBt + 𝜆4t LEVt + 𝜀. (2)

Given this construction, the higher the C-Score, the timelier the loss recognition relative to gain recognition. Fur-
thermore, to better understand the sources of conservatism, I decompose the C-Score into two components: (i) loan
loss provision conservatism (C-Score_LLP) and (ii) earnings before provision conservatism (C-Score_EBP), following
Andreou et al. (2017). Specifically, I follow the same approach described above but replace the dependent variable,
NI, with either loan loss provision (LLP) or earnings before provision (EBP), and then rerun equation (1). To reduce the
measurement errors, both C-Score_LLP and C-Score_EBP are defined as indicator variables if each value is greater than
the median during the quarter, and zero otherwise. Because conservatism is expected to manifest through managers’
use of discretionary loan loss provision, CON_LLP is expected to drive the results.

3.1.2 Loan loss provision asymmetry measure (LLP_ASY)

Although the Basu model is widely used for testing conditional conservatism, it has some weaknesses. For example,
Cano-Rodriguez and Nunez-Nicel (2015) find that the Basu model produces an omitted variable bias and a truncated-
sample bias because it uses the aggregated market returns of the period rather than separating good news from
bad news in the period under consideration. Therefore, I develop another conservatism measure by modifying the
approaches employed by Beatty and Liao (2011) and Bushman and Williams (2012). In developing a measure of delay
in loss provisioning, Beatty and Liao (2011) estimate equations in which loan loss provisions are regressed on lagged,
contemporaneous and future changes in NPLs along with capital and earnings before provisions. Following Basu’s
(1997) piecewise linear regression in which net income is regressed on positive and negative stock returns, I similarly
modify Beatty and Liao’s (2011) model by decomposing the changes in NPLs into increases in NPLs ( = bad news) and
decreases in NPLs ( = good news) and estimate the extent to which loss provisions reflect the asymmetric timeliness of
bad news relative to good news for NPLs.7 Specifically, I estimate the following piecewise linear model. To extract the

7 The LLP_ASY measure builds on the assumption that increases (decreases) in NPLs can be interpreted as bad news (good news). However, to the extent
that NPLs may decrease because they are charged off, interpreting a decrease in NPLs as good news poses a problem. I address this issue in two ways. First,
I calculate the correlation coefficients among the change in NPLs and the contemporaneous and four-quarters-ahead levels of net charge-offs (NCOs) to
confirm that there is no correlation between these variables. If decreased NPLs are predominantly caused by increased charge-offs, then the correlation
coefficient between ΔNPL and NCO should be negative. However, untabulated results show that the correlation coefficients between ΔNPL and current as
well as four-quarters-ahead NCOs are positive and significant, suggesting that decreased NPLs would not coincide with or be followed by increased charge-
offs. More specifically, the correlation coefficient between ΔNPLt and NCOt is 0.024 (p-level: 0.0004) while the correlation coefficient between ΔNPLt and
NCOt+4 is 0.165 (p-level: < 0.0001). This result alleviates the concern that decreased NPLs predominantly capture charge-offs. Second, to further ensure that
the effects of charge-offs do not interfere with the interpretation of an increase (a decrease) in NPLs as an indicator of bad news (good news), I partition the
original sample into two subsamples, one with the below-median charge-offs and the other with the above-median charge-offs, and run the main regression
separately for each subsample. This analysis is based on the premise that to the extent that changes in NPLs indicate something other than the direction of
news (either bad or good), then the construct validity of LLP_ASY as a conservatism measure is weakened. If decreases in NPLs are mostly attributable to
charge-offs and not to an improved loan portfolio quality (thus good news), then the effect of conservative reporting on loan quality will be less pronounced
for the above-median charge-offs subsample, and vice versa. Untabulated results show that the coefficient of CON is negative and significant at the 1% level
HA 9

bank-quarter-level conservatism measure, I run the rolling regressions for each bank quarter using the observations
of the past three years and requiring 12 observations for each regression8 :

( ∗ )
LLPt+4 = 𝛼0 + 𝛼1 ΔNPLt−2 + 𝛼2 DΔNPLt−2 + 𝛼3 DΔNPLt−2 ΔNPLt−2 + 𝛼4 ΔNPLt−1 + 𝛼5 DΔNPLt−1
( ∗ ) ( ∗ )
+ 𝛼6 DΔNPLt−1 ΔNPLt−1 + 𝛼7 ΔNPLt + 𝛼8 DΔNPLt + 𝛼9 DΔNPLt ΔNPLt + 𝛼10 ΔNPLt+1
( ∗ )
+ 𝛼11 DΔNPLt+2 + 𝛼12 DΔNPLt+1 ΔNPLt+1 + 𝛼13 CAPITALt + 𝛼14 EBPt + 𝜀t , (3)

where DΔNPLt+x is a dummy variable that is set to one for bad news for credit losses, that is, when ΔNPLt+x is positive,
and zero otherwise. The capital ratio relative to total assets (CAPITAL) and earnings before provision (EBP) are included
to control for banks’ incentives to manipulate provisioning to meet regulatory requirements or to smooth earnings.
I focus on the association between loan loss provisions and lagged, contemporaneous and future changes in NPL.
More specifically, I use the lagged increase in NPL (DΔNPLt-1 * ΔNPLt-1 ), contemporaneous increase in NPL (DΔNPLt *
ΔNPLt) , and future increase in NPL (DΔNPLt+1 * ΔNPLt+1 ) to capture timelier recognition of loan losses versus gains.
The more positive association between loan loss provisions and increases in NPL for each time period is consistent
with more conservatism, that is, bank managers who report conservatively will incorporate future increases in NPL in
provisions sooner than future decreases in NPL.
Therefore, the coefficients α3, α6, α9, and α12 capture the degree of the conservative provisioning practice. I refer
to the average of these four coefficients as loan loss provision asymmetry, denoted as LLP_ASY.
The coefficients on DΔNPLt-2 * ΔNPLt-2 (α3 ) and DΔNPLt-1 * ΔNPLt-1 (α6 ) capture the extent to which current pro-
visions reflect past deteriorations in the performance of the loan portfolios. The coefficient on DΔNPLt * ΔNPLt (α9 )
picks up the extent to which current provisions incorporate contemporaneous deteriorations in the quality of the loan
portfolio, while the coefficient on DΔNPLt+1 * ΔNPLt+1 (α12 ) picks up the extent to which current provisions explic-
itly anticipate future deteriorations in the performance of the loan portfolio. Because the current GAAP requires an
‘incurred loss’ methodology for recognizing credit losses that delays recognition until it is probable that a loss has been
incurred, as robustness checks, I exclude α9 and α12 and include only the average of the coefficients on lagged changes
in NPLs. Untabulated results show that the inferences remain unchanged using this alternative measure.
To reduce the measurement errors, I use an indicator variable that is equal to one if each conservatism measure is
greater than the median during the quarter, and zero otherwise.

3.2 Empirical models

3.2.1 Test of hypothesis H1

I use the following equation (4) to test hypothesis H1. I have included quarter and bank fixed effects for all regres-
sion analyses, and statistical significance is based on heteroskedasticity-robust standard errors clustered by bank and
quarter.

LQit+4 = 𝛽0 + 𝛽1 CONit + 𝛽2 LQit + 𝛽3 TIMELINESS + 𝛽4 SIZEit + 𝛽5 ABSGAPit + 𝛽6 FEEit

+ 𝛽7 ROAVOLit + 𝛽8 EFFit + 𝛽9 EBLLPit + 𝛽10 LOANit + 𝛽11 LGit + 𝛽12 COMMit + 𝛽13 REit

+ 𝛽14 CONSit + 𝛽15 CAPit + 𝛽16 LIQUIDit + 𝛽17 ΔGDPit + Year − Quarterdummies + 𝜀it (4)

in both subsamples. The results continue to hold in both the top and bottom decile charge-off subsamples. This analysis mitigates the concern that a decrease
in NPLs is driven by charge-offs and strengthens the construct validity of LLP_ASY as a measure of conservatism.
8 Using an international sample, Bushman and Williams (2012) estimate a similar equation for each country and use the coefficient on ΔNPL
t+ 1 as a proxy for
forward-looking discretion in loan provisions and the coefficient on EBP as a proxy for smoothing discretion in loan provisions.
10 HA

The unit of observation is the bank quarter, with i indexing banks and t indexing quarters. LQit+4 represents loan
portfolio quality measured by four-quarters-ahead NPLs or net charge-offs (NCOs), which are respectively a stock and
a flow measure.9 CON is one of the conservatism measures C-Score, C-Score_EBP, C-Score_LLP, and LLP_ASY. Additionally,
I later use two additional proxies for bank monitoring quality—the Z-Score and the industry-median-adjusted salary
expense—as dependent variables for robustness checks. A more detailed description of these measures can be found
in Section 6.3.
I control for variables known to affect loan quality, largely based on Bhat and Desai (2017). LQ at time t is included
to control for serial correlation in NPLs and NCOs. TIMELINESS is the ratio of allowance (reserve for credit losses) over
nonperforming loans as used by Beatty and Liao (2011). This variable is controlled to establish that the disciplinary
effects of asymmetric timeliness are above and beyond those of timely loan loss recognition. SIZE is the natural log-
arithm of the bank’s total assets, ABSGAP is the absolute value of the one-year maturity gap deflated by total assets,
FEE is the ratio of fee income to total income, ROAVOL is the standard deviation of ROA over the past five years, EFF
is the ratio of non-interest expense to income before extraordinary items, EBLLP is earnings before LLP, LOAN is loans
deflated by total assets, LG is loan growth deflated by total assets, COMM is agricultural and commercial loans deflated
by total assets, RE is real estate loans deflated by total assets, CONS is consumer loans deflated by total assets, CAP
is equity capital deflated by total assets, and LIQUID is liquid assets deflated by total assets. The qualities of loans
are expected to deteriorate during slow economic cycles, so the change in GDP is controlled. ΔGDP is the quarterly
GDP growth rate. Year-quarter dummy variables are included to capture the static level of loan quality that is due to
macroeconomic factors and to account for the unobservable time-invariant heterogeneity.
Four-quarters-ahead loan quality is used as a dependent variable rather than a loan quality measure contempo-
raneous with independent variables to allow some time for the monitoring effect of conservatism to take effect on
loan quality and to mitigate endogeneity. As bank monitoring effort is reflected in the bank asset quality (mainly loan
portfolios), I use loan quality as a main proxy for the bank monitoring effort. Berger and DeYoung (1997) argue that
NPLs serve as the most commonly agreed-upon definition of problem loans in both academic research and the trade
press and are difficult for managers to manipulate because loans must be reported as nonperforming if they are at least
90 days past due. While there is evidence that banks used discretionary NCOs to manage capital prior to the 1989 reg-
ulatory change (Beatty, Chamberlain, & Magliolo, 1995; Collins, Shackelford, & Wahlen, 1995), charge-offs are often
driven by exogenous factors, and some loans such as consumer loans are automatically charged off when they become
delinquent for a certain number of days. Therefore, NCOs also provide a relatively nondiscretionary measure of bank
loan quality.
Additionally, my choice of NPLs and NCOs as dependent variables is firmly grounded in conservatism research.
Conditional conservatism results in losses being reported sooner than gains, which incentivizes managers to make
efficient investments. Prior research on conservatism uses managers’ choice of value-enhancing investments in the
first place as well as early abandonment of poorly performing projects as a sign of efficient investments (for example,
Bushman, Piotroski, & Smith, 2011; García Lara et al., 2016; Kravet, 2014). In the banking context, value-enhancing
investments are akin to having low NPLs, while early abandonment of poorly performing investments would lead to
low NCOs. For example, if loan officers make more prudent loan decisions under conservative accounting, they are
less likely to make risky loans that are predicted to be nonperforming. Also, if loan officers promptly act on poorly
performing loans by charging off bad loans on a timely basis, this will prevent a large buildup of charge-offs, which
subsequently leads to low NCOs.
Note that both NPLs and NCOs are a negative function of loan quality, so a significantly negative coefficient β1 on
CON is consistent with the hypothesis that greater conservatism is positively associated with a bank’s own monitoring
effort. All variables and their construction are detailed in Appendix A.

9 I also consider one-quarter-ahead NPLs and NCOs. This does not change the inferences.
HA 11

3.2.2 Test of hypotheses H2A and H2B

I modify equation (4) to test whether the association between conservatism and loan quality is stronger for firms with
high risk shifting incentives than firms with low risk shifting incentives. I estimate the following model:

( ∗ )
LQit+4 = 𝛽0 + 𝛽1 CONit + 𝛽2 RISKSHIFTit + 𝛽3 CONit RISKSHIFTit + 𝛽n Control Variablesit

∗
+ RISKSHIFTit (𝜇n ControlVariablesit ) + Year − Quarter dummies + 𝜀it , (5)

where RISKSHIFT is a proxy of either information asymmetry or distress risk. I also use a fully interacted model by
including interaction variables for RISKSHIFT and the control variables to allow the coefficients on control variables to
vary with RISKSHIFT10 .
Information asymmetries are measured in two ways. The first measure, INFASY, is a market-based measure, which
is the average of the standardized bid–ask spread, stock return volatility, and idiosyncratic risk. The second measure,
JDC, is information asymmetries between the firm and debtholder as reflected in the cost of borrowing. Borrow-
ers’ perceptions of a bank’s risk taking are reflected in the cost of borrowing because rational debtholders recognize
managers’ incentives to engage in risky activities and incorporate the value implications of managers’ future actions
into the pricing of debt. Specifically, I examine the imputed interest rate on large time deposits (jumbo certificates of
deposit, or JCD).11 I choose the JCD market to proxy for borrowers’ perceptions of a bank’s risk shifting incentives
because most banks use this market as their primary source of marginal funds (Kishan & Opiela, 2012).
Distress risks are measured in two ways, following Bushman and Williams (2012). The first measure, POORPERF,
captures distress risks reflected in the ROE. Because risk shifting incentives will go up as banks’ performance deterio-
rates, a low ROE represents a high distress risk. I designate banks with ROEs in the bottom quintile as poor performers.
The second measure, LOWCAP, is a balance-sheet-based measure that captures distress risks reflected in capital lev-
els. Because banks’ shareholders will benefit more from risk shifting as banks move closer to a violation of regulatory
capital requirements, low capital levels indicate high distress risk. I classify banks with a capital ratio less than 7% as
low capital banks.
A significantly positive coefficient on β3 is consistent with hypotheses H2A and H2B, which state that the bene-
fit of conservatism in improving banks’ own monitoring effort is more pronounced for banks with high risk shifting
incentives.

3.2.3 Test of hypothesis H3

The monitoring benefits of conservatism are likely to vary depending on the degree of information asymmetry
between managers and capital providers. Presumably, during low and high lending growth cycles, which coincide with
economic booms and busts, information asymmetry regarding banks’ asset quality is more severe because the poten-
tial risk of generating problem loans is higher than it is in normal times. Therefore, the monitoring benefits of conser-
vatism are likely to be more prominent during low and high lending growth cycles because conservatism can lower the
negative consequences of information asymmetry by imposing timely reporting of expected loan losses.
To test this conjecture, I modify equation (4) and include dummy variables that capture the different states of the
lending growth cycles (high, moderate, or low) as well as the interaction between these dummy variables and CON to
see whether the impact of conservatism on loan monitoring efforts changes under the different lending growth cycles.

10 The results do not change without the interaction terms between the conditioning variable and the control variables.

11 JCDs (larger than $100,000) were fully uninsured until Q3 2008 and partly uninsured (over $250,000) thereafter.
12 HA

Following Andreou et al. (2017), I employ the following model:

∗ ∗
LQit+4 = 𝛽0 + 𝛽1 CONit + 𝛽2 (CONt LOWt ) + 𝛽3 (CONt HIGHt ) + 𝛽4 LOWt + 𝛽5 HIGHt

+ 𝛽n Control Variablesit + Year − Quarter dummies + 𝜀it , (6)

where HIGH and LOW are indicator variables that correspond to high and low lending growth cycles, respectively.
The high lending growth cycle variable gets the value one for the years 2000, 2005–2008, and 2011–2015, and zero
otherwise. The low lending growth cycle variable is set to one for the years 2001–2004 and 2009–2010, and zero
otherwise. The moderate lending growth cycle is for the years 1994–1999, but it is excluded from the model in order
to be used as a comparison group. Negative coefficients for CON * LOW and CON * HIGH are consistent with H3, which
states that conservatism plays a more prominent role in helping a bank’s own loan monitoring efforts during low and
high lending growth cycles compared to moderate lending growth cycles. All the regression models include bank fixed
effects to control for omitted time-invariance variables.

4 DATA

The sample includes publicly traded bank holding companies in the United States during the period 1994–2015. The
sample period starts with 1994 because Compustat Bank does not report quarterly NPLs prior to 1993. I focus on
publicly traded banks since the stock price is an input for my conservatism measures. I obtained the stock prices of
banks from CRSP, and I extracted financial statement data from Compustat Bank Quarterly and the quarterly call
report data (Y9C) from the Federal Reserve Bank of Chicago website. After removing bank-quarter observations with
missing data for any variables used in the main analysis, the final sample consists of 18,158 bank-quarters represent-
ing 635 bank holding companies (BHCs) when C-Score is used and 17,049 bank-quarters representing 600 unique
BHCs when LLP_ASY is used. All independent variables are winsorized at the 1% level to help control for outliers
in the sample. The descriptive statistics are provided in Panel A of Table 1. The mean NPL relative to total assets is
0.008, while the mean NCO relative to total assets is 0.002. The average bank in my sample has assets of $14.92 billion
(median = $14.60 billion), and the mean (median) equity–capital ratio is 9.1% (8.9%). The average bank derives about
25% of its income from fee-generating activities such as advisory services. For the average bank in the sample, earn-
ings before loan loss provisions represent 0.008 of the total assets, and the standard deviation of return on assets is
0.004. The average loan relative to assets is 66%.
Table1, Panel B reports Pearson correlations. The correlation coefficient between the two loan quality measures,
NPL and NCO, is 0.58, indicating that while the loan quality proxies are substantially positively correlated, they cap-
ture different aspects of default risks. As predicted, both proxies of CON, namely Cscore_LLP and LLP_ASY, are signifi-
cantly negatively correlated with the proxy for loan portfolio quality, which provides initial evidence that conservatism
improves bank asset quality through the discretionary channels of loan loss provisions. In addition, both proxies of CON
are positively and significantly correlated. However, as the correlation coefficient is 0.08, the proxies capture some-
what different dimensions of CON. Both CON proxies are positively related to SIZE, FEE, EFF, EBLLP, CONS and LIQUID,
indicating that banks with greater conservatism are likely to be large, profitable and liquid and to generate a larger
proportion of their income from fees and consumer loans. The CON proxies are negatively related to LOAN, LG, COMM
and RE, indicating that banks with greater conservatism have lower loans relative to assets, lower loan growth, and
lower commercial and real estate lending.
 HA

TA B L E 1 Descriptive statistics and correlations

Panel A: Descriptive Statistics

Variable N Mean Median Q1 Q3 Std Dev
NPL 18158 0.008 0.005 0.001 0.011 0.010
NCO 18158 0.002 0.001 0.000 0.002 0.003
C-Score_LLP 18158 0.504 0.556 0.444 0.556 0.272
LLP_ASY 17049 0.501 0.500 0.250 0.750 0.351
TIMELINESS 18158 0.504 0.556 0.222 0.778 0.318
SIZE 18158 14.921 14.595 13.884 15.695 1.329
ABSGAP 18158 0.673 0.659 0.518 0.815 0.225
FEE 18158 0.247 0.233 0.171 0.306 0.109
ROAVOL 18158 0.004 0.003 0.003 0.004 0.001
EFF 18158 0.025 0.023 0.013 0.034 0.013
EBLLP 18158 0.008 0.008 0.004 0.011 0.005
LOAN 18158 0.662 0.674 0.604 0.734 0.107
LG 18158 0.122 −0.037 −0.106 0.041 0.631
COMM 18158 0.122 0.109 0.070 0.161 0.072
RE 18158 0.454 0.455 0.359 0.556 0.140
CONS 18158 0.056 0.037 0.009 0.093 0.056
CAP 18158 0.091 0.089 0.077 0.103 0.021
LIQUID 18158 0.230 0.219 0.160 0.290 0.100
ΔGDP 18158 0.025 0.027 0.018 0.038 0.017
INFASY 10714 −0.013 −0.214 −0.549 0.302 0.801
JCD 16763 0.523 0.500 0.250 0.750 0.346
ROE 18158 0.062 0.062 0.032 0.099 0.080

(Continues)
13
 (Continued)
14

TA B L E 1

Panel B: Correlation Analysis

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Variables NCO C-Score_LLP LLP_ASY TIMELINESS SIZE ABSGAP FEE ROAVOL EFF EBLLP LOAN LG COMM RE CONS CAP LIQUID ΔGDP
NPL 0.58 −0.14 −0.48 −0.30 0.00 0.02 −0.07 0.12 −0.19 −0.06 0.26 0.17 −0.03 0.35 −0.29 0.11 −0.10 −0.54
(0.00) (0.00) (0.00) (0.00) (0.48) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
NCO −0.14 −0.38 −0.19 0.13 0.11 0.01 0.19 0.12 0.27 0.22 0.11 0.07 0.12 −0.01 −0.02 −0.17 −0.39
(0.00) (0.00) (0.00) (0.00) (0.00) (0.06) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.05) (0.00) (0.00) (0.00)
C-Score_LLP 0.08 0.11 0.03 −0.09 0.06 0.01 0.05 0.08 −0.12 −0.01 −0.03 −0.09 0.02 0.18 0.08 0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.06) (0.00) (0.00) (0.00) (0.08) (0.00) (0.00) (0.00) (0.00) (0.00) (0.70)
LLP_ASY 0.37 0.20 −0.13 0.13 −0.02 0.13 0.04 −0.15 −0.05 −0.01 −0.21 0.14 −0.01 0.07 0.39
(0.00) (0.00) (0.00) (0.00) (0.03) (0.00) (0.00) (0.00) (0.00) (0.27) (0.00) (0.00) (0.28) (0.00) (0.00)
TIMELINESS 0.18 −0.11 0.10 0.02 0.05 0.06 −0.09 −0.03 0.05 −0.14 0.05 0.06 0.05 0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.95)
SIZE −0.04 0.54 0.08 0.08 0.14 −0.19 0.09 0.19 −0.41 0.14 −0.01 −0.01 0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.13) (0.03) (0.90)
ABSGAP −0.02 0.06 −0.02 0.00 0.21 −0.02 0.22 0.05 0.05 −0.14 −0.21 0.01
(0.01) (0.00) (0.01) (0.81) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.34)
FEE 0.02 0.29 0.13 −0.26 0.03 0.08 −0.38 0.10 −0.03 0.11 0.00
(0.01) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.66)
ROAVOL 0.07 0.18 0.10 0.04 0.04 0.06 −0.03 0.11 −0.07 −0.02
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
EFF 0.77 −0.04 −0.05 0.06 −0.16 0.16 0.06 −0.04 0.16
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
EBLLP 0.05 −0.03 0.09 −0.10 0.16 0.14 −0.08 0.08
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

(Continues)
HA
 HA

TA B L E 1 (Continued)

Panel B: Correlation Analysis

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Variables NCO C-Score_LLP LLP_ASY TIMELINESS SIZE ABSGAP FEE ROAVOL EFF EBLLP LOAN LG COMM RE CONS CAP LIQUID ΔGDP
LOAN 0.06 0.22 0.71 0.03 0.06 −0.60 −0.13
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
LG 0.00 0.08 −0.11 0.06 −0.01 −0.12
(0.50) (0.00) (0.00) (0.00) (0.18) (0.00)
COMM −0.33 0.02 0.01 −0.12 0.04
(0.00) (0.00) (0.30) (0.00) (0.00)
RE −0.37 0.07 −0.36 −0.24
(0.00) (0.00) (0.00) (0.00)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Variables NCO C-Score_LLP LLP_ASY TIMELINESS SIZE ABSGAP FEE ROAVOL EFF EBLLP LOAN LG COMM RE CONS CAP LIQUID ΔGDP
CONS −0.11 −0.16 0.26
(0.00) (0.00) (0.00)
CAP 0.00 −0.11
(0.77) (0.00)
LIQUID 0.07
(0.00)

Notes: Panel A of Table 1 presents descriptive statistics for the full sample of winsorized (1% of top/bottom) key variables. The sample spans the 1994–2015 period and excludes the
financial and utilities industries. Variable definitions are presented in the Appendix. Panel B of Table 1 presents the Pearson correlations for key variables. P-values are below the coefficient
estimates, in parentheses.
15
16 HA

5 RESULTS

5.1 Baseline results: Testing hypothesis H1

Table 2 summarizes the estimation results of the multivariate regressions that test the effect of conservatism on loan
quality. Note that a negative coefficient on CON is consistent with the hypothesis that greater conservatism leads
banks to improve their loan quality, presumably through enhanced monitoring of bank managers’ lending decisions.
In Panel A, when NPL is used as a dependent variable, Column (1) reports the results from regressing NPL on C-
Score using ordinary least squares. After controlling for bank characteristics, the coefficient on CON is negative and
statistically significant at the 1% level. In terms of economic significance, changing the C-Score dummy from zero to
one while holding the other independent variables at their means decreases NPL relative to the total loans by about
0.4%. Column (2) and Column (3) report the results from regressing NPL on the decomposed C-Score, that is, the C-
Score associated with EBP and the C-Score associated with LLP, respectively. Column (2) shows that when C-Score_EBP
is used, the coefficient on CON is positive and insignificant (coeff = 0.001, T-stat = 0.29), the opposite of the predicted
sign, while Column (3) shows that when C-Score_LLP is used, the coefficient on CON is negative and highly significant at
less than the 1% level (coeff = −0.002, T-stat = −8.12), confirming that the results reported in Column (1) are driven
by C-Score_LLP and not C-Score_EBP. Changing the C-Score_LLP dummy from zero to one decreases NPL by 0.2%, repre-
senting an almost 25% decrease in NPL. Similarly, Column (4) shows that when LLP_ASY is used, the coefficient on CON
is negative and highly significant (coeff = −0.003, T-stat = −14.97), representing an almost 38% decrease in NPL.
In Panel B, where NCO is used as a dependent variable, the inferences remain unchanged. Column (1) reports the
results from regressing NCO on C-Score using ordinary least squares. The coefficient on C-Score_LLP is negative and
statistically significant at the 1% level. In terms of economic significance, changing C-Score_LLP from zero to one while
holding the other independent variables at their means decreases NCO relative to total assets by about 0.1%. Given
the mean value of NCO (0.002), this represents a decrease of 50% in NCO. The results that C-Score_LLP is significant
while C-Score_EBP is insignificant support the hypothesis that conservatism mainly operates through the discretionary
application of LLP and not EBP, which is relatively nondiscretionary. Similar results using LLP_ASY are shown in Col-
umn (4). The coefficient on CON is −0.001 and statistically significant at the 1% level, consistent with the hypothesis
that conservatism is associated with better loan portfolio quality. In sum, the effect of CON on loan portfolio qual-
ity is statistically and economically significant in improving loan portfolio quality, manifesting the monitoring role of
conservatism.
Turning to the control variables, the coefficient on LQ is positive and significant across all columns, confirming serial
autocorrelation in loan quality. None of the coefficients on TIMELINESS are significant with NPL as a dependent vari-
able. On the other hand, the coefficient on TIMELINESS is negative and significant with NCO as a dependent variable
with C-Score, C-Score_EBP, and C-Score_LLP as a proxy for CON, but not significant with LLP_ASY as a proxy. A negative
sign suggests that more timely provisioning practices improve loan quality. While the coefficient on SIZE is positive
and significant for only LLP_ASY as a CON proxy with NPL as a dependent variable, it is positive and significant for all
four CON proxies with NCO as a dependent variable. A positive sign suggests that larger banks have poorer loan port-
folio quality. The coefficients on ABSGAP are positive and significant, implying that banks with higher interest risks
have lower loan portfolio quality. The coefficient on FEE is negative for only LLP_ASY as a CON proxy with NPL as a
dependent variable and negative and significant for all four CON proxies with NCO as a dependent variable. A negative
sign suggests that banks drawing a higher percentage of their income from fee-generating activities rather than tradi-
tional deposit-taking and lending activities have better loan portfolio quality. The coefficients on ROAVOL are positive,
implying that banks with greater operating volatility have poorer loan portfolio quality. The coefficients on EFF are
negative when NPL is used as a dependent variable, but positive when NCO is used as a dependent variable. A negative
sign indicates that banks with higher cost efficiency have better loan portfolio quality, and vice versa. More specif-
ically, these coefficients on EFF suggest that banks with higher cost efficiency have low NPLs and high NCOs. The
HA 17

TA B L E 2 Analysis of the effect of conservatism on loan quality

LQit+4 = 𝛽0 + 𝛽1 CONit + 𝛽2 LQit + 𝛽3 TIMELINESS + 𝛽4 SIZEit + 𝛽5 ABSGAPit + 𝛽6 FEEit + 𝛽7 ROAVOLit + 𝛽8 EFFit
+ 𝛽9 EBLLPit +10 LOANit + 𝛽11 LGit + 𝛽12 COMMit + 𝛽13 REit + 𝛽14 CONSit + 𝛽15 CAPit + 𝛽16 LIQUIDit
+ 𝛽17 ΔGDPit + Year − quarter dummies + 𝜀it (4)

Panel A: Using NPL as the dependent variable

Dep Var = NPL at t + 4
CON = C-Score CON = C-Score_EBP CON = C-Score_LLP CON = LLP_ASY
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept 0.000 −0.06 0.000 −0.09 −0.005 −3.95 ***
0.001 0.29
CON −0.004 −3.11*** 0.001 0.29 −0.002 −8.12*** −0.003 −14.97***
LQ 0.682 41.58*** 0.690 42.80*** 0.667 41.08*** 0.670 39.32***
TIMELINESS 0.000 −0.13 0.000 −0.14 0.000 0.11 0.000 0.92
SIZE 0.000 1.13 0.000 0.96 0.000 1.02 0.000 3.43***
ABSGAP 0.001 4.13*** 0.001 4.19*** 0.002 4.63*** 0.001 4.16***
FEE −0.001 −0.99 −0.001 −1.05 −0.001 −0.71 −0.002 −2.11**
*** *** ***
ROAVOL 0.274 4.47 0.291 4.72 0.312 5.22 0.304 4.87***
EFF −0.027 −2.96*** −0.025 −2.78*** −0.026 −2.98*** −0.018 −1.97**
*** ** ***
EBLLP 0.076 2.77 0.067 2.44 0.077 2.87 0.062 2.19**
LOAN 0.004 2.25** 0.004 2.27** 0.002 0.83 0.004 2.03**
LG 0.000 4.26*** 0.000 4.41*** 0.000 4.04*** 0.000 3.80***
COMM −0.002 −1.15 −0.003 −1.21 0.000 −0.21 −0.003 −1.63
RE 0.000 0.05 0.000 0.05 0.002 1.09 0.000 −0.16
CONS −0.003 −1.24 −0.003 −1.29 0.001 0.24 −0.003 −1.32
CAP −0.004 −0.96 −0.005 −1.10 −0.002 −0.63 −0.004 −1.00
LIQUID −0.002 −2.13** −0.002 −2.21** −0.003 −2.75*** −0.003 −2.37**
ΔGDP −0.142 −0.81 −0.130 −0.73 0.138 10.57 ***
−0.197 −1.11
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 76.19% 76.11% 77.30% 76.08%
No. obs. 18,158 18,158 18,158 17,049

Panel B: Using NCO as the dependent variable

Dep Var = NCO at t+4
CON = C-Score CON = C-Score_EBP CON = C-Score_LLP CON = LLP_ASY
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.005 −3.82*** −0.005 −3.70*** −0.001 −2.15** −0.005 −3.30***
CON −0.003 −4.95 ***
−0.001 −0.83 −0.001 −9.65 ***
−0.001 −17.35***
LQ 0.584 37.33*** 0.596 37.95*** 0.582 39.90*** 0.575 38.55***
TIMELINESS 0.000 −4.42 ***
0.000 −4.45 ***
0.000 −3.83 ***
0.000 −1.07
*** *** ***
SIZE 0.000 4.29 0.000 3.93 0.000 3.93 0.000 7.75***
ABSGAP 0.000 3.71*** 0.000 3.83*** 0.000 3.95*** 0.000 3.59***
FEE −0.001 −3.62 ***
−0.001 −3.62 ***
−0.001 −3.42 ***
−0.001 −5.29***

(Continues)
18 HA

TA B L E 2 (Continued)

Panel B: Using NCO as the dependent variable

Dep Var = NCO at t+4
CON = C-Score CON = C-Score_EBP CON = C-Score_LLP CON = LLP_ASY
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
ROAVOL 0.131 6.30*** 0.142 6.81*** 0.142 7.04*** 0.145 7.38***
** ** *
EFF 0.007 2.10 0.008 2.20 0.006 1.76 0.014 4.00***
EBLLP 0.056 4.92*** 0.050 4.39*** 0.058 5.14*** 0.041 3.86***
** ** ***
LOAN 0.002 2.50 0.002 2.47 0.002 2.69 0.002 3.49***
LG 0.000 4.83*** 0.000 5.03*** 0.000 5.09*** 0.000 4.52***
COMM 0.000 −0.48 0.000 −0.51 −0.001 −0.91 −0.001 −1.51
RE −0.001 −0.88 −0.001 −0.85 −0.001 −1.25 −0.001 −1.57
CONS 0.001 0.69 0.001 0.62 0.000 0.08 0.000 0.53
CAP −0.005 −4.03 ***
−0.006 −4.39 ***
−0.004 −3.54 ***
−0.005 −4.48***
LIQUID 0.000 −1.40 0.000 −1.56 0.000 −1.41 0.000 −0.96
ΔGDP 0.111 2.12** 0.112 2.13** −0.035 −3.80*** 0.074 1.33
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 61.99% 61.60% 63.18% 62.24%
No. obs. 18,158 18,158 18,158 17,049

Notes: Panels A and B of Table 2 report the baseline OLS regression results for the effect of conservatism (CON) on loan qual-
ity (LQ). The unit of observation is a bank quarter. All variables are winsorized at the 1st and 99th percentiles. LQ is either
nonperforming loans deflated by total assets (in Panel A), or net charge-offs deflated by total assets (in Panel B). The main
coefficient of interest is CON, which is expected to be negative and statistically significant. CON is one of the conservatism
measures C-Score, C-Score_EBP, C-Score_LLP, and LLP_ASY. Detailed calculation of these variables is provided in the Appendix.
TIMELINESS is the ratio of the reserve for credit losses deflated by the nonperforming loans. SIZE is the log of total assets. ABS-
GAP is the absolute value of the one-year maturity gap deflated by total assets. FEE is the ratio of fee income deflated by total
interest income. ROAVOL is the standard deviation of return on assets, calculated on a moving average basis over the preced-
ing 12 quarters. EFF is the ratio of non-interest expense deflated by total assets. EBLLP is earnings before provision deflated
by total assets. LOAN is loan deflated by total assets. LG is quarterly loan growth. COMM is agricultural and commercial loans
deflated by total assets. RE is real estate loans deflated by total assets. CONS is consumer loans deflated by total assets. CAP
is equity capital deflated by total assets. LIQUID is liquid assets deflated by total assets. ΔGDP is the quarterly change in GDP
obtained from the St. Louis Federal Reserve website. Detailed definitions of all variables are provided in the Appendix. All the
specifications include year-quarter dummy variables. Heteroskedasticity-robust t-statistics clustered by bank and quarter are
reported. Significance at the 10%, 5%, and 1% level are indicated by *, **, and ***, respectively.

coefficients on EBLLP, LOAN and LG are mostly positive and significant, indicating that profitable banks, banks with
higher loan to total asset ratios, and banks experiencing rapid growth in loans have poorer loan portfolio quality, con-
sistent with Foos et al. (2010). None of the coefficients on COMM, RE and CONS are significant. The coefficients on CAP
are negative but only significant when NCO is used as a dependent variable, indicating that bank capitalization is pos-
itively associated with loan portfolio quality, which is generally consistent with Bhat and Desai (2017). This indicates
that bank capital strengthens monitoring incentives, which in turn improves loan portfolio quality. The coefficients on
LIQUID are mostly negative but only significant when NPL is used as a dependent variable, indicating that liquid banks
have better loan portfolio quality. The results for the coefficient on the change in GDP are mixed. When NPL is used as
a dependent variable, the coefficients on ∆GDP are mostly insignificant except that it is positive and significant with
C-Score_LLP as a CON proxy. When NCO is used as a dependent variable, the coefficients on ∆GDP are positive and
HA 19

significant with C-Score and C-Score_EBP as a CON proxy, whereas the coefficient on ∆GDP is negative and significant
with C-Score_LLP as a CON proxy and insignificant with LLP_ASY as a CON proxy. A positive (negative) coefficient on
∆GDP implies increased (decreased) risk-taking during an economic upturn. Therefore, these coefficients on ∆GDP
suggest that banks are likely to experience increased NPLs but decreased NCOs during the economic boom.
For the remainder of the analyses, I employ only C-Score_LLP and LLP_ASY as independent variables because the
main results confirm that these two variables drive the results.

5.2 Conservatism and risk-taking incentive: Testing hypothesis H2

In hypotheses H2A and H2B, I predict that ex ante information asymmetry or distress risk can contribute to risk-taking
incentives and, thus, the disciplinary effect of conservatism on loan portfolio quality should be more pronounced for
banks that have greater risk-taking incentives. The variable RISKSHIFT is a proxy of either information asymmetry
or distress risk. Table 3, Panels A and B present the results of estimating equation (5) for each of the two informa-
tion asymmetry proxies, INFASY and JCD, respectively.12 Both H2A and H2B predict that the coefficient β3 on the
interaction between CON and RISKSHIFT should be negative. In Panel A, when INFASY is used as a proxy for informa-
tion asymmetry, the coefficient β3 on the interaction between CON and RISKSHIFT loads in the predicted direction
across both the CON measures, C-Score_LLP and LLP_ASY, and both the LQ measures, NPL and NCO, and it is significant
except for Column (1), where C-Score_LLP is used as a proxy for CON. In Panel B, where JCD is used as a proxy for infor-
mation asymmetry, the coefficient β3 loads negatively and significantly across both the CON measures C-Score_LLP
and LLP_ASY and both the LQ measures NPL and NCO, suggesting that conservatism makes a greater contribution to
improving loan quality when information asymmetry is high. Overall, these findings suggest that the monitoring bene-
fits of conservatism are greater for firms subject to larger information asymmetries, as predicted.
Hypothesis H2B predicts that a greater distress risk affects risk shifting incentives and, in turn, the disciplinary
effect of conservatism on loan portfolio quality. I examine this hypothesis in Table 4, Panels A and B, using two different
measures of distress risk, LOWCAP and POORPERF, respectively. In Panel A, when a low capital ratio (LOWCAP) is used
as a proxy for distress risk, the coefficient β3 on the interaction between CON and RISKSHIFT is negative and significant
for both LQ measures when C-Score_LLP is used as a proxy for CON, while the coefficient β3 is not significant when
LLP_ASY is used as a proxy for CON. In Panel B, when a low ROE (POORPERF) is used as a proxy for distress risk, the
coefficient β3 on the interaction between CON and RISKSHIFT loads negatively and significantly across both of the
CON measures C-Score_LLP and LLP_ASY and both of the LQ measures NPL and NCO, suggesting that the contribution
of conservatism to improving loan portfolio quality is greater when distress risk is high. In both Table 3 and Table 4, the
main effect of CON on improving loan portfolio quality is still present.
The results for H2B can directly address the possibility that the association between CON and loan quality could
also be due in part to a causal link from loan portfolio quality to conservative accounting—for example, through the
effect discussed by Liu and Ryan (1995), who point out that loan loss provisions can signal the good news that the bank
is in a sound enough financial position to absorb a hit to its earnings. Similarly, Vyas (2011) proposes that ‘signaling’
can be an alternative explanation for why some firms are timelier in write-downs. If conservatism is a signal of banks’
financial health, banks in distress will report less conservatively. Assuming the loan portfolio represents the financial
health of banks, this argument implies a positive coefficient of the interaction term between CON and distress risk
(because the dependent variable is a negative function of loan quality). However, the results are the opposite, with
the coefficients for the interaction term between CON and RISKSHIFT showing negative signs for both LOWCAP and
POORPERF as proxies for distress risk. Therefore, the results appear not to be attributable to the signaling explanation.

12 The coefficients on the interaction variables between RISKSHIFT and the control variables are estimated but not reported, for the sake of brevity, in Tables

3 and 4.
20 HA

TA B L E 3 The conditional effect of information asymmetry on the relation between conservatism and loan quality
∗
LQit+ 4 = 𝛽0 + 𝛽1 CONit + 𝛽2 RISKSHIFTit + 𝛽3 (CONit RISKSHIFTit ) + 𝛽n Control Variablesit
∗
+ RISKSHIFTit (𝜇n Control Variablesit ) + Year − Quarter dummies + 𝜀it (5)

Panel A: Information asymmetry measured by the market-based proxies

Dep Var = NPL at t+4 CON = C-Score_LLP
CON = C-Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
RISKSHIFT = INFASY
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.004 −1.06 −0.011 −6.65 ***
−0.002 −0.39 −0.010 −6.16***
CON −0.001 −5.28 ***
−0.001 −7.60 ***
−0.002 −11.15 ***
−0.001 −13.58***
RISKSHIFT 0.000 0.15 −0.003 −3.37*** 0.001 0.45 −0.003 −3.28***
CON × RISKSHIFT 0.000 −0.45 0.000 −1.96* −0.001 −2.94*** 0.000 −3.72***
*** *** ***
LQ 0.718 34.56 0.578 24.05 0.684 30.48 0.576 24.75***
TIMELINESS 0.000 0.55 0.000 −3.48 ***
0.000 0.60 0.000 −3.91***
*** ***
SIZE 0.000 0.61 0.000 4.10 0.000 2.72 0.000 6.65***
ABSGAP 0.002 4.17*** 0.000 2.13** 0.001 3.66*** 0.000 2.15**
FEE 0.000 0.09 −0.001 −2.65 ***
0.000 −0.44 −0.001 −3.43***
*** *** ***
ROAVOL 0.258 3.95 0.083 3.67 0.259 3.96 0.098 4.25***
EFF −0.040 −3.85 ***
0.007 1.75 *
−0.036 −3.33 ***
0.011 2.75***
EBLLP 0.063 2.03** 0.067 5.37*** 0.053 1.64 0.056 4.66***
**
LOAN 0.001 0.67 0.001 2.10 0.000 0.12 0.002 2.32**
** *** **
LG 0.000 2.50 0.000 4.32 0.000 2.10 0.000 3.95***
COMM 0.000 0.16 0.000 0.26 0.001 0.46 0.000 −0.30
RE 0.003 2.02** 0.000 −0.03 0.004 2.37** 0.000 −0.28
CONS 0.000 0.03 0.000 0.34 0.001 0.54 0.000 0.18
CAP 0.004 0.89 −0.003 −1.95* 0.000 0.10 −0.004 −2.95***
LIQUID −0.002 −1.41 0.000 −0.31 −0.003 −1.93* 0.000 −0.24
ΔGDP 0.044 0.27 0.356 5.40*** −0.095 −0.61 0.290 4.57***
Control variables × Yes Yes Yes Yes
RISKSHIFT
Year−quarter dummies Yes Yes Yes Yes
Adj. R square 79.48% 67.92% 78.98% 68.51%
No. obs. 10,714 10,714 10,588 10,588
Panel B: Information asymmetry measured by JCD rate
Dep Var = NPL at t+4 Dep Var = NCO at t+4
CON = C−Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
RISKSHIFT = JDC
(1) (2) (3) (4)
Coeff t−stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept 0.001 0.66 0.002 2.71*** −0.012 −1.81* 0.003 0.88
CON −0.001 −3.04 ***
0.000 −2.91 ***
−0.002 −5.43 ***
0.000 −4.76***
RISKSHIFT −0.011 −4.54 ***
−0.005 −5.00 ***
−0.013 −4.03 ***
−0.005 −4.10***
CON × RISKSHIFT −0.001 −2.47 **
−0.001 −2.20 **
−0.002 −4.48 ***
−0.001 −7.26***

(Continues)
HA 21

TA B L E 3 (Continued)

Panel B: Information asymmetry measured by JCD rate

Dep Var = NPL at t+4 Dep Var = NCO at t+4
CON = C−Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
RISKSHIFT = JDC
(1) (2) (3) (4)
Coeff t−stat Coeff t-stat Coeff t-stat Coeff t-stat
LQ 0.720 35.46*** 0.518 17.64*** 0.694 29.50*** 0.506 17.31***
TIMELINESS 0.000 1.11 0.000 0.72 0.000 1.52 0.000 0.72
SIZE 0.000 −1.78* 0.000 −1.30 0.000 −0.28 0.000 −0.04
ABSGAP 0.001 3.23*** 0.000 1.35 0.001 2.35** 0.000 1.14
FEE 0.001 0.64 0.000 −0.99 0.000 0.14 −0.001 −1.18
ROAVOL 0.052 0.49 0.088 2.22** 0.047 0.41 0.082 2.29**
EFF −0.052 −3.21*** 0.004 0.51 −0.045 −2.55** 0.009 1.30
EBLLP 0.120 2.47** 0.054 2.30** 0.106 1.93* 0.041 1.82*
LOAN 0.003 1.20 0.000 0.15 0.002 0.68 0.001 0.59
LG 0.000 0.01 0.000 1.82* 0.000 0.50 0.000 1.53
COMM −0.007 −2.49** −0.001 −0.81 −0.006 −1.88* 0.000 −0.27
RE −0.002 −0.79 0.000 −0.24 −0.001 −0.30 0.000 0.23
CONS −0.004 −1.02 0.000 0.10 −0.002 −0.57 0.001 0.50
CAP 0.002 0.32 −0.006 −3.30*** −0.001 −0.08 −0.007 −3.29***
LIQUID −0.004 −2.97*** −0.001 −2.57** −0.005 −2.42** 0.000 −0.95
ΔGDP 0.139 7.23*** −0.020 −1.83* 0.679 2.33** −0.128 −0.93
Control variables × Yes Yes Yes Yes
RISKSHIFT
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 76.02% 62.59% 76.59% 63.06%
No. obs 16,763 16,763 15,496 15,496

Notes: Panels A and B of Table 3 report the OLS regression results for testing hypothesis H2A, whether the effect of conser-
vatism on loan quality is greater if there is greater information asymmetry. LQ is either nonperforming loans deflated by total
assets (in Columns (1) and (2) in both Panel A and B) or net charge-offs deflated by total assets (in Columns (3) and (4) in both
Panel A and B). CON is one of the conservatism measures C-Score_LLP, and LLP_ASY. Detailed calculation of these variables is
provided in the Appendix. RISKSHIFT is a proxy of information asymmetry. Information asymmetries are measured in two ways.
The first measure used in Panel A is a market-based measure, which is the average of the standardized bid–ask spread, stock
return volatility, and idiosyncratic risk (INFASY). The second measure used in Panel B is information asymmetries between the
firm and debtholder as reflected in the imputed interest rates for large time deposits (jumbo certificates of deposits, or JCD).
The main coefficient of interest is CON × RISKSHIFT, which is expected to be negative and statistically significant. TIMELINESS
is the ratio of the reserve for credit losses deflated by the nonperforming loans. SIZE is the log of total assets. ABSGAP is the
absolute value of the one-year maturity gap deflated by total assets. FEE is the ratio of fee income deflated by total interest
income. ROAVOL is the standard deviation of return on assets, calculated on a moving average basis over the preceding 12
quarters. EFF is the ratio of non-interest expense deflated by total assets. EBLLP is earnings before provision deflated by total
assets. LOAN is loans deflated by total assets. LG is quarterly loan growth. COMM is agricultural and commercial loans deflated
by total assets. RE is real estate loans deflated by total assets. CONS is consumer loans deflated by total assets. CAP is equity
capital deflated by total assets. LIQUID is liquid assets deflated by total assets. ΔGDP is the quarterly change in GDP
obtained from the St. Louis Federal Reserve website. I also use a fully interacted model by including interaction variables
for RISKSHIFT and the control variables to allow the coefficients on control variables to vary with RISKSHIFT (untabulated).
Detailed definitions of all variables are provided in the Appendix. All the specifications include year-quarter dummy variables.
Heteroskedasticity-robust t-statistics clustered by bank and quarter are reported. Significance at the 10%, 5%, and 1% level
are indicated by * , ** , and *** , respectively.
22 HA

TA B L E 4 The conditional effect of distress risk on the relation between conservatism and loan quality
∗
LQit+4 = 𝛽0 + 𝛽1 CONit + 𝛽2 RISKSHIFTit + 𝛽3 (CONit RISKSHIFTit ) + 𝛽n Control Variablesit
∗
+ RISKSHIFTit (𝜇n Control Variablesit ) + Year − quarter dummies + 𝜀it (5)

Panel A: Distress risk measured by capital ratio (CAP)

Dep Var = NPL at t+4 Dep Var = NCO at t+4
CON = C-Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
RISKSHIFT = LOWCAP
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.005 −3.49*** 0.000 −0.01 −0.001 −2.86*** −0.005 −3.87***
CON −0.002 −6.70 ***
−0.003 −13.74 ***
−0.001 −8.07 ***
−0.001 −16.20***
RISKSHIFT −0.003 −0.87 0.000 0.14 0.000 −0.34 −0.001 −0.94
CON × RISKSHIFT −0.001 −1.94* 0.000 −0.96 −0.001 −2.33** 0.000 0.78
LQ 0.682 38.17*** 0.668 37.79*** 0.580 33.42*** 0.571 34.41***
TIMELINESS 0.000 0.43 0.000 1.33 0.000 −3.38*** 0.000 −1.03
***
SIZE 0.000 0.24 0.000 2.76 0.000 3.07*** 0.000 6.56***
*** *** ***
ABSGAP 0.002 4.62 0.002 4.12 0.000 3.98 0.000 3.88***
FEE −0.001 −0.65 −0.002 −1.75* −0.001 −2.43** −0.001 −4.35***
ROAVOL 0.315 4.73*** 0.309 4.45*** 0.130 5.73*** 0.137 5.98***
EFF −0.032 −3.17 ***
−0.023 −2.21 **
0.005 1.20 0.012 3.38***
*** *** ***
EBLLP 0.106 3.51 0.083 2.62 0.062 5.23 0.043 3.83***
LOAN 0.002 0.82 0.005 2.12** 0.002 3.07*** 0.003 3.76***
LG 0.000 4.20*** 0.000 3.88*** 0.000 4.22*** 0.000 3.55***
COMM 0.000 −0.15 −0.002 −0.91 −0.001 −1.06 −0.001 −1.62
RE 0.002 1.03 0.000 0.05 −0.001 −1.46 −0.001 −1.99**
CONS 0.000 0.20 −0.002 −1.03 0.000 0.21 0.000 0.12
LIQUID −0.003 −2.59*** −0.002 −1.81* 0.000 −0.77 0.000 −0.54
ΔGDP 0.139 8.12*** −0.179 −1.01 −0.033 −3.49*** 0.084 1.51
Control variables × Yes Yes Yes Yes
RISKSHIFT
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 75.47% 76.14% 61.96% 62.32%
No. obs. 18,158 17,049 18,158 17,049
Panel B: Distress risk measured by return on equity (ROE)
Dep Var = NPL at t+4 Dep Var = NCO at t+4
CON = C-Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
RISKSHIFT = POORPERF
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.005 −2.18** 0.001 0.17 −0.003 −3.28*** 0.000 −0.22
CON −0.002 −5.09*** −0.002 −12.78*** −0.001 −6.81*** −0.002 −12.46***
RISKSHIFT 0.003 1.18 0.003 0.60 0.000 0.17 −0.004 −3.02***
CON × RISKSHIFT −0.001 −1.81* −0.002 −4.96*** −0.001 −3.30*** −0.001 −5.93***
*** *** ***
LQ 0.658 34.99 0.670 31.36 0.454 18.05 0.273 12.25***
TIMELINESS 0.028 0.86 0.000 1.95 *
0.006 0.14 0.000 −3.32***
SIZE 0.000 0.60 0.000 3.14*** 0.000 1.84* 0.000 5.44***
ABSGAP 0.001 0.82 0.001 3.57*** 0.000 1.46 0.001 2.47**

(Continues)
HA 23

TA B L E 4 (Continued)

Panel B: Distress risk measured by return on equity (ROE)

Dep Var = NPL at t+4 Dep Var = NCO at t+4
CON = C-Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
RISKSHIFT = POORPERF
(1) (2) (3) (4)
FEE 0.000 0.11 −0.002 −2.34** −0.002 −2.22** 0.002 2.99***
ROAVOL −0.109 −0.75 0.337 5.34 ***
0.281 5.11 ***
0.425 9.47***
EFF −0.012 −0.55 −0.015 −1.47 0.055 6.11 ***
−0.083 −13.07***
EBLLP 0.006 0.11 0.089 2.91 ***
−0.016 −0.63 0.070 3.46***
LOAN 0.000 −0.03 0.004 1.85* −0.003 −1.72* 0.007 4.53***
LG 0.000 −1.35 0.000 3.88 ***
0.000 0.30 0.000 0.86
COMM −0.006 −1.19 −0.002 −0.98 −0.001 −0.34 −0.003 −1.69*
RE 0.001 0.15 0.000 −0.17 0.003 1.69* −0.003 −1.80*
CONS 0.003 0.58 −0.002 −0.98 0.000 −0.10 0.002 1.05
CAP −0.008 −0.87 −0.003 −0.82 −0.003 −1.10 −0.006 −2.75***
LIQUID −0.002 −1.23 −0.002 −1.99 **
−0.001 −1.45 0.001 1.04
ΔGDP −0.035 −2.67*** −0.195 −1.07 −0.004 −0.76 −0.236 −5.50***
Control variables × Yes Yes Yes Yes
RISKSHIFT
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 75.49% 76.33% 63.45% 63.10%
No. obs. 18,158 17,049 18,158 17,049

Notes: Panels A and B of Table 4 report the OLS regression results for testing hypothesis H2B, whether the effect of conser-
vatism on loan quality is greater if there is greater distress risk. LQ is either nonperforming loans deflated by total assets (in
Columns (1) and (2) in both Panel A and B), or net charge-offs deflated by total assets (in Columns (3) and (4) in both Panel
A and B). CON is one of the conservatism measures C-Score_LLP, and LLP_ASY. Detailed calculation of these variables is pro-
vided in the Appendix. RISKSHIFT is a proxy of distress risk. Distress risk is measured by two dummy variables, a low balance
sheet capital ratio (LOWCAP) and a low ROE (POORPERF). LOWCAP is an indicator variable that gets one if a bank’s level of
balance sheet capital deflated by total assets is less than 7%, and zero otherwise. POORPERF is an indicator variable that gets
one if a bank’s ROE is in the bottom quintile each quarter, and zero otherwise. ROE is calculated by earnings before provision
deflated by equity capital. The main coefficient of interest is CON × RISKSHIFT, which is expected to be negative and statisti-
cally significant. TIMELINESS is the ratio of the reserve for credit losses deflated by the nonperforming loans. SIZE is the log of
total assets. ABSGAP is the absolute value of the one-year maturity gap deflated by total assets. FEE is the ratio of fee income
deflated by total interest income. ROAVOL is the standard deviation of return on assets, calculated on a moving average basis
over the preceding 12 quarters. EFF is the ratio of non-interest expense deflated by total assets. EBLLP is earnings before pro-
vision deflated by total assets. LOAN is loans deflated by total assets. LG is quarterly loan growth. COMM is agricultural and
commercial loans deflated by total assets. RE is real estate loans deflated by total assets. CONS is consumer loans deflated by
total assets. CAP is equity capital deflated by total assets. LIQUID is liquid assets deflated by total assets. ΔGDP is the quarterly
change in GDP obtained from the St. Louis Federal Reserve website. I also use a fully interacted model by including interaction
variables for RISKSHIFT and the control variables to allow the coefficients on control variables to vary with RISKSHIFT (untab-
ulated). Detailed definitions of all variables are provided in the Appendix. All the specifications include year-quarter dummy
variables. Heteroskedasticity-robust t-statistics clustered by bank and quarter are reported. Significance at the 10%, 5%, and
1% level are indicated by * , ** , and *** , respectively.

Overall, these findings suggest that the monitoring benefits of conservatism are incrementally greater for banks
subject to larger information asymmetry, as these banks are likely to take more risk. In addition to being of interest in
themselves, the interaction tests also provide additional credence to H1 by alleviating correlated omitted variables or
reinforcing causality, thereby demonstrating that the effect is more pronounced in subsamples in which the effect is
expected to be stronger.
24 HA

5.3 Conservatism and bank lending growth cycle: Testing hypothesis H3

The results for hypothesis H3 are reported in Table 5. Hypothesis H3 states that the monitoring benefit of conser-
vatism in facilitating a bank’s own monitoring effort is more pronounced during extreme lending growth cycles, when
banks will likely have lower-quality loan portfolios than during normal times. H3 predicts that both the coefficients β2
on the interaction between CON and LOW and β3 on the interaction between CON and HIGH should be negative. The
coefficient on CON is negative and significant across all columns, indicating that conservatism is positively associated
with loan quality during moderate lending growth cycles. Both the coefficients on CON * LOW and CON * HIGH are
negative and significant across both the CON measures C-Score_LLP and LLP_ASY and both the LQ measures NPL and
NCO. The results imply that the benefits of conservatism for improving loan quality are more pronounced in both low-
and high-credit growth cycles compared to moderate credit growth cycles, consistent with H3.

6 ADDITIONAL ANALYSES

In this section, I report results for additional tests that lend robustness and extend the reported results.

6.1 Corporate governance subsample analysis

Leventis et al. (2013) argue that effective governance structures result in better monitoring of management and
hence will favor implementation of conservative accounting, because greater accounting conservatism will reduce
the likelihood of managers’ manipulations and enhance the quality of accounting information. Consistent with this
view, Leventis et al. (2013) find that well-governed banks engage in significantly higher levels of conservatism in
their financial reporting practices. If greater conservatism is associated with strong corporate governance, then one
might wonder whether the present study’s results are driven by corporate governance and not conservatism per
se. Therefore, I examine whether the improved loan portfolio quality associated with conservatism differs between
well-governed and poorly governed firms. If corporate governance drives the results, then the results should be
stronger for well-governed banks.
I partition the sample into weak and strong governance subsamples and estimate equation (4) for each subsample.
I use Gompers, Ishii, and Metrick’s (2003) G-score as a summary (inverse) measure of the strength of firms’ corpo-
rate governance. Firms whose G-score equals or exceeds the median for the full sample are classified as having weak
governance, and the remainder are included in the strong governance subsample.
The results for the corporate governance subsamples are shown in Panels A and B of Table 6. The coefficient on
CON loads negatively and significantly in both the weak and strong corporate governance subsamples across both of
the CON measures C-Score_LLP (in Panel A) and LLP_ASY (in Panel B) and both of the LQ measures NPL and NCO. This
confirms that the results documented in this paper are attributable to the benefits of conservatism and not corporate
governance.

6.2 The effect of loan portfolio composition

The disciplinary effect of conservatism on loan quality should partly depend on the extent to which managers can
exercise their discretion in establishing loan loss provisions. For example, consumer loans are relatively homogeneous
loans, and there is less room for bank management to apply conservatism when establishing allowances for those
loans. For these loans, the connection between conservatism and loan portfolio quality could be weakened. To address
HA 25

TA B L E 5 Analysis of the relation between conservatism and loan quality depending on lending growth cycles
∗ ∗
LQit+4 = 𝛽0 + 𝛽1 CONit + 𝛽2 (CONt LOWt ) + 𝛽3 (CONt HIGHt ) + 𝛽4 LOWt + 𝛽5 HIGHt + 𝛽n Control Variablesit
+ Year − quarter dummies + 𝜀it (6)

Dep Var = NPL at t+4 Dep Var = NCO at t+4

CON = C-Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.006 −4.67*** −0.001 −2.76*** 0.002 0.46 −0.005 −3.27***
CON −0.001 −2.39** 0.000 −3.30*** −0.002 −7.54*** −0.001 −8.95***
CON × LOW 0.000 −2.02 **
0.000 −2.89 ***
0.000 −3.53 ***
0.000 −2.74***
CON × HIGH −0.004 −6.99 ***
−0.001 −5.67 ***
−0.005 −9.92 ***
−0.001 −9.89***
LOW 0.014 16.68 ***
0.001 1.15 −0.005 −0.55 0.005 1.91*
HIGH 0.017 26.28*** 0.003 8.45*** 0.006 1.24 0.005 3.44***
*** *** ***
LQ 0.667 41.10 0.583 39.77 0.665 39.03 0.571 38.57***
TIMELINESS 0.000 0.08 0.000 −3.66 ***
0.000 1.26 0.000 −0.84
SIZE 0.000 1.20 0.000 4.14*** 0.000 2.91*** 0.000 7.44***
ABSGAP 0.002 4.46*** 0.000 3.85*** 0.001 3.87*** 0.000 3.26***
FEE 0.000 −0.56 −0.001 −3.29 ***
−0.001 −1.57 −0.001 −4.89***
*** *** ***
ROAVOL 0.306 5.12 0.140 6.96 0.290 4.69 0.140 7.21***
EFF −0.027 −3.11 ***
0.006 1.75 *
−0.023 −2.52 **
0.012 3.63***
EBLLP 0.080 3.04*** 0.058 5.23*** 0.067 2.38** 0.043 4.05***
LOAN 0.001 0.77 0.002 2.62*** 0.004 1.91* 0.002 3.38***
*** *** ***
LG 0.000 3.96 0.000 5.04 0.000 3.63 0.000 4.36***
COMM 0.000 −0.04 −0.001 −0.77 −0.003 −1.55 −0.001 −1.5
RE 0.002 1.16 −0.001 −1.16 0.000 0.03 −0.001 −1.48
CONS 0.000 0.14 0.000 0.03 −0.002 −1.24 0.000 0.63
CAP −0.003 −0.72 −0.004 −3.67 ***
−0.004 −0.95 −0.005 −4.57***
LIQUID −0.003 −2.62 ***
0.000 −1.21 −0.003 −2.3 **
0.000 −0.85
ΔGDP 0.139 11.19*** −0.034 −3.72*** −0.233 −1.35 0.065 1.19
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 75.53% 61.95% 76.46% 62.64%
No. obs. 18,158 17,049 18,158 17,049

Notes: Table 5 reports the OLS regression results for testing hypothesis H3, whether the effect of conservatism on loan quality
depends on the lending growth cycles. LQ is either nonperforming loans deflated by total assets (in Columns (1) and (2)), or
net charge-offs deflated by total assets (in Columns (3) and (4)). CON is one of the conservatism measures C-Score_LLP, and
LLP_ASY. Detailed calculation of these variables is provided in the Appendix. The main coefficients of interest are CON × LOW
and CON × HIGH, which are expected to be negative and statistically significant. HIGH and LOW are indicator variables that
correspond to high and low lending growth cycles, respectively. The high lending growth cycle variable gets the value one for
the years 2000, 2005–2008, and 2011–2015, and zero otherwise. The low lending growth cycle variable is set to one for the
years 2001–2004 and 2009–2010, and zero otherwise. The moderate lending growth cycle is for the years 1994–1999, but it
is excluded from the model in order to be used as a comparison group. TIMELINESS is the ratio of the reserve for credit losses
deflated by the nonperforming loans. SIZE is the log of total assets. ABSGAP is the absolute value of the one-year maturity gap
deflated by total assets. FEE is the ratio of fee income deflated by total interest income. ROAVOL is the standard deviation of
return on assets, calculated on a moving average basis over the preceding 12 quarters. EFF is the ratio of non-interest expense
deflated by total assets. EBLLP is earnings before provision deflated by total assets. LOAN is loans deflated by total assets. LG
is quarterly loan growth. COMM is agricultural and commercial loans deflated by total assets. RE is real estate loans deflated
by total assets. CONS is consumer loans deflated by total assets. CAP is equity capital deflated by total assets. LIQUID is liquid
assets deflated by total assets. ΔGDP is the quarterly change in GDP obtained from the St. Louis Federal Reserve website. All
the specifications include year-quarter dummy variables. Heteroskedasticity-robust t-statistics clustered by bank and quarter
are reported. Significance at the 10%, 5%, and 1% level are indicated by * , ** , and *** , respectively.
26 HA

TA B L E 6 Corporate governance subsample analysis

LQit+4 = 𝛽0 + 𝛽1 CONit + 𝛽2 LQit + 𝛽3 TIMELINESS + 𝛽4 SIZEit + 𝛽5 ABSGAPit + 𝛽6 FEEit + 𝛽7 ROAVOLit + 𝛽8 EFFit
+ 𝛽9 EBLLPit +10 LOANit + 𝛽11 LGit + 𝛽12 COMMit + 𝛽13 REit 𝛽14 CONSit + 𝛽15 CAPit + 𝛽16 LIQUIDit
+ 𝛽17 ΔGDPit + Year − quarter dummies + 𝜀it (4)

Panel A: Using C-Score_LLP as a conservatism measure (CON)

Dep var = NPL at t+4 Dep Var = NCO at t+4
CON = C-Score_LLP
Strong Governance Poor Governance Strong Governance Poor Governance
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.005 −1.99* −0.006 −3.34*** 0.000 0.07 −0.001 −1.16
CON −0.001 −1.94* −0.001 −3.73*** 0.000 −2.63** −0.001 −3.62***
*** *** ***
LQ 0.678 12.67 0.584 14.13 0.633 21.20 0.583 21.21***
TIMELINESS 0.000 −0.89 0.000 −0.28 0.000 −0.83 0.000 0.10
SIZE 0.000 1.20 0.000 2.28** 0.000 0.91 0.000 1.66
ABSGAP 0.000 −0.72 0.001 1.83* 0.000 0.21 0.001 2.67***
FEE 0.001 0.32 0.002 1.44 0.000 −0.51 −0.001 −0.99
ROAVOL 0.429 2.82*** 0.127 1.04 0.081 1.66 0.086 2.04**
EFF −0.038 −2.30 **
−0.035 −1.77 *
−0.009 −1.24 −0.003 −0.42
EBLLP 0.107 2.07** 0.078 1.32 0.094 3.58*** 0.072 3.29***
LOAN 0.005 1.04 −0.007 −1.84* 0.003 1.61 0.004 2.67***
LG 0.000 1.03 0.000 1.23 0.000 1.17 0.000 3.62***
COMM −0.002 −0.28 0.009 1.94 *
−0.001 −0.70 −0.003 −2.24**
RE −0.001 −0.15 0.011 3.09*** −0.002 −1.20 −0.002 −1.89*
CONS 0.000 −0.10 0.009 1.90* −0.002 −0.96 −0.001 −0.63
CAP −0.024 −2.36 **
−0.004 −0.50 −0.006 −1.98 *
−0.008 −2.75***
LIQUID −0.002 −1.33 −0.003 −2.20 **
0.000 −0.81 0.000 −0.31
ΔGDP 0.150 5.84*** 0.107 7.77*** −0.037 −1.93* −0.043 −3.51***
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 76.95% 78.05% 68.31% 69.99%
No. obs. 3,594 3,593 3,594 3,593
Panel B: Using LLP_ASY as a conservatism measure (CON)
Dep var = NPL at t+4 Dep Var = NCO at t+4
CON = LLP_ASY
Strong Governance Poor Governance Strong Governance Poor Governance
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.010 −1.62 −0.010 −1.54 −0.011 −2.73*** −0.006 −2.50**
CON −0.003 −7.67*** −0.003 −7.03*** −0.001 −8.80*** −0.001 −7.93***
*** *** ***
LQ 0.708 14.29 0.633 15.31 0.624 14.90 0.638 23.92***
TIMELINESS 0.000 −0.56 0.000 0.27 0.000 −0.49 0.000 0.39
SIZE 0.000 3.12*** 0.000 3.14*** 0.000 3.08*** 0.000 3.09***
ABSGAP 0.000 0.33 0.001 1.21 0.000 0.94 0.001 2.37**
FEE 0.000 −0.30 0.001 0.50 −0.001 −2.01 **
−0.001 −1.78*
***
ROAVOL 0.422 2.72 0.115 0.92 0.100 1.62 0.105 2.54**
EFF −0.026 −1.74* −0.020 −0.93 0.001 0.26 0.006 1.00

(Continues)
HA 27

TA B L E 6 (Continued)

Panel B: Using LLP_ASY as a conservatism measure (CON)

Dep var = NPL at t+4 Dep Var = NCO at t+4
CON = LLP_ASY
Strong Governance Poor Governance Strong Governance Poor Governance
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
** ***
EBLLP 0.102 2.11 0.051 0.80 0.075 2.99 0.051 3.08***
LOAN 0.007 1.58 −0.002 −0.44 0.003 1.72 *
0.003 2.43**
LG 0.000 0.54 0.000 1.65 0.000 0.91 0.000 3.72***
COMM −0.005 −1.04 0.002 0.51 −0.002 −1.19 −0.003 −2.31**
RE −0.003 −0.72 0.006 1.88 *
−0.003 −1.45 −0.002 −1.77*
CONS −0.003 −0.68 0.002 0.55 −0.002 −0.89 0.001 0.82
CAP −0.019 −1.98* −0.005 −0.58 −0.007 −2.36** −0.007 −2.71***
LIQUID −0.001 −0.65 −0.003 −1.52 0.000 −0.48 0.000 −0.26
ΔGDP 0.189 0.84 0.180 0.78 0.382 2.42 **
0.148 1.72*
Year−quarter dummies Yes Yes Yes Yes
Adj. R square 82.37% 80.88% 69.75% 73.90%
No. obs. 3,029 3,116 3,029 3,116

Notes: Panels A and B of Table 6 report the OLS regression results for the effect of conservatism on loan quality depending
on corporate governance. I divide the sample into weak and strong governance subsamples and run the regression analysis
separately on these two subsamples. I use Gompers et al.’s (2003) G-score as a summary (inverse) measure of the strength
of firms’ corporate governance. Firms whose G-score equals or exceeds the median for the full sample are classified as ‘weak
governance,’ and the remainder are placed in the ‘strong governance’ subsample. LQ is either nonperforming loans deflated by
total assets (in Columns (1) and (2) in both Panel A and B), or net charge-offs deflated by total assets (in Columns (3) and (4)
in both Panel A and B). CON is one of the conservatism measures C-Score_LLP (used in Panel A), and LLP_ASY (used in Panel B).
Detailed calculation of these variables is provided in the Appendix. The main coefficient of interest is CON, which is expected
to be negative and statistically significant for both strong and poor corporate governance subsamples. TIMELINESS is the ratio
of the reserve for credit losses deflated by the nonperforming loans. SIZE is the log of total assets. ABSGAP is the absolute value
of the one-year maturity gap deflated by total assets. FEE is the ratio of fee income deflated by total interest income. ROAVOL
is the standard deviation of return on assets, calculated on a moving average basis over the preceding 12 quarters. EFF is the
ratio of non-interest expense deflated by total assets. EBLLP is earnings before provision deflated by total assets. LOAN is loans
deflated by total assets. LG is quarterly loan growth. COMM is agricultural and commercial loans deflated by total assets. RE
is real estate loans deflated by total assets. CONS is consumer loans deflated by total assets. CAP is equity capital deflated by
total assets. LIQUID is liquid assets deflated by total assets. ΔGDP is the quarterly change in GDP obtained from the St. Louis
Federal Reserve website. All the specifications include year-quarter dummy variables. Heteroskedasticity-robust t-statistics
clustered by bank and quarter are reported. Significance at the 10%, 5%, and 1% level are indicated by * , ** , and *** , respectively.

the possibility that some banks can exercise discretion to a lesser extent, I test whether the link between conservatism
and loan quality is weaker for banks with a larger proportion of consumer loans among the total loans. I create a dummy
variable, HOMOGENOUS, which gets the value 1 if the ratio of consumer loans to total loans is in the top tercile, and 0
if this ratio falls in the bottom two terciles. I modify equation (4) by including HOMOGENOUS and its interaction with
CON.13 The results reported in Table 7 show that CON*HOMOGENEOUS is positive and significant with the exception
of Column (2), where LLP_ASY is used as a CON proxy and NPL as a dependent variable, suggesting that the effect of
conservatism on improving loan quality is lower for banks that have a more homogenous loan portfolio. This is consis-
tent with the expectation that conservatism improves monitoring over loan portfolios through managerial discretion
over provisioning decisions.

13 The coefficients on the interaction variables between HOMOGENEOUS and the control variables are estimated but not reported, for the sake of brevity, in

Table 7.
28 HA

TA B L E 7 The relation between conservatism and loan quality depending on loan portfolio composition
∗
LQit+ 4 = 𝛽0 + 𝛽1 CONit + 𝛽2 HOMOGENEOUSit + 𝛽3 (CONit HOMOGENEOUSit ) + 𝛽n Control Variablesit
∗
+ HOMOGENEOUSit (𝜇n Control Variablesit ) + Year − quarterdummies + 𝜀it (modified Eq.4)

Dep Var = NPL at t+4 Dep Var = NCO at t+4

CON = C-Score_LLP CON = LLP_ASY CON = C-Score_LLP CON = LLP_ASY
(1) (2) (3) (4)
Coeff t-stat Coeff t-stat Coeff t-stat Coeff t-stat
Intercept −0.005 −2.79 ***
0.003 0.57 −0.001 −1.04 −0.005 −3.10***
CON −0.002 −7.57 ***
−0.003 −12.56 ***
−0.001 −8.45 ***
−0.001 −15.97***
HOMOGENEOUS 0.002 1.04 0.000 −0.03 −0.001 −0.90 0.000 0.15
CON × 0.001 2.82*** 0.000 1.13 0.000 2.32** 0.000 1.92*
HOMOGENEOUS
LQ 0.681 41.03*** 0.673 37.30*** 0.564 35.05*** 0.554 35.32***
TIMELINESS 0.000 0.92 0.000 1.80* 0.000 −3.45*** 0.000 −1.01
SIZE 0.000 0.46 0.000 2.52** 0.000 3.40*** 0.000 6.53***
*** *** ***
ABSGAP 0.002 4.21 0.002 3.95 0.000 3.55 0.000 3.04***
FEE −0.001 −0.60 −0.002 −1.98** −0.001 −3.24*** −0.001 −4.66***
ROAVOL 0.350 4.83*** 0.323 4.39*** 0.157 6.91*** 0.152 6.85***
EFF −0.021 −2.04 **
−0.016 −1.49 0.007 1.69 *
0.014 3.50***
* ***
EBLLP 0.060 1.91 0.051 1.57 0.061 4.69 0.041 3.30***
**
LOAN 0.001 0.55 0.003 1.09 0.002 2.04 0.002 2.27**
LG 0.000 3.70*** 0.000 3.50*** 0.000 4.72*** 0.000 4.57***
COMM −0.002 −0.67 −0.004 −1.59 −0.001 −1.34 −0.001 −1.01
RE 0.002 1.01 0.000 −0.06 −0.001 −1.45 −0.001 −0.93
CONS 0.017 3.88*** 0.010 1.95* 0.001 0.46 0.000 −0.27
CAP −0.001 −0.23 −0.005 −0.93 −0.004 −3.01*** −0.005 −3.47***
LIQUID −0.003 −2.31 **
−0.004 −2.39 **
−0.001 −2.16 **
0.000 −0.92
ΔGDP 0.133 8.41*** −0.198 −1.11 −0.042 −4.28*** 0.064 1.16
Control variables × Yes Yes Yes Yes
HOMOGENEOUS
Year-quarter dummies Yes Yes Yes Yes
Adj. R square 75.58% 76.21% 61.94% 62.50%
No. obs. 18,158 17,049 18,158 17,049

Notes: Table 7 reports the OLS regression results for the effect of conservatism on loan quality depending on bank loan portfo-
lio composition. LQ is either nonperforming loans deflated by total assets (in Columns (1) and (2)), or net charge-offs deflated
by total assets (in Columns (3) and (4)). CON is one of the conservatism measures C-Score_LLP (used in Columns (1) and (3)),
and LLP_ASY (used in Columns (2) and (4)). Detailed calculation of these variables is provided in the Appendix. A dummy vari-
able, HOMOGENOUS, gets the value one if the ratio of consumer loans to total loans is in the top tercile, and zero if this ratio
is in the bottom two terciles. The main coefficient of interest is CON × HOMOGENEOUS, which is expected to be positive and
statistically significant. TIMELINESS is the ratio of the reserve for credit losses deflated by the nonperforming loans. SIZE is the
log of total assets. ABSGAP is the absolute value of the one-year maturity gap deflated by total assets. FEE is the ratio of fee
income deflated by total interest income. ROAVOL is the standard deviation of return on assets, calculated on a moving aver-
age basis over the preceding 12 quarters. EFF is the ratio of non-interest expense deflated by total assets. EBLLP is earnings
before provision deflated by total assets. LOAN is loans deflated by total assets. LG is quarterly loan growth. COMM is agricul-
tural and commercial loans deflated by total assets. RE is real estate loans deflated by total assets. CONS is consumer loans
deflated by total assets. CAP is equity capital deflated by total assets. LIQUID is liquid assets deflated by total assets. ΔGDP is
the quarterly change in GDP obtained from the St. Louis Federal Reserve website. I also use a fully interacted model by includ-
ing interaction variables for HOMOGENOUS and the control variables to allow the coefficients on control variables to vary with
HOMOGENOUS (untabulated). Detailed definitions of all variables are provided in the Appendix. All the specifications include
year-quarter dummy variables. Heteroskedasticity-robust t-statistics clustered by bank and quarter are reported. Significance
at the 10%, 5%, and 1% level are indicated by * , ** , and *** , respectively.
HA 29

6.3 Other analyses

6.3.1 Subsample analysis according to size

I now examine whether the impact of conservatism on loan portfolio quality varies across banks depending on their
size. Because the type of information used in the evaluation of loans as well as the organizational structure are sig-
nificantly different depending on the size of a bank, it is important to check whether the relation between CON and
LQ documented so far applies to all banks, regardless of their size.14 Following Beatty and Liao (2011), I classify banks
as small if they have total assets greater than $500 million but less than $1 billion, medium if they have total assets
between $1 billion and $3 billion, and large if they have total assets in excess of $3 billion, and I estimate equation (4)
for each subsample.
In an untabulated analysis, I find that the coefficient on CON is significantly negative for all banks, regardless of size,
for both LQ proxies. The same inferences hold when LLP_ASY is used as a conservatism proxy. The results suggest that
monitoring benefits are associated with conservatism regardless of a bank’s size.

6.3.2 Change specifications

The results are consistent with the notion that conservatism affects banks’ monitoring efforts regarding their loan
portfolios. However, other variables could be correlated with both conservatism and loan portfolio quality. Further-
more, if both conservatism and loan portfolio quality are the banks’ choices, my analyses could suffer from endogene-
ity. To further rule out correlated omitted variables and endogeneity as potential concerns, I estimate equation (4)
and equation (5) with first differences, where the change (from t+1 to t+2) in loan quality is regressed on the change
(from t to t+1) in CON and in the control variables. Employing a first-differenced specification mitigates the effect of
bank-specific characteristics that are relatively constant over time.
Untabulated results show that no inferences are affected by a change specification. Regardless of the choice of con-
servatism proxy or loan quality proxy, the coefficient on ΔCON is negative and significant, indicating that an increase
in CON is associated with an increase in monitoring efforts. Also, the results of testing the conditional effect of ex ante
information asymmetry and the conditional effect of distress risk on the relation between conservatism and loan qual-
ity are robust to this specification.

6.3.3 Use of alternative asset quality measures

My analysis assumes that better loan portfolio quality is the outcome of increased monitoring efforts. To shed further
light on whether this premise is reasonable, in this section I use two alternative measures to capture the degree of
bank monitoring.
The first measure is the Z-Score, the distance from insolvency, based on the premise that bank risk is expected
to decrease with heightened monitoring efforts. The Z-Score is defined as the inverse of the probability of insolvency.
Insolvency is defined as a state where (ROA + CAR) ≤ 0, with ROA being the bank’s returns on assets and CAR its capital-
asset ratio. Following Laeven and Levine (2009), the Z-Score is estimated as ROA plus the CAR divided by the standard
deviation of ROA (ROAVOL) calculated on a moving average basis over the preceding 12 quarters. I use the natural

14 The relation between conservatism and loan monitoring efforts may be greater for smaller banks than for larger banks because smaller banks are inher-

ently more subject to greater information asymmetry. In the case of firms with more unverifiable information or greater information asymmetry, it is also
costly for outsiders to evaluate managers’ investment decisions and the overall firm performance (Gaver & Gaver, 1993; LaFond & Watts, 2008; Smith &
Watts, 1992). Conservatism is thus likely to play an even more important role in reducing managers’ incentives to make unprofitable and risky investment
decisions in smaller banks.
30 HA

logarithm of the Z-Score to address skewness. For brevity, the label ‘Z-Score’ is applied to indicate the natural loga-
rithm of the Z-score throughout the paper.
As another alternative proxy for monitoring quality, I use salary expenses, following Coleman, Esho, and Sharpe
(2006) and Bhat and Desai (2017). The main presumption here is that the quality and quantity of a bank’s staff reflect
its loan screening/monitoring efforts and ability. I use the ratio of the median adjusted salary expense to the total
noninterest expense as an alternative dependent variable for capturing banks’ loan monitoring efforts15 .
Untabulated results show that the coefficient on CON is statistically significant and positive across both CON mea-
sures C-Score_LLP and LLP_ASY and both alternative monitoring quality measures Z-Score and the ratio of the median
adjusted salary expense to the total noninterest expense, corroborating a positive relation between conservatism and
monitoring efforts.

7 CONCLUSION

This study provides evidence regarding the monitoring benefits of bank financial transparency defined as conser-
vatism or timelier recognition of expected loan losses. I find robust evidence that the quality of banks’ loan portfolios
measured by future levels of nonperforming loans or charge-offs increases with conservatism, consistent with market
discipline-enhancing lending practices. The results also indicate that a positive association between conservatism and
loan quality is more pronounced for banks with potentially greater incentives for risk-shifting or risk-taking. Further-
more, the findings suggest that this effect is greater when banks are likely to experience deteriorating loan portfolio
quality, such as during low and high lending growth cycles. Moreover, banks benefit from conservatism regardless of
their size.
While the banking literature posits that transparency can promote bank stability by enhancing the market disci-
pline of banks’ risk-taking decisions, no existing research looks directly into the relation between transparency and
a bank’s own monitoring effort. Using financial reporting conservatism as a proxy of transparency, this study finds
robust evidence that conservatism improves banks’ own monitoring efforts. Additionally, despite ample evidence that
conservatism improves investment efficiency in nonfinancial firms, to date no study has investigated this relation in
the banking sector. Viewing loan portfolio quality as the outcome of banks’ main investment decisions, this study con-
tributes to the investment efficiency literature the finding that the benefit of conservatism in improving investment
efficiency extends to the banking sector. Given that the benefits of conservatism are shared by lenders as well as bor-
rowers, these findings should be of interest to regulators and policymakers who debate how to incentivize banks to
use the discretion inherent in their loan loss provisioning in ways that are more informative and less opportunistic.

ACKNOWLEDGEMENTS
I am highly grateful to the Senior Editor, Andrew Stark, and to the anonymous referee, who provided comments that
helped us make a substantial improvement to the paper. I am also grateful to Christopher Williams (discussant), Wei
Wang (discussant) and seminar participants at the 2018 Journal of Accounting, Auditing and Finance Conference,
2018 European Accounting Association Annual Congress, and the 2019 Hawai’i Accounting Research Conference for
helpful discussions.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are from subscription sources discussed in Section 4 of the paper.

15 Because bank size and loan composition significantly affect salary expenses, the median salary is calculated for each size tercile within the tercile based

on the ratio of commercial loans to total assets for each quarter. Unlike consumer loans, which are mostly homogenous and monitored as pools, commercial
loans are heterogeneous and require more individual effort to monitor, entailing a higher salary expense for the same dollar amount of loans.
HA 31

CONFLICT OF INTEREST
The author has no conflict of interest to declare.

REFERENCES
Acharya, V. V., & Mora, N. (2015). A crisis of banks as liquidity providers. The Journal of Finance, 70(1), 1–43.
Acharya, V. V., & Richardson, M. (2009). Restoring financial stability: How to repair a failed system. Hoboken, NJ: John Wiley &
Sons.
Akins, B., Dou, Y., & Ng, J. (2017). Corruption in lending: The role of timely loan loss recognition. Journal of Accounting and
Economics, 64(2-3), 453–478.
Andreou, P. C., Cooper, I., Louca, C., & Philip, D. (2017). Bank loan loss accounting treatment, credit cycles and crash risk. The
British Accounting Review, 49(5), 474–492.
Armstrong, C., Guay, W., & Weber, J. (2010). The role of information and financial reporting in corporate governance and debt
contracting. Journal of Accounting and Economics, 50(2-3), 179–234.
Balakrishnan, K., Watts, R., & Zuo, L. (2016). The effect of accounting conservatism on corporate investment during the global
financial crisis. Journal of Business Finance & Accounting, 43(5), 513–542.
Basu, S. (1997). The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting and Economics,
24(1), 3–37.
Beatty, A., Chamberlain, S., & Magliolo, J. (1995). Managing financial reports of commercial banks: The influence of taxes, reg-
ulatory capital, and earnings. Journal of Accounting Research, 33(2), 231–261.
Beatty, A., & Liao, S. (2011). Do delays in expected loss recognition affect banks’ willingness to lend? Journal of Accounting and
Economics, 52, 1–20.
Beaver, W., & Ryan, S. (2005). Conditional and unconditional conservatism, concepts and modeling. Review of Accounting Stud-
ies, 10(2–3), 269–309.
Berger, A., & Udell, G. (2004). The institutional memory hypothesis and the procyclicality of bank lending behavior. Journal of
Financial Intermediation, 13(4), 438–495.
Berger, A. N., & DeYoung, R. (1997). Problem loans and cost efficiency in commercial banks. Journal of Banking and Finance,
21(6), 849–870.
Bhat, G., & Desai, H. (2017). Bank capital structure and loan Monitoring. Retrieved from [Link]
Bhattacharya, S., & Thakor, A. (1993). Contemporary banking theory. Journal of Financial Intermediation, 3(1), 2–50.
Brockman, P., Ma, T., & Ye, J. (2015). CEO compensation risk and timely loss recognition. Journal of Business Finance & Account-
ing, 42(1 & 2), 204–236.
Bushman, M., Piotroski, J., & Smith, A. (2011). Capital allocation and timely accounting recognition of economic losses. Journal
of Business Finance & Accounting, 38(1 & 2), 1–33.
Bushman, R., & Smith, A. (2001). Financial accounting information and corporate governance. Journal of Accounting and Eco-
nomics, 32(1-3), 237–333.
Bushman, R., & Williams, C. (2012). Accounting discretion, loan loss provisioning, and discipline of banks’ risk-taking. Journal of
Accounting and Economics, 54, 1–18.
Bushman, R., & Williams, C. (2015). Delayed expected loss recognition and the risk profile of banks. Journal of Accounting
Research, 53(3), 511–553.
Bushman, R. (2016). Transparency, accounting discretion, and bank stability. Economic Policy Review, 22(1), 129–149.
Cano-Rodriguez, M., & Nunez-Nickel, M. (2015). Aggregation bias in estimates of conditional conservatism: Theory and evi-
dence. Journal of Business Finance & Accounting, 42(1 & 2), 51–78.
Coleman, A. D. F., Esho, N., & Sharpe, I. G. (2006). Does bank monitoring influence loan contract terms? Journal of Financial
Services Research, 30, 177–198.
Collins, J., Shackelford, D., & Wahlen, J. (1995). The coordination of regulatory capital, earnings, and taxes for banks. Journal of
Accounting Research, 33, 263–292.
Dell’Ariccia, G., & Marquez, R. (2006). Lending booms and lending standards. Journal of Finance, 61, 2511–2546.
Dugan, J. (2009). Loan loss provisioning and procyclicality (Remarks before the Institute of International Bankers). Retrieved
from [Link]
Financial Accounting Standards Board (FASB). (2016). Accounting Standards Codification 326, available at Retrieved from
[Link] Norwalk, CT.
Foos, D., Norden, L., & Weber, M. (2010). Loan growth and riskiness of banks. Journal of Banking and Finance, 32, 2929–2940.
Francis, J. R., & Martin, X. (2010). Acquisition profitability and timely loss recognition. Journal of Accounting and Economics,
49(1-2), 161–178.
Freixas, X., & Rochect, J. C. (1997). Macroeconomics of banking. Cambridge, MA: MIT Press.
García Lara, J. M., Garcia Osma, B., & Penalva, F. (2016). Conditional conservatism and firm investment efficiency. Journal of
Accounting and Economics, 61, 221–238.
32 HA

Gaver, J. J., & Gaver, K. M. (1993). Additional evidence on the association between the investment opportunity set and corpo-
rate financing, dividend, and compensation policies. Journal of Accounting and Economics, 16, 125–160.
Gebhardt, G., & Novotny-Farkas, Z. (2011). Mandatory IFRS adoption and accounting quality of European banks. Journal of
Business Finance & Accounting, 38(3 & 4), 389–333.
Goldstein, I., & Sapra, H. (2014). Should Banks’ Stress Test Results be Disclosed? An Analysis of the Costs and Benefits. Foun-
dations and Trends in Finance, 8(1), 1–54.
Gompers, P., Ishii, J., & Metrick, A. (2003). Corporate governance and equity prices. The Quarterly Journal of Economics, 118(1),
107–156.
Jensen, M., & Meckling, W. (1976). Theory of the firm: Managerial behavior, agency costs and ownership structure. The Journal
of Financial Economics, 3, 305–360.
Khan, M. S., & Watts, R. L. (2009). Estimation and empirical properties of a firm-year measure of accounting conservatism.
Journal of Accounting and Economics, 48(2-3), 132–150.
Kishan, R. P., & Opiela, T. P. (2012). Monetary policy, bank lending, and the risk-pricing channel. Journal of Money, Credit and
Banking, 44(4), 573–602.
Kravet, T. (2014). Accounting conservatism and managerial risk taking: Corporate acquisitions. Journal of Accounting and Eco-
nomics, 57(2-3), 218–240.
Laeven, L., & Levine, R. (2009). Bank governance, regulation and risk taking. Journal of Financial Economics, 93, 259–275.
LaFond, R., & Watts, R. L. (2008). The information role of conservative financial statements. The Accounting Review, 83, 447–
478.
Leventis, S., Dimitropoulos, P., & Owusu-Ansah, S. (2013). Corporate governance and accounting conservatism: Evidence from
the banking industry. Corporate Governance: An International Review, 21(3), 264–286.
Levine, R. (2004). The corporate governance of banks: A concise discussion of concepts and evidence. World Bank Policy
Research Working Paper, 3404. Washington, DC: The World Bank.
Li, X. (2015). Accounting conservatism and the cost of capital: An international analysis. Journal of Business Finance & Account-
ing, 42(5), 555–582.
Liu, C., & Ryan, S. (1995). The effect of bank loan portfolio composition on the market reaction to and anticipation of loan loss
provisions. Journal of Accounting Research, 33, 77–94.
McDonough, R., Panaretou, A., & Shakespeare, C. (2020). Fair value accounting: Current practice and perspectives for future
research. Journal of Business Finance & Accounting, 47(3-4), 303–332.
Manganaris, P., Beccalli, E., & Dimitropoulos, P. (2017). Bank transparency and the crisis. The British Accounting Review, 49,
121–137.
Martin, G. W., Thomas, W. B., & Wieland, M. M. (2016). S&P 500 Membership and Managers’ Supply of Conservative Financial
Reports. Journal of Business Finance & Accounting, 43(5–6), 543–571.
Morgan, D. (2002). Rating banks: Risk and uncertainty in an opaque industry. American Economic Review, 92, 874–888.
Myers, S., & Rajan, R. (1998). The paradox of liquidity. The Quarterly Journal of Economics, 113, 733–771.
Nichols, D., Wahlen, J., & Wieland, M. (2009). Publicly traded versus privately held: Implications for conditional conservatism
in bank accounting. Review of Accounting Studies, 14, 88–122.
Smith, C. W., Jr., & Watts, R. L. (1992). The investment opportunity set and corporate financing, dividend, and compensation
policies. Journal of Financial Economics, 32, 263–292.
Stephanou, C. (2010). Rethinking market discipline in banking: Lessons from the financial crisis (World Bank Policy Research
Working Paper 5227). Retrieved from [Link]
Turner, A. (2010). Banks are Different: Should Accounting Reflect that Fact?. Speech Given at the International Chartered Accoun-
tants in England and Wales (ICAEW). London, UK. [Link]
Vyas, D. (2011). The timeliness of accounting write-downs by U.S. financial institutions during the financial crises of 2007–
2008. Journal of Accounting Research, 49, 823–860.
Watts, R. L. (2003). Conservatism in accounting, Part I: Explanations and implications. Accounting Horizons, 3, 207–221.

How to cite this article: Ha J. Bank accounting conservatism and bank loan quality. J Bus Fin Acc. 2020;1–35.
[Link]
HA 33

APPENDIX
Variable Descriptions

Variable Description
Dependent Variables:
NPL Nonperforming loans (bhck5525+bhck5526) deflated by total assets (bhck2170) or Compustat
‘npatq/atq’
CO Net charge-offs (bhck4635-bhck4605) deflated by total assets (bhck2170) or Compustat ‘ncoq/atq’
Z-Score The negative value of the natural logarithm of the moving average of the return on assets
(bhck4300/bhck2170 or Compustat ‘piq/atq’) over the preceding
12 quarters plus the capital to asset ratio (bhck3210/bhck2170 or Compustat ‘ceqq/atq’) divided
by the standard deviation of return on assets. The standard deviation of return on assets is
calculated on a moving average basis over the preceding 12 quarters.
AdjSalExp The median adjusted salary expense to total noninterest expense ratio, where the median salary is
calculated for each size tercile within the tercile based on the ratio of commercial loans to total
assets for each quarter or Compustat ‘xstfwsq/xnitbq’

Conservatism (CON) measures:

NIit = 𝛽0 + 𝛽1 Dit + Rit (𝜇1 + 𝜇2 SIZEit + 𝜇3 MBit + 𝜇4 LEVit ) + Dit Rit (𝜆1t + 𝜆2t SIZEit
C-Score + 𝜆3t MBit + 𝜆4t LEVit ) + (𝛿1t SIZEit + 𝛿2t MBit + 𝛿3t LEVit + 𝛿4t Dit SIZEit + 𝛿5t Dit MBit
+ 𝛿6t Dit LEVit ) + 𝜀t (1)
where NI: Net income (Compustat ‘niq’) divided by lagged market value of equity (Compustat
‘cshoq’ × share price at the fiscal quarter end),
R: Quarterly returns compounded from monthly returns beginning the second month after the
prior fiscal quarter end,
D: An indicator variable that equals one for negative returns, and zero otherwise,
MV: Market value of equity, calculated as the natural log of market value of equity (Compustat
‘cshoq’ × share price at the fiscal quarter end).
MTB: Market value of equity divided by book value of equity (Compustat ‘ceqq’),
LEV: Long-term debt (Compustat ‘dlttq’) divided by market value of equity (Compustat ‘cshoq’ *
share price at the fiscal quarter end).
After equation (1) is estimated, C-Score is constructed using the estimated coefficients, as follows:
C−Scoreit = 𝛽3 = 𝜆1t + 𝜆2t SIZEt + 𝜆3t MBt + 𝜆4t LEVt + 𝜀.
To reduce the measurement errors, I use an indicator variable that is equal to one if C-Score is
greater than the median during the quarter, and zero otherwise.
C-Score_LLP Conservatism associated with loan loss provisions
LLPit = 𝛽0 + 𝛽1 Dit + Rit (𝜇1 + 𝜇2 SIZEit + 𝜇3 MBit + 𝜇4 LEVit )
+ Dit Rit (𝜆1t + 𝜆2t SIZEit + 𝜆3t MBit + 𝜆4t LEVit )
+ (𝛿1t SIZEit + 𝛿2t MBit + 𝛿3t LEVit + 𝛿4t Dit SIZEit
+ 𝛿5t Dit MBit + 𝛿6t Dit LEVit ) + 𝜀t (1A)
where LLP is provision (Compustat ‘rclq’ or bhck4230) deflated by total assets (Compustat ‘atq’ or
bhck2170).
After equation (1A) is estimated, C-Score_LLP is constructed using the estimated coefficients, as
follows:
C−Score_LLPit = 𝛽3 = 𝜆1t + 𝜆2t SIZEt + 𝜆3t MBt + 𝜆4t LEVt + 𝜀.
To reduce the measurement errors, I use an indicator variable that is equal to one if C-Score_LLP is
greater than the median during the quarter, and zero otherwise.

(Continues)
34 HA

Variable Description
C-Score_EBP Conservatism associated with earnings before provision conservatism
EBPit = 𝛽0 + 𝛽1 Dit + Rit (𝜇1 + 𝜇2 SIZEit + 𝜇3 MBit + 𝜇4 LEVit )
+ Dit Rit (𝜆1t + 𝜆2t SIZEit + 𝜆3t MBit + 𝜆4t LEVit )
+ (𝛿1t SIZEit + 𝛿2t MBit + 𝛿3t LEVit + 𝛿4t Dit SIZEit
+ 𝛿5t Dit MBit + 𝛿6t Dit LEVit ) + 𝜀t (1B)
where EBP is earnings before provision (bhck 4300 + bhck 4230) deflated by total assets (bhck
2170), or Compustat ‘(piq+ pllq)/atq.’
After equation (1B) is estimated, C-Score_EBP is constructed using the estimated coefficients, as
follows:
C−Score_EBPit = 𝛽3 = 𝜆1t + 𝜆2t SIZEt + 𝜆3t MBt + 𝜆4t LEVt + 𝜀.
To reduce the measurement errors, I use an indicator variable that is equal to one if C-Score_EBP is
greater than the median during the quarter, and zero otherwise.
LLP_ASY I run the rolling regressions for each bank quarter using the observations of the past three years
and requiring 12 observations for each regression:
∗
LLPt+4 = 𝛼0 + 𝛼1 ΔNPLt−2 + 𝛼2 DΔNPLt−2 + 𝛼3 (DΔNPLt−2 ΔNPLt−2 ) + 𝛼4 ΔNPLt−1 + 𝛼5 DΔNPLt−1
∗ ∗
+ 𝛼6 (DΔNPLt−1 ΔNPLt−1 ) + 𝛼7 ΔNPLt + 𝛼8 DΔNPLt + 𝛼9 (DΔNPLt ΔNPLt ) + 𝛼10 ΔNPLt+1
∗
+ 𝛼11 DΔNPLt+2 + 𝛼12 (DΔNPLt+1 ΔNPLt+1 ) + 𝛼13 CAPITALt + 𝛼14 EBPt + 𝜀t (3)
where DΔNPLt+x is a dummy variable that is set to one for bad news for credit losses, that is, when
ΔNPLt+x is positive, and zero otherwise. The capital ratio relative to total assets (CAPITAL) and
earnings before provision (EBP) are included to control for banks’ incentives to manipulate
provisioning to meet regulatory requirements or to smooth earnings.
LLP_ASY is an indicator variable that is equal to one if the average of the coefficients α3, α6, α9, and
α12 is above the median, and zero otherwise.
Conditioning Variables (RISKSHIFT)
INFASY An average of standardized values of bid-ask spread, volatility, and idiosyncratic risk, where the
bid-ask spread equals the annual average of the daily spread scaled by the midpoint between bid
and ask, volatility equals the annualized variance of the individual stock returns for the previous
250 trading days obtained from CRSP, and idiosyncratic risk equals the natural logarithm of the
standard deviation of the residuals from a market-model regression of excess returns on
value-weighted market excess returns for 60 months (with at least 24 observations).
JCD (Jumbo The JCD rate equals interest expense on JCDs divided by the quarterly average of the JCDs
Certificates (RIAD4174 (or RIADA517)/RCON3345 (or RCONA415) × 4).
of Deposit) Interest expense on JCDs: until 1966Q4: RIAD4174; from 1997Q3a: RIADA517.
rate Quarterly average of JCDs: until 1996Q4: RCON3345; from 1997Q1: RCONA514.
The interest expenses on JCDs (RIAD4174 and RIADA517) cumulate yearly, and therefore I take
first differences within each year to obtain quarterly interest expenses. Furthermore, I multiply
the fraction by four to obtain the annual CD rates. The definition of the CD rate follows Acharya
and Mora (2015) and Kishan and Opiela (2012).
LOWCAP An indicator variable that gets one if a bank’s level of balance sheet capital (equity capital deflated by
total assets (bhck3210/bhck2170) or Compustat ‘ceqq/atq’) is less than 7%, and zero otherwise.
POORPERF An indicator variable that gets one if a bank’s return on equity (ROE) is in the bottom quintile, and
zero otherwise, where ROE is calculated by earnings before provision (bhck4300+bhck4230)
deflated by equity capital (bhck3210) or Compustat ‘(piq+pclq)/ceqq.’

(Continues)
HA 35

Variable Description
Control Variables:
TIMELINESS The ratio of the reserve for credit losses (bhck3123 or Compustat ‘rclq’) divided by the
nonperforming loans (bhck5525+bhck5526 or Compustat ‘npatq’).
SIZE Log of total assets (bhck2170) or Compustat ‘atq.’
ABSGAP Absolute value of the one-year maturity gap (bhck3197+bhck3296+bhck3298+bhck3409)
deflated by total assets (bhck2170).
FEE Ratio of fee income (bhck4079) deflated by total interest income (bhck4074+bhck4079) or
Compustat ‘(fccq+ idilbcq)/niintq.’
ROAVOL The standard deviation of return on assets (bhck4300/bhck2170), calculated on a moving
average basis over the preceding 12 quarters, or Compustat ‘piq/atq’
EFF Ratio of non-interest expense (bhck4093+bhck4300) deflated by total assets (bhck2170) or
Compustat ‘xnitbq/atq.’
EBLLP Earnings before provision (bhck4300+bhck4230) deflated by total assets (bhck2170) or
Compustat ‘(piq+pclq)/atq.’
LOAN Loan (bhck2122) deflated by total assets (bhck2170) or Compustat ‘lntalq/atq.’
LG Change in loans deflated by total assets ((bhck2122t –bhck2122t-1 )/bhck2170t-1 ) or
Compustat ‘(lntalqt − lntalqt−1 )/atq.’
COMM Agricultural and commercial loans (bhck1763+bhck1764+bhck1590) deflated by total
assets (bhck2170).
RE Real estate loans (bhck1410) deflated by total assets (bhck2170).
CONS Consumer loans (bhck2008+bhck2011+bhckb538+bhckb539) deflated by total assets
(bhck2170).
CAP Equity capital (bhck3210) deflated by total assets (bhck2170) or Compustat ‘ceqq/atq.’
LIQUID Liquid assets
(bhck0081+bhck0395+bhck0397+bhck1773+bhck1350+bhck0276+bhck0277+
bhdmB987) deflated by total assets (bhck2170).
ΔGDP Quarterly change in GDP from St. Louis Federal Reserve website,
[Link]
HOMOGENEOUS HOMOGENOUS is coded as one if the ratio of consumer loans
(bhck2008+bhck2011+bhckb538+bhckb539) to total loans (bhck2122) is in the top
tercile, and zero if this ratio is in the bottom two terciles.

a Bushman and Williams (2012), using an international sample, estimate a similar equation to the above for each coun-

try and use the coefficient on ΔNPLt+1 as a proxy for forward-looking discretion in loan provision and the coefficient
on EBP as a proxy for smoothing discretion in loan provision.