Determinants of Signaling by Banks Through Loan Loss Provisions

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Journal of Business Research 58 (2005) 312 – 320

Determinants of signaling by banks through loan loss provisions


Kiridaran Kanagaretnama, Gerald J. Lobob,*, Dong-Hoon Yangc
a
DeGroote School of Business, McMaster University, Hamilton, ON, Canada L8S 4M4
b
Whitman School of Management, Syracuse University, Syracuse, NY 13244, USA
c
School of Business, Information and Communications University, Taejon, Korea
Received 10 April 2001; accepted 3 June 2003

Abstract

This study investigates whether bank managers use their discretion in estimating loan loss provisions (LLP) to convey information about
their banks’ future prospects. Bank managers’ propensities to signal their private information vary cross sectionally because they face
different conditions and have different incentives. This study hypothesizes that the propensity to signal varies negatively with bank size and
positively with earnings variability, future investment opportunities, and degree of income smoothing. The empirical evidence supports these
predictions. It suggests that the propensity to signal is positively related to the degree of information asymmetry and that bank managers
attenuate perceived undervaluation of their banks by communicating their private information about their banks’ favorable future prospects.
D 2003 Elsevier Inc. All rights reserved.

Keywords: Signaling; Loan loss provision; Earnings variability; Investment opportunity set; Income smoothing

1. Introduction The need for signaling arises when managers with


information indicating that bank values are larger than those
Accounting principles for commercial banks require that assessed by the market desire to have them revised upward.
losses of loan principal that are probable and reasonably One approach to doing so is to indicate that the bank is
estimable be recorded as loan loss provisions (LLP). This strong enough to absorb future potential losses by increasing
recognition of provisions for loan losses results in an the LLP. Beaver et al. (1989) suggest that an increase in LLP
increase in loan loss allowance and a reduction in current can indicate that ‘‘management perceives the earnings
net income. LLP is the expense representing management’s power of the bank to be sufficiently strong that it can
estimate of the year’s net change in probable loan losses. withstand a ‘hit to earnings’ in the form of additional loan
Bank managers have superior information relative to invest- loss provisions.’’ Implicit in their reasoning is that the
ors and other stakeholders regarding default risks inherent in increase in LLP conveys a ‘‘good news’’ signal about the
their loan portfolios; consequently, they can use that infor- strength of a bank’s future earnings. However, an increase in
mation for estimating the LLP in each period. Additionally, LLP can also be viewed as bad news, especially if it is not
because bank managers have latitude in using their judg- accompanied by other, more timely indicators of loan
ment to estimate the LLP, they can exercise discretion over default because LLP will then serve as the primary source
the timing of recognition of certain loan losses. This paper of information on loan default. Banks generally disclose
investigates (1) whether bank managers’ discretionary deci- information on the change in nonperforming loans, which is
sions over LLP signal information about future operating an important indicator of loan default. By controlling for
results and (2) whether cross-sectional differences in the timely indicators of loan default such as change in nonper-
propensity to signal are related to certain bank-specific forming loans, loan loss allowance and loan charge-offs,
economic characteristics. Prior research on this topic has any excess LLP will contain only the ‘‘good news’’ com-
not addressed the latter issue. ponent (Liu and Ryan, 1995; Wahlen, 1994).
Signaling theory indicates that management’s primary
* Corresponding author. Tel.: +1-315-443-3583; fax: +1-315-443-
motivation for signaling is to address the adverse selection
5457. problem (Akerlof, 1970) and that the credibility of the signal
E-mail address: [email protected] (G.J. Lobo). is positively related to its cost. To be credible for strong

0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.jbusres.2003.06.002
K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320 313

banks, the cost of a false signal must be sufficiently high for have been studied (Jensen et al., 1992; Kao and Wu, 1994;
banks that are not expected to perform well. In the case of Yoon and Starks, 1995), the determinants of the propensity
LLP, poorly performing banks (i.e., banks with low earn- to signal with accounting accruals remain unexplored.
ings) will report even lower earnings if they increase LLP. Considering the emerging importance of research on banks’
This, in turn, will increase the probability that they will be major accrual (i.e., LLP), it clearly is important to identify
audited by bank regulatory agencies and that the FDIC will underlying factors that explain banks’ signaling through
increase their insurance premiums. Furthermore, since man- LLP. This study attempts to do so. Knowledge of such
agement’s incentive compensation is often based on factors will enable users to better understand and interpret
reported earnings, an increase in LLP will reduce reported the information conveyed by reported accounting earnings.
earnings and result in lower management compensation. Additionally, by understanding the conditions under which
The increased probability of regulatory scrutiny along with managers use their discretion over LLP to communicate
the reduction in management compensation that could result their private information, regulators can distinguish between
from false signaling through increasing LLP suggest that managers’ use of accounting discretion for opportunistic
there are strong disincentives for managers of poor banks to purposes and their use of accounting discretion to increase
engage in false signaling. the informativeness of reported earnings.
Another disincentive for poor-performing banks to signal Most prior studies in this line of research estimate the
falsely by increasing LLP is reputation cost. Our study relationship between LLP and bank managers’ underlying
analyzes 78 large U.S. bank holding companies that have incentives using pooled time-series cross-sectional regres-
been in business for a long time. These banks are likely to sions. While such an approach is computationally simple
place a high value on the credibility of their actions in the and parsimonious, the assumption of a single, sample-wide
financial community and, consequently, are unlikely to risk relation fails to incorporate any form of heterogeneity
losing their credibility in exchange for potential short-term among banks. To the extent that individual bank differences
gains that may result from false signaling. are relevant, those differences could alter the outcome of the
Several studies hypothesize and find that bank managers tests if they are not specifically modeled because estimates
use discretionary components of provisions to signal their of the signaling parameter will be less precise and the
private information about future prospects of banks. Beaver related test statistics downward biased. Prior research
et al. (1989) recognize that investors interpret an unexpected reports that bank managers’ discretion over LLP is bank-
increase in LLP as a signal of a bank’s financial strength. specific, suggesting that parameters of interest should be
Scholes et al. (1990) suggest that bank managers can lower measured relative to within-bank means rather than relative
their cost of capital by exercising discretion over LLP to to annual or pooled cross-sectional means as in most of
convey their private information to investors. Other studies those prior studies. This study uses a bank-specific time-
(Grammatikos and Saunders, 1990; Musumeci and Sinkey, series approach in an attempt to address the above concerns.
1990; Elliott et al., 1991; Griffin and Wallach, 1991) Additionally, using a bank-specific approach facilitates
demonstrate that the capital market responded positively gaining insights into factors contributing to differences in
to banks’ increases in LLP in response to less developed the propensity to signal through LLP. This research exam-
countries’ debt payment suspensions. They attribute this ines four bank-specific factors, which reflect the degree of
finding to investors interpreting the increase in LLP posi- information asymmetry across banks: bank size, earnings
tively because it signals a bank’s intention and ability to variability, future investment opportunities, and degree of
resolve its problem debt situation. income smoothing.
Wahlen (1994) offers evidence that bank managers in- The empirical evidence indicates that the propensity to
crease the discretionary component of current LLP when signal differs across banks based upon the degree of infor-
future cash flow prospects improve. Beaver and Engel mation asymmetry. It varies negatively with bank size and
(1996) provide additional evidence of positive effects of positively with earnings variability, future investment op-
discretionary LLP on stock price, which is consistent with portunities, and degree of income smoothing. These findings
Wahlen’s (1994) findings. Liu et al. (1997) as well document have at least two important implications. First, they indicate
a positive market reaction to the unexpected increase in LLP, that using a bank-specific time-series approach that allows
but only for banks with low regulatory capital levels (‘‘at for differences in the propensity to signal rather than a cross-
risk’’ banks) in the fourth quarter. More recently, Kanagar- sectional approach that constrains the signaling propensity to
etnam et al. (2003) find that managers of undervalued banks be constant across banks leads to more precise parameter
use LLP to signal their banks’ future earnings prospects. In estimates and less biased test statistics. Second, these results
contrast to the aforementioned empirical findings, Ahmed et have implications for interpreting the findings of prior
al. (1999) find no support for the signaling hypothesis. research and for the design of future research on income
To date, however, economic factors affecting the propen- smoothing and capital management through discretionary
sity to signal with accounting accruals have not been LLP. By incorporating the bank-specific factors affecting
investigated. Although firm-specific economic factors af- signaling propensity that are identified in this study in their
fecting managers’ tendency to signal with financial policies research designs, that research can better control for differ-
314 K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320

ential signaling incentives, thereby facilitating more power- for the increased riskiness perceived from a more variable
ful tests of income smoothing and capital management. earnings stream. This suggests that managers of banks with
In the next section, we discuss potential determinants of higher earnings variability will have higher incentives to
signaling through LLP and develop testable hypotheses. We signal their future earnings prospects.
describe the data and sample selection in section 3, and the Theoretical support for the relation between earnings
empirical specifications and testing approach in section 4. variability and the propensity to signal is also provided by
We discuss the results in section 5 and present the con- two related streams of research. First, market microstruc-
clusions of the study in section 6. ture theory maintains that more noisy earnings tends to
exacerbate information asymmetries between the bank’s
management and outside investors, and between privately
2. Determinants of LLP signaling informed investors and market makers. To compensate for
this informational disadvantage, the market maker’s in-
Managers’ incentives to signal are motivated by their crease in the bid-ask spread will be greater for banks with
concerns about stock undervaluation resulting from infor- less predictable earnings (Affleck-Graves et al., 2002). The
mation asymmetry. Prior research attributes interfirm differ- resulting increase in the adverse selection component of the
ences in information asymmetry to the following factors: bid-ask spread will increase the bank’s cost of capital. If the
firm size, earnings variability, investment opportunity, own- manager of a bank with high historical variability in
ership concentration, strength of governance/monitoring earnings can reduce the adverse selection component of
mechanisms, and degree of income smoothing. Addressing the bid-ask spread, then the bank’s trading costs will be
the governance and ownership explanations are beyond the lower and the market for its securities more liquid, making
scope of this research. Consequently, this study focuses on the securities themselves more valuable (Callahan et al.,
the remaining four firm-specific characteristics: firm size, 1997). This suggests that managers of banks with relatively
earnings variability, investment opportunity, and degree of high earnings variability are more likely to engage in
income smoothing. signaling through LLP to reduce the adverse selection
component of the bid-ask spread and, consequently, their
2.1. Bank size cost of capital.
From a different perspective, if an increase in a firm’s
The information environment hypothesis suggests that earnings variability increases information asymmetry due to
larger banks are more closely monitored by regulatory the noise created by low predictability of earnings prospects,
agencies and followed by more analysts. Such banks are then the management of such a firm will have a greater
more likely to have strong linkages with analysts and incentive to signal. According to signaling rationales for
institutional investors. As a result, managers of larger banks voluntary disclosure proposed by Ross (1977) and Verrec-
will have less private information to signal through LLP chia (1983), managers with information about their firm
and, consequently, are less likely to use signaling devices being undervalued will disclose it credibly so that their stock
including LLP to communicate their private information. price will be revised upward, whereas managers with infor-
Similarly, the extant literature on determinants of firms’ mation that implies values below market will withhold their
disclosure policies documents a positive association be- information. Accordingly, if earnings variability increases
tween disclosure quality and firm size (Imhoff, 1992; Lang information asymmetry and thus results in undervaluation of
and Lundholm, 1993). This suggests that enhanced disclo- the firm, then managers of banks with high earnings vari-
sure quality will mitigate the information asymmetry be- ability will have greater incentives to signal their information
tween inside managers and outside investors, thereby about their banks’ future favorable prospects. The above
resulting in a weaker incentive to signal through accounting discussion suggests the following hypothesis:
accruals. Hence, we hypothesize the following:
Hypothesis 2: The propensity to signal is positively related
Hypothesis 1: The propensity to signal is negatively related to earnings variability.
to bank size.
2.3. Investment opportunity set
2.2. Earnings variability
Cross-sectional differences in propensity to signal may
Gebhardt et al. (2001) report that the implied risk also be related to investment opportunities. Generally, firms
premium is consistently higher for commercial banks and with higher investment opportunities or growth potential
that earnings variability and predictability are important have more flexibility in their choice of future investments
factors in explaining cross-sectional differences in implied (Titman and Wessels, 1988). The assets of growth firms are
risk premium. Financial practitioners also regard earnings generally more difficult to observe because they are primar-
variability as an indicator of risk. Barth et al. (1995) argue ily represented by future (unrecognized) investments. If so,
that bank shareholders will demand a higher risk premium there is likely to be more information asymmetry for firms
K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320 315

with higher investment opportunities. Arguably, because the In sum, the above discussion implies that income
degree of information asymmetry is more pronounced for smoothing enhances the quality of signaling; therefore, it
firms with more discretion in their choice of investment complements the signaling through LLP. Thus, we posit the
decisions, firms that derive a relatively higher proportion of following:
their market value from growth options relative to assets in
place will have stronger incentives to signal. Hypothesis 4: The propensity to signal is positively related
Relying on the foregoing, Ballester et al. (1998), hy- to the degree of income smoothing.
pothesize that growth firms have higher levels of abnormal
earnings than other firms because they have greater levels
of unrecorded economic assets. Conceptually, growth firms 3. Empirical models
should enjoy greater abnormal earnings than those expected
by the market because they have some economic assets that This section presents the model used for estimating the
are not fully reflected in financial statements. Furthermore, propensity to signal through LLP and the model for testing
the concurrent nature of uncertainty surrounding a larger the determinants of signaling. Next, it describes operational
portion of their assets will motivate bank managers to measures of the signaling determinants.
signal more to communicate to investors that the future
earnings prospects of their banks will exceed current market 3.1. Model for estimating propensity to signal
expectations.
The implication of the above arguments is that banks The following model is used for estimating the propen-
with positive changes in net investments will signal more sity to signal:
through LLP to mitigate the underpricing of securities
resulting from adverse selection by the market. This sug- LLPit ¼ a0i þ a1i LCOit þ a2i LLAit1 þ a3i CHNPLit
gests the following hypothesis:
þ di EBTPit þ ki CHEBTPitþ1 þ ci CAPit
Hypothesis 3: The propensity to signal is positively related
þ /i CAPit  REG þ eit ð1Þ
to the size of the investment opportunity set.

2.4. Degree of income smoothing where, for firm i in period t, LLPit = loan loss provision,
LCOit = net loan charge-offs, LLAit  1 = beginning loan
There are alternative views in the literature about why loss allowance, CHNPL it = change in nonperforming
managers engage in income smoothing. One is that if the loans, EBTP it = earnings before tax and provision,
income smoothing arises because of a need to reduce CHEBTPit + 1 = 1-year ahead change in earnings before tax
information asymmetry, then income smoothing will make and provision, CAPit = ratio of actual regulatory capital
the reported earnings more reflective of future perfor- before loan loss reserves to required regulatory capital,
mance (e.g., Warfield et al., 1995; Beatty and Harris, REG = one in the new capital regime (after 1990) and zero
1999). An alternative view is that, if the income smooth- in the old regime (1990 or earlier), ki = signaling coefficient,
ing arises to circumvent contracts designed to mitigate di = income-smoothing coefficient, ci, /i = capital manage-
agency costs, then, the earnings management will make ment coefficients. All variables other than CAP are deflated
the reported earnings less reflective of future performance. by beginning market value of equity.
Since our focus is on signaling for information enhance- If managers use LLP to signal future prospects, then the
ment, we do not discuss the effect of opportunistic income signaling parameter, ki, in Eq. (1) is expected to be positive.
smoothing. For signaling, LLP will be high when 1-year-ahead change
According to the information enhancement viewpoint, in premanaged earnings (CHEBTPt + 1) is high. To examine
income smoothing occurs when managers have superior bank managers’ signaling behavior through LLP, we com-
information about the value of their firm and want to pute a t statistic for the signaling parameter (ki) for each
communicate it to outsiders. Consequently, income smooth- sample bank. The t statistics are then aggregated across
ing is viewed as an additional signaling device for convey- sample banks to obtain a z statistic, as described in Dechow
ing managers’ favorable private information about their et al. (1994). This z statistic, which is distributed asymptot-
firms. Chaney and Lewis (1995) develop a model in which ically as a standard normal variate, is used to assess the
managers of high quality firms smooth reported earnings to statistical significance of our signaling hypothesis test.
convey their private information. They hypothesize that We include earnings before tax and provision (EBTPit)
managers have private information about their firm type and primary capital ratio (CAPit) to control for the potential
indicating that the expected value of the cash flow distri- effects on discretionary LLP of motivations related to
bution is larger for higher quality firms. Kao and Wu (1994) income smoothing and capital management. If discretionary
demonstrate that smoothing enhances the quality of signal- components of unexpected provisions are used by bank
ing because it reduces noise. managers to smooth earnings, then the income smoothing
316 K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320

parameter (di) is expected to be positive. The capital Accordingly, all regression variables in Eq. (2), including ki
management hypothesis (Moyer, 1990; Beatty et al., 1995; and di, are deflated by wi.
Ahmed et al., 1999) posits that managers of banks with low
regulatory capital have incentives to increase LLP because 3.3. Measures of signaling determinants
primary capital includes loan loss reserve. This suggests that
the coefficient of primary capital (ci) will be negative. Section 2 discussed the rationales for the hypothesized
However, the 1990 change in bank capital adequacy regu- relationships between the propensity to signal and its four
lations substantially reduced the incentive to manage capital determinants—bank size, earnings variability, investment
via LLP. Therefore, if the capital management incentive has opportunity set, and degree of income smoothing. Measures
declined in the new capital regime, the parameter (ci + /i) of the degree of income smoothing (di) are obtained from
should not differ from zero. Eq. (1). In this subsection, we describe the measurement of
The model also includes net loan charge-offs, change in the remaining three determinants which are estimated using
nonperforming loans, and beginning loan loss allowance to annual data over the entire sample period (1981 –1996).
explicitly account for the nondiscretionary portion of LLP.
Net loan charge-offs (LCOit) is related to LLP by construc- 3.3.1. Bank size (SIZE)
tion. Change in nonperforming loans (CHNPLit) will also Bank size is measured as the natural logarithm of total
have a positive effect on LLP, whereas beginning loan loss loans outstanding at the end of each year. Unlike alternative
allowance (LLAit  1) will have a negative effect on LLP. A size proxies such as market value of equity and total assets,
large beginning allowance will require a smaller provision total loans outstanding is well-suited for this study primarily
in the current period, and vice versa. Our choice of these because the room for managerial discretion over LLP
variables for estimating nondiscretionary LLP is consistent mainly depends upon the magnitude of outstanding loans.
with prior studies (Wahlen, 1994; Beaver and Engel, 1996;
Kim and Kross, 1998). 3.3.2. Earnings variability (EVAR)
Earnings variability is defined as the standard deviation of
3.2. Model for testing the determinants of signaling earnings per share over the sample period. Variability in the
earnings series increases the bank’s probability of bankrupt-
The signaling parameter estimated for each sample bank cy as well as its perceived risk. This, in turn, increases its cost
in Eq. (1) represents the bank’s propensity to signal. The of capital. We calculate earnings variability as follows:
purpose of the second-stage regression analysis presented in vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uN
this subsection is to examine the relationships between the uP
u ðEPSit  EPSi Þ2
strength of LLP signaling and bank-specific economic t t¼1
factors. Accordingly, we estimate the following cross-sec- EVARi ¼
N
tional regression (expected signs for the coefficient esti-
mates are in parentheses): where, for bank i, EVARi = earnings variability over the
sample period, EPSit = earnings per share excluding extraor-
ðÞ ðþÞ ðþÞ dinary items in year t, EPS i = mean earnings per share
ki ¼ x0 þ x1 SIZEi þ x2 EVARi þ x3 INVi
excluding extraordinary items over the sample period,
ðþÞ
þ x 4 di þ li ð2Þ N = number of observations over the sample period.

where, for bank i, ki = signaling coefficient (from Eq. (1)), 3.3.3. Investment opportunity set (INV)
SIZEi = average size over the sample period, EVARi = earn- Kallapur and Trombley (1999) provide a comprehensive
nings variability over the sample period, INVi = investment evaluation of the association between alternative investment
opportunities over the sample period, di = degree of income opportunity set proxies and realized growth. Based on an
smoothing (from Eq. (1)). analysis of annual samples comprising all Compustat firms
Note that Eq. (2) includes ki as the dependent variable centered on 1978 through 1991 as base years, they document
and di as an independent variable. Because di is estimated that the book-to-market ratio is the measure most highly
from the first-stage model, it is stochastic. This causes OLS (negatively) correlated with future growth. Following Kalla-
parameter estimates to be inefficient and statistical infer- pur and Trombley (1999), we employ the ratio of the market
ences to be unreliable (Pindyck and Rubenfeld, 1991). The value to the book value of total assets as a proxy for the
appropriate technique for obtaining efficient coefficient relative importance of investment opportunities, where the
estimates is to use weighted least squares estimation with market value of total assets is defined as the ending market
‘‘weighting’’ factors that take into account the variance – value of common equity plus the book value of preferred
covariance structure of these stochastic variables. The equity and liabilities. This measure has previously been used
weights are w i = Var[ki ]  (Cov[ki ,di ])/Var[di ], where by Smith and Watts (1992) and Mian (1996), among others,
Var[ki], Var[di]) and (Cov[ki, di]) are the variances and the to differentiate between assets in place and growth opportu-
covariance of ki and di, generated from estimating Eq. (1). nities. It relies on the notion that, if firms’ growth prospects
K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320 317

are at least partially impounded in stock prices, then growth tion for the period 1980 – 1997 available in the Keefe,
firms will have higher market values relative to assets in Bruyette, and Woods (KBW) database. The 1997 Bank
place. For example, banks with highly competent derivatives Compustat file contains 14,080 bank-year observations for
experts will have an unrecorded economic asset. This will 705 banks for the period 1978 – 1997. Eliminating banks
cause higher abnormal earnings and will, in time, result in lacking the necessary financial data leaves 3068 bank-year
relatively higher stock prices. Since the market-to-book ratio observations. From these, 1410 bank-year observations are
indicates the discrepancy between market price and book deleted due to first-differencing in total outstanding loans
value, banks with higher ratios will have relatively higher and 1-year-ahead changes in earnings before tax and provi-
information asymmetries and, not surprisingly, higher un- sion. To facilitate the firm-specific time-series regression
certainty in their future income streams. Therefore, the need analysis, we also require that sample firms have at least 10
for communicating favorable future prospects by managers years of data available. This requirement leaves 1253 obser-
of such banks will be relatively greater. vations (102 banks). Merging the Compustat and KBW
databases results in a final sample of 1120 observations,
spanning 16 years (1981 –1996) and including 78 banks.
4. Sample selection and descriptive statistics Table 1 provides descriptive statistics of key variables. As
shown in Panel A, the mean LLP is greater than the mean
Our sample includes banks from the 1997 Bank Compu- loan charge-off indicating that there is, on average, discre-
stat annual file that also have nonperforming loans informa- tionary addition to the loan loss allowance by charging more

Table 1
Descriptive statistics
Panel A: Unscaled variables in the first-stage model
Totals (US$MM except for CAP)
Variable Mean S.D. Min 25% Median 75% Max
LLP 111.6 642.0  60 7.21 19.02 64.80 4410
LCO 97.74 297.7  16.08 5.70 15.36 58.94 4934
LLAt  1 245.1 591.9 0 22.08 50.93 167.0 5368
CHNPL 8.360 253.99  2391  8.14 1.100 13.95 3463
NI 360.58 751.0  259.2 47.83 102.2 314.6 7999
CAP 1.51 0.305 0.706 1.284 1.509 1.714 2.759
CHEBTPt + 1  4.433 423.9  7789 1.692 11.71 37.0 2229
MV 1669.9 3406.7 6.75 223.18 522.1 1490.9 47,711

Panel B: Ratios
Ratio Mean Median S.D.
NPL/MV 0.17 0.08 0.32
LLP/MV 0.08 0.04 0.16
LCO/MV 0.06 0.03 0.11
NPL/LLA 0.96 0.81 0.73
NPL/LLP 3.32 2.03 9.84
NPL/LCO 2.75 2.60 53.4
LLP/LCO 1.47 1.23 7.94
LLP/NI 1.57 0.29 23.0

Panel C: Variables in the second-stage model


Variable Mean S.D. Min 25% Median 75% Max
k 0.050 0.187  0.346  0.032 0.020 0.093 0.884
SIZE 8.441 1.199 5.714 7.524 8.187 9.536 11.75
EVAR 1.720 1.510 0.173 0.791 1.318 2.096 9.357
INV 1.026 0.039 0.757 1.016 1.026 1.037 1.126
d 0.117 0.362  0.803  0.011 0.100 0.245 1.293
Variables are defined as follows: LLPit = loan loss provision (Bank Compustat #135), LCOit = net loan charge-offs (Bank Compustat #190),
LLAit  1 = beginning loan loss allowance (Bank Compustat #70), NPLit = beginning nonperforming loans, CHNPLit = change in nonperforming loans,
CAP = ratio of actual regulatory capital before loan loss reserves to required regulatory capital, EBTPit = earnings before tax and provision (Bank Compustat
#161+#135), CHEBTPit + 1 = one-year ahead change in earnings before tax and provision (Bank Compustat #145+#135), MVit = market value of equity (Bank
Compustat #90  #205), k = signaling coefficient estimated from Eq. (1), SIZE = average size, EVAR = earnings variability, INV = investment opportunities,
d = degree of income smoothing estimated from Eq. (1).
Variables are estimated over the period 1981 – 1996.
Descriptive statistics in panel C are based on 78 observations.
318 K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320

than needed for current write-offs. The ratios of average LLP, Table 3
Pearson correlations among second-stage regression variables (N = 78)
loan charge-offs and nonperforming loans to average begin-
ning market value of common equity reported in Panel B are k SIZE EVAR INV
0.08, 0.06, and 0.17, respectively. These ratios show im- SIZE  .009 (.936) –
provement relative to Wahlen’s (1994) sample largely be- EVAR .384 (.001) .418 (.001) –
INV .074 (.518)  .292 (.009)  .279 (.013) –
cause the overall quality of loans has improved with the
d .281 (.012)  .140 (.221)  .280 (.013)  .047 (.678)
economy over this period (Edwards and Mishikin, 1995).
See Table 1 for variable definitions.
Panel C of Table 1 reports descriptive statistics of
Values in parentheses denote p values.
variables used in the second-stage regression model that
examines the hypothesized determinants of the propensity to
signal. On average, earnings are relatively unstable as cient on capital management (c) is positive; however, the z
indicated by the earnings variability measure (EVAR). This statistic is not sufficiently large to reject the null hypothesis
is so because the sample period of our study ranges from the on capital management. Furthermore, roughly half of the 78
eighties when most banks suffered from ‘‘disintermedia- firms have positive capital management coefficients. This
tion’’ and a resulting decline in earnings to the nineties result is consistent with the results of Collins et al. (1995)
when their profits increased sharply due to growth of and inconsistent with Moyer (1990), Beatty et al. (1995), and
noninterest income derived from off-balance-sheet activities Ahmed et al. (1999). Plausible explanations for these incon-
such as fee and trading income. sistencies in tests of the capital management hypothesis
include the following: different sample periods, different
models to separate out the discretionary and nondiscretion-
5. Results ary components of LLP, and use of cross-sectional pooled
regression models that are less powerful in isolating this
5.1. Evidence on signaling through LLP managerial motive. The mean coefficient on the interaction
term (/) is negative, consistent with the argument that for the
Consistent with the signaling hypothesis, the bank-specif- period after 1990 the parameter (c + /) should not differ from
ic OLS regression results reported in Table 2 indicate a zero. The coefficients of all the control variables exhibit
significant positive association between LLP and change in statistically significant and predicted associations with LLP.
earnings before tax and provision. The mean (median) Finally, the mean (median) R2 is 0.875 (0.938) suggesting
signaling coefficient (k) is 0.059 (0.025); and 52 of the 78 that the model has, on average, high explanatory power in
firms exhibit positive coefficients. The z statistic for testing explaining banks’ LLP decisions.
the null hypothesis that the signaling coefficient equals zero is
3.568, significant at the 1% level. This result for signaling is 5.2. Evidence on determinants of signaling
largely consistent with Elliott et al. (1991), Griffin and
Wallach (1991), Wahlen (1994), and Kanagaretnam et al. 5.2.1. Correlation analysis
(2003). Table 3 reports Pearson correlation coefficients among the
The mean (median) coefficient on income smoothing (d) variables included in the second-stage regression model that
is 0.134 (0.119); and 61 of the 78 firms exhibit positive examines the relationships between the propensity to signal
coefficients. This result is consistent with those of prior and firm characteristics. As hypothesized, all the determi-
studies by Scholes et al. (1990), Wahlen (1994), Collins et al. nants of signaling exhibit the predicted signs. The simple
(1995), and Kanagaretnam et al. (2003). The mean coeffi- correlations of propensity to signal with earnings variability

Table 2
First-stage bank-specific OLS regressions: results are reported for 78 bank-specific regressions
LLPit ¼ a0i þ a1i LCOit þ a2i LLAit1 þ a3i CHNPLit þ di EBTPit þ ki CHEBTPitþ1 þ ci CAPit þ /i CAPit REG þ eit

a0 a1 a2 a3 d k c / Adjusted R2
(?) (+) () (+) (+) (+) () (+)
Mean 0.012 1.092  0.456 0.089 0.134 0.059 0.437  0.178 0.875
Median  0.001 1.055  0.400 0.048 0.119 0.025 0.164  0.050 0.938
Third Quartile 0.045 1.455  0.199 0.154 0.257 0.089 0.823 0.095 0.967
First Quartile  0.042 0.729  0.795  0.035 0.014  0.025  0.386  0.222 0.844
Maximum 0.670 2.872 2.497 1.143 1.293 0.884 6.501 1.956 0.995
Minimum  0.511  0.610  1.973  0.799  0.803  0.652  4.142  4.048 0.304
Number > 0 39 75 10 47 61 52 46 30 –
z statistics2 0.809 39.287  14.337 4.151 10.391 3.568 1.216  2.646 –
One-tailed significance level 0.1867 0.0001 0.0001 0.0377 0.0001 0.0001 0.1131 0.0041 –
See Table 1 for variable definitions.
REG = 1 after 1990 and 0 otherwise.
K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320 319

Table 4
WLS regression results for determinants of signaling (N = 78)
ki ¼ x0 þ x1 SIZEi þ x2 EVAR þ x3 INVi þ x4 di þ li

Predicted sign Intercept (?) SIZE (  ) EVAR (+) INV (+) SMOOTHING (+)
Coefficient  4.003  0.093 0.144 4.572 0.416
t value  3.082  4.098 9.557 3.785 8.887
p value 0.0029 0.0001 0.0001 0.0001 0.0001
Adjusted R2=.6788
See Table 1 for variable definitions.
Significance levels denoted by p values are based on a one-tail test.

(EVAR) and degree of income smoothing (d) are statistically private information about their banks’ future prospects.
significant while those with bank size (SIZE) and investment Unlike prior research which assumes that the propensity to
opportunity set (INV) are not. The correlations between signal through LLP is uniform across banks, this study
earnings variability (EVAR) and bank size (SIZE), invest- posits that, because bank managers face different conditions
ment opportunity set (INV) and degree of income smoothing and have different incentives, their propensities to signal
(d) are also significant as is the correlation between bank size their private information vary systematically with bank size,
(SIZE) and investment opportunity set (INV). earnings variability, future investment opportunities, and
degree of income smoothing.
5.2.2. Regression analysis Our empirical analysis indicates that the propensity to
Table 4 reports the weighted least squares regression signal is not cross-sectionally constant; it varies negatively
results for the relation between the propensity to signal (k) with bank size and positively with earnings variability, future
and the selected bank-specific economic variables. The investment opportunities, and degree of income smoothing.
coefficients of all four determinants exhibit the predicted These results are consistent with the hypothesis that the
signs. Bank size (SIZE) has the predicted negative sign and propensity to signal varies positively with the degree of
is statistically significant. Thus, the hypothesis that larger information asymmetry. They also are consistent with the
banks have weaker incentives to signal than smaller banks is notion that managers have incentives to attenuate perceived
statistically supported by the sample data. undervaluation of their banks by communicating their private
Earnings variability (EVAR) has a positively significant information about their banks’ favorable future prospects.
association with the strength of signaling at less than the 1% Our findings have implications for users of financial
level. This result supports the hypothesis that managers of statements and for policymakers. Knowledge of factors that
banks with high earnings variability have more incentive to explain differences in the propensity to signal will enable
signal their banks’ future favorable prospects, presumably users to better understand and interpret the information
because an increase in earnings variability increases the conveyed by reported accounting earnings. Furthermore,
information asymmetry stemming from unpredictability of by understanding the conditions under which managers
performance. Thus, the propensity to signal is greater when use their discretion over LLP to communicate their private
outsiders do not know as much about the corporation and its information, regulators can better distinguish between man-
managers’ efforts as do insiders. agers’ use of accounting discretion for opportunistic pur-
The results in Table 4 indicate that the coefficient on the poses and their use of accounting discretion to increase the
investment opportunity set (INV) is significantly positive at informativeness of reported earnings.
the 1% level, which is consistent with the hypothesis that Our findings also have implications for interpreting the
banks with greater investment opportunities signal more findings of prior research and for future research in this area.
than do banks with fewer investment opportunities. Finally, First, they demonstrate that using a bank-specific time-series
the estimated regression coefficient on the degree of income approach that recognizes firm-specific differences in the
smoothing (d) is significant and positive at less than the 1% propensity to signal rather than the commonly used cross-
level. This result is consistent with the hypothesis that bank sectional approach that ignores such differences results in
managers’ smoothing through LLP is, in general, an effec- more precise parameter estimates and less biased test
tive means of enhancing their signaling. statistics on signaling propensity. Second, our results have
implications for interpreting the findings of research on
income smoothing and capital management through discre-
6. Summary and conclusions tionary LLP. That research typically controls for the signal-
ing incentive by assuming that it is cross-sectionally
This study extends the literature on bank managers’ use uniform. Given that there are systematic differences across
of their discretion in estimating LLP to communicate their banks in the propensity to signal through LLP, as demon-
320 K. Kanagaretnam et al. / Journal of Business Research 58 (2005) 312–320

strated in this study, such controls will be less efficacious. changes in bank loan loss reserve levels. Account Rev 1991;66(Octo-
ber):847 – 61.
Incorporating the factors affecting the propensity to signal
Gebhardt WR, Lee CMC, Swaminathan B. Toward an implied cost of
that were documented in this study can help improve the capital. J Acc Res 2001;39(June):135 – 76.
research designs of those studies. Grammatikos T, Saunders S. Additions to loan-loss reserves: good news or
bad news? J Monet Econ 1990;25(March):289 – 304.
Griffin P, Wallach S. Latin American lending by major U.S. banks: the
Acknowledgements effect of disclosures about nonaccrual loans and loan loss provisions.
Account Rev 1991;66(October):831 – 47.
Imhoff EA. The relation between perceived accounting quality and eco-
We thank the two anonymous reviewers for their con- nomic characteristics of the firm. J Account Public Policy 1992;
structive suggestions. Gerald Lobo acknowledges the 11(Summer):97 – 118.
financial support provided by the Office of the Vice President Jensen GR, Solberg PD, Zorn TS. Simultaneous determination of insider
of Research and Computing, the School of Management, and ownership, debt, and dividend policies. J Financ Quant Anal 1992;
27(June):247 – 63.
the George E. Bennett Center at Syracuse University.
Kallapur S, Trombley MA. The association between investment opportunity
set and realized growth. J Bus Finance Account 1999;26(April/May):
505 – 19.
References Kanagaretnam K, Lobo GJ, Yang DH. Joint tests of signaling and income
smoothing through bank loan loss provisions. Syracuse University
Affleck-Graves JF, Callahan CM, Chipaklatti N. Earnings predictability, in- working paper; 2003.
formation asymmetry, and market liquidity. J Acc Res 2002;40(June): Kao C, Wu C. Tests of dividend signaling using the Marsh – Merton model:
561 – 84. a generalized friction approach. J Bus 1994;67(January):45 – 68.
Ahmed AS, Takeda C, Thomas S. Bank loan loss provisions: a reexami- Kim M, Kross W. The impact of the 1989 change in bank capital standards
nation of capital management, earnings management and signaling ef- on loan loss provisions and loan write-offs. J Account Econ 1998;
fects. J Account Econ 1999;28(November):125. 25(February):69 – 99.
Akerlof G. The market for ‘lemons’: quality uncertainty and the market Lang M, Lundholm R. Cross-sectional determinants of analysts’ ratings of
mechanism. Q J Econ 1970;89(August):488 – 500. corporate disclosures. J Acc Res 1993;31(Autumn):246 – 71.
Ballester M, Livnat J, Seethamraju C. Individual-firm style loadings, unre- Liu C, Ryan SG. The effect of bank loan portfolio composition on the
corded economic assets, and systematic risk. J Account Audit Financ market reaction to and anticipation of loan loss provision. J Acc Res
1998;14(Summer):275 – 96. 1995;33(Spring):77 – 94.
Barth ME, Landsman WR, Wahlen JM. Fair value accounting: effects on Liu C, Ryan SG, Wahlen JM. Differential valuation implications of loan
banks’ earnings volatility, regulatory capital, and value of contractual loss provisions across banks and fiscal quarters. Account Rev 1997;
cash flows. J Bank Financ 1995;9(June):577 – 605. 72(January):133 – 46.
Beatty A, Harris DG. The effects of taxes, agency costs and information Mian SL. Evidence on corporate hedging policy. J Financ Quant Anal
asymmetry on earnings management: a comparison of public and pri- 1996;31(September):419 – 39.
vate firms. Rev Acc Stud 1999;4(December):299 – 326. Moyer SE. Capital adequacy ratio regulations and accounting choices in
Beatty A, Chamberlain SL, Magliolo J. Managing financial reports of commercial banks. J Account Econ 1990;13:123 – 54.
commercial banks: the influence of taxes, regulatory capital, and earn- Musumeci JJ, Sinkey JF. The international debt crisis and bank loan-loss
ings. J Acc Res 1995;33(Autumn):231 – 61. reserve decisions: the signaling content of partially anticipated events. J
Beaver WH, Engel EE. Discretionary behavior with respect to allowance for Money Credit Bank 1990;22(August):379 – 87.
loan losses and the behavior of securities prices. J Account Econ 1996; Pindyck RS, Rubenfeld DL. Econometric Models and Economic Forecasts.
22:177 – 206. New York (NY): McGraw-Hill; 1991.
Beaver WH, Eger C, Ryan SG, Wolfson M. Financial reporting, supple- Ross S. The determination financial structure: the incentive signaling ap-
mental disclosures, and bank share price. J Acc Res 1989;27(Autumn): proach. Bell J Econ 1977;8(Spring):23 – 40.
157 – 78. Scholes MS, Wilson GP, Wolfson MA. Tax planning, regulatory capital
Callahan C, Lee CM, Yohn T. Accounting information and bid-ask spreads. planning, and financial reporting strategy for commercial banks. Rev
Account Horiz 1997;11(December):50 – 60. Financ Stud 1990;3:625 – 50.
Chaney PK, Lewis CM. Earnings management and firm valuation under Smith CW, Watts RL. The investment opportunity set and corporate fi-
asymmetric information. J Corp Finance Contract Gov Organ 1995; nancing, dividend, and compensation policies. J Financ Econ 1992;32
1(April):319 – 45. (December):263 – 92.
Collins JH, Shackelford DA, Wahlen JM. Bank differences in the coor- Titman S, Wessels R. The determinants of capital structure choice. J Fi-
dination of regulatory capital, earnings, and taxes. J Acc Res 1995; nance 1988;43(March):1 – 20.
33(Autumn):263 – 91. Verrecchia RE. Discretionary disclosure. J Account Econ 1983;5(Decem-
Dechow P, Huson MR, Sloan RG. The effect of restructuring charges ber):179 – 94.
on executives’ cash compensation. Account Rev 1994;69(January): Wahlen JM. The nature of information in commercial bank loan loss dis-
138 – 56. closures. Account Rev 1994;69(July):455 – 78.
Edwards FR, Mishikin FS. The decline of traditional banking: implications Warfield TD, Wild JJ, Wild KL. Managerial ownership, accounting choices,
for financial stability and regulatory policy. Fed Reserve Bank N Y and informativeness of earnings. J Account Econ 1995;20(July):61 – 91.
Econ Policy Rev 1995;(July):27 – 45. Yoon PS, Starks LT. Signaling, investment opportunities, and dividend
Elliott J, Hanna JD, Shaw W. The evaluation by the financial markets of announcements. Rev Financ Stud 1995;8(Winter):995 – 1018.

You might also like