Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: We examine whether and how measures of market and credit risk modeling identified from banks’
Received 5 June 2014 financial reports enhance the returns-relevance of their estimated annual unrealized fair value gains and
Revised 15 April 2015
losses for financial instruments. To capture differences in market liquidity and fair valuation difficulties
Accepted 16 April 2015
across types of financial instruments, we distinguish unrealized gains and losses that are recorded in net
Available online 7 May 2015
income versus recorded in other comprehensive income versus calculable using financial statement note
disclosures. We predict and generally find that banks’ market (credit) risk modeling enhances the returns-
relevance of their unrealized fair value gains and losses, more so for less liquid instruments subject to
greater market-risk-related (credit-risk-related) valuation difficulties and during periods for which market
(credit) risk is higher. We obtain these findings both for banks’ unadjusted risk modeling measures and
for the portions of these measures that we model as attributable to banks’ risk modeling activities, but
not for the portions we model as attributable to banks’ disclosure of these activities.
© 2015 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.aos.2015.04.004
0361-3682/© 2015 Elsevier Ltd. All rights reserved.
82 G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95
uity to interest rate movements. It also involves simulating the ef- riod. Since all banks must engage in at least minimal levels of
fects of interest rate movements on the prepayment of fixed-rate MRM and CRM to make investment and financing decisions and to
mortgages and exercise of other interest-rate options. Credit risk estimate the fair values of financial instruments for which market
is variability of the value of funded assets attributable to uncer- data do not suffice for the task, it appears that some banks do not
tainty about default probabilities, losses given default, and timing disclose these activities. Hence, non-disclosure of a risk modeling
of default, as well as variability of the value of unfunded loan com- activity does not imply absence of the activity. We assume, how-
mitments due to uncertainty about draws on those commitments, ever, that our MRM and CRM measures capture meaningful varia-
which are more likely to occur during worse economic times. CRM tion in risk modeling intensity across banks and time.
involves analysis of these parameters based on attributes of the We test all hypotheses using both one-stage and two-stage ap-
borrowers, borrowing contracts, borrowers’ performance to date on proaches. The one-stage approach regresses returns for the twelve
the contracts, and relevant economic conditions. months ending four months after the fiscal year end on net in-
We argue that banks’ MRM and CRM activities enhance the come before FVGL recorded in net income3 and the three types
quality of their estimates of FVGLs on financial instruments when of FVGL (recorded in net income, recorded in other comprehen-
two conditions hold: (1) the relevant markets for those instru- sive income, and disclosed), separately and interacted with the un-
ments are sufficiently illiquid that prices or other information from adjusted MRM and CRM measures, as well as control variables.
these markets do not substantially determine the instruments’ fair We frame and test our hypotheses as restrictions on the one or
values; and (2) the instruments exhibit features, such as embed- more coefficients on the interactions of banks’ unadjusted MRM
ded options or complex structuring, that increase the difficulty of and CRM measures with specific types of FVGL. Empirical results
estimating the instruments’ fair values. As a first cut to capture using this approach support our main hypotheses with one ex-
the applicability of these conditions, we distinguish FVGL that are plainable exception.
recorded in net income versus recorded in other comprehensive in- We use the two-stage approach to help ensure that the one-
come versus calculable using financial statement note disclosures stage approach results are attributable to banks’ risk modeling ac-
(cȁdiscloseddȁ). Fig. 1 summarizes current relevant U.S. generally tivities rather than to their choice to disclose these activities. In
accepted accounting principles, under which FVGL are recorded in this approach, we first regress banks’ unadjusted MRM and CRM
net income for most trading and risk management instruments measures on proxies for their discipline over risk modeling, techni-
and in other comprehensive income for available-for-sale securi- cal sophistication, risk exposures, and risk tolerance, which we ex-
ties and cash-flow-hedge derivatives. FVGL are disclosed for most pect primarily indicate banks’ risk modeling activities rather than
of banks’ other primary types of financial instruments, including their disclosure of those activities. We use the explained (unex-
their largest asset, loans, and largest liability, deposits. We propose plained) portions of banks’ unadjusted MRM and CRM measures
three main hypotheses below that we test by examining whether from these first-stage models as measures of banks’ risk model-
and how MRM and CRM enhance the returns-relevance of these ing activities (disclosure of these activities) in second-stage re-
three types of FVGL from 2002 to 2013. turns models. The estimated coefficients on the MRM and CRM
Our first and most general hypothesis is that banks’ MRM and activity measures in the two-stage approach yield the same in-
CRM enhance the returns-relevance of their FVGL, more so for ferences as the estimated coefficients on the unadjusted measures
FVGL on less liquid and more difficult-to-fair-value financial instru- in the one-stage approach, whereas the estimated coefficients on
ments. In testing this hypothesis, we exploit the fact that banks’ the MRM and CRM disclosure measures generally are insignificant.
financial instruments for which FVGL are disclosed, such as loans These results are consistent with the one-stage approach results
and deposits, usually are less liquid and more difficult to fair being driven by banks’ risk modeling rather than their disclosure
value than their other instruments. Our second hypothesis is that of that modeling.
banks’ MRM also enhances the returns-relevance of their FVGL We further hypothesize that MRM more strongly impacts the
recorded in other comprehensive income. Available-for-sale securi- returns-relevance of FVGL that are recorded in other comprehen-
ties and cash-flow-hedge derivatives typically are near credit risk- sive income or disclosed in years with high interest rate volatility,
less. Moreover, to the limited extent that banks experience credit and that CRM more strongly impacts the returns-relevance of dis-
losses on these instruments, banks typically record these losses closed FVGL during the financial crisis. To test these predictions,
in net income under impairment accounting rules. Hence, interest we interact the primary test variables with indicator variables for
rate risk is the primary risk reflected in FVGL recorded in other years with above-median interest rate volatility or the crisis period
comprehensive income. We expect this hypothesis to hold only for 2007–2009. Empirical results for the unadjusted MRM and CRM
available-for-sale securities and cash-flow-hedge derivatives that measures and the MRM and CRM activity measures generally sup-
are both less than highly liquid and exhibit fair valuation difficul- port these further hypotheses.
ties, such as mortgage-backed and asset-backed securities, so that Our study contributes to the extensive literature beginning
MRM is essential to estimate the fair values of the instruments with Barth (1994) that empirically examines the extent and de-
accurately. Our third hypothesis is that banks’ CRM primarily im- terminants of the value-relevance of fair values and the returns-
pacts the returns-relevance of their disclosed FVGL, because banks relevance of FVGL for financial instruments. Our study is most re-
assume credit risk primarily through their funded loans and un- lated to recent papers examining disclosures of fair valuation in-
funded loan commitments. puts and other measures of the reliability of recognized fair value
To test these hypotheses, we identify banks’ risk modeling ac- estimates under Statement of Financial Accounting Standards (FAS)
tivities from disclosures in their Form 10-K filings. As described in 157 (2006, Accounting Standards Codification (ASC) 820), which
the Appendix cȁRisk modeling measures and chief risk officer indi- became effective in 2008. In particular, Chung, Goh, Ng, and Yong
catordȁ, we hand collect disclosures of five MRM activities (inter-
est rate gap analysis, interest rate sensitivity analysis, Value-at-Risk
analysis, stress testing, and backtesting) and four CRM activities 3
Net income before FVGL recorded in net income includes realized gains and
(statistical credit risk measurement, credit scoring, internal credit losses that are distinct from FVGL except for two types of impairment write-down
risk rating, and stress testing). We equally weight these activities that are included in net income and thus are accounted for in the same fashion
as realized losses. First, net income includes all or the credit loss portion of other-
to construct indices of banks’ MRM and CRM. This approach raises
than-temporary impairment write-downs of available-for-sale and held-to-maturity
the issue that many bank-year financial reports include little about securities. Second, net income includes losses on loans held for sale recognized at
risk modeling activities, particularly CRM early in our sample pe- fair value under the lower-of-cost-or-fair-value measurement basis.
G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95 83
Fig. 1. Summary of three financial reporting treatments for fair value gains and losses on banks’ various types of financial instruments.
(2014) develop measures of the reliability of recognized fair value 133.4 We denote FVGL that are recorded in net income (other com-
estimates from textual analysis of banks’ and insurers’ financial prehensive income) by NIGL (OCIGL).
statement notes. Chung et al. provide evidence that these measures FAS 107 (1991, ASC 825.10.50) requires firms to disclose the fair
are associated with enhanced market pricing and lower informa- and carrying values of most types of financial instruments in the
tion risk for recognized level 3 fair values. Our study complements notes to financial statements. FAS 107 disclosures are the only in-
Chung et al. by examining an alternative measure of the reliability formation in banks’ financial reports about the fair values of their
of fair value estimates, a longer and more diverse sample period, on-balance sheet loans, deposits (excluding core deposit intangi-
disclosed as well as recognized FVGL, and returns-relevance rather bles), and debt as well as off-balance sheet instruments such as
than value-relevance. loan commitments; FAS 115 requires these and other disclosures
for held-to-maturity securities. We denote FVGL that can only be
calculated from disclosures by DISCGL. FAS 107 also requires firms
to disclose the method(s) and significant assumptions used to es-
Financial report information about estimated fair values of timate the fair value of financial instruments. Banks fulfilled this
financial instruments, literature review, and hypothesis requirement in boilerplate and minimally informative fashions un-
development til the effective date of FAS 157.
FAS 157 defines fair value as cȁthe price that would be received
In this section, we first describe the various bases in account- to sell an asset or paid to transfer a liability in an orderly trans-
ing standards, accounting research, and other sources for our main action between market participants at the measurement date,dȁ
hypotheses about the effects of MRM and CRM on the returns- and it creates a three-level hierarchy of fair value inputs. 5 Level
relevance of FVGL. We then support and state our further hypothe- 1 (highest quality) inputs are quoted prices in active markets for
ses about the distinct conditions under which we expect MRM and the identical item. Level 2 inputs are quoted prices in markets that
CRM to be particularly useful. are not active for the identical item or in active markets for similar
items and most other observable information. Level 3 (lowest qual-
ity) inputs are unobservable reporting-firm-supplied inputs. The
Required information about estimated fair values and FVGL level of a fair value estimate is determined by the level of its low-
est quality significant input. FAS 157, as amended by FASB Staff Po-
Under current U.S. generally accepted accounting principles, sition (FSP) FAS 157-4 (2009, ASC 820.10.50) and Accounting Stan-
firms recognize certain types of financial instruments at fair value dards Update (ASU) 2010-06 (2010, ASC 820.10.50), requires quar-
on the balance sheet. They record FVGL in net income for some of terly disclosures of the amounts of each major category of assets
these instruments and in other comprehensive income otherwise. and liabilities that is recognized at fair value on the balance sheet
Banks’ primary types of financial instruments currently recognized
at fair value with unrealized gains and losses recorded in net in-
4
come are: trading securities under FAS 115 (1993, ASC 320); non- Other standards require certain types of financial instruments to be recognized
accounting-hedge and fair value hedge derivatives as well as fair at fair value at either: (1) inception, e.g., FAS 166 (2009, ASC 860) for retained inter-
ests in securitizations; or (2) the time of (other than temporary) impairment write-
value hedged items under FAS 133 (1998, ASC 815); most other
downs, e.g., FAS 65 (1982, ASC 948.310.35) and SOP 01-6 (2001, ASC 310.10.35) for
trading instruments under industry accounting principles or prac- held-for-sale loans and FAS 115, as amended by FSP FAS 115-2 & FAS 124-2 (2009,
tices; and financial instruments for which the fair value option ASC 320.10.35), for available-for-sale and held-to-maturity securities.
5
is selected under FAS 155 (2006, ASC 815.15) and FAS 159 (2007, Paragraphs C21–C22 of FAS 157 indicates that generally accepted accounting
ASC 825.10). The primary types of financial instruments that cur- principles include various practicability exceptions to fair value measurement that
FAS 157 did not change. While banks do not often invoke these exceptions in their
rently are recognized at fair value with unrealized gains and losses financial reports, the exceptions may have some effect on our FVGL measures or be
recorded in other comprehensive income are available-for-sale se- correlated with our risk modeling measures; e.g., a bank with better risk modeling
curities under FAS 115 and cash-flow-hedge derivatives under FAS may have less need to invoke a practicability exception.
84 G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95
in each of the three levels, separately for items fair valued each Prior empirical research on the differential value-relevance of fair
reporting period versus on a non-recurring basis. For level 2 and values of financial instruments by FAS 157 level and other reliability
level 3 fair values, firms must disclose their valuation techniques disclosures
and inputs. For level 3 fair values, firms must disclose rollforwards
of the fair values from the beginning to the end of the period, dis- Using FAS 157-required disclosures, several studies provide ev-
tinguishing total gains and losses, purchases, sales, issuances, set- idence that banks’ recognized fair values of financial instruments
tlements, and transfers in and out of level 3.6 estimated using higher quality inputs are more value-relevant
(Goh, Li, Ng, & Yong, 2015; Kolev, 2009; Song, Thomas, & Yi, 2010).
These studies all find that level 1 and 2 fair values are more value-
relevant than level 3 fair values. Goh et al. find that level 1 fair
Prior empirical research on the differential value-relevance of fair values are more value-relevant than level 2 fair values. Song et al.
values and returns-relevance of FVGL by type of financial instrument find that level 3 fair values are more value-relevant for banks with
and by recognition versus disclosure better corporate governance.
Chung et al. (2014) develop binary (disclosure made versus not)
An extensive literature beginning with Barth (1994) empirically and continuous (number of words in the disclosure divided by to-
examines the value-relevance of estimated fair values and returns- tal number of words in the Form 10-K filing) measures of FAS 157-
relevance of FVGL on financial instruments. Most of these studies encouraged disclosures in banks’ and insurers’ financial statement
limit their samples to banks because financial instruments dom- notes about the controls, processes, and procedures used to assure
inate banks’ balance sheets, banks’ financial instruments exhibit the reliability of their recognized fair value estimates. Chung et al.
varying levels of liquidity and other factors associated with the (2014) provide evidence that these measures are associated with
need or ability to exercise discretion over fair value estimation, and enhanced market pricing (higher share price and lower priced risk)
analysis of a single industry mitigates concerns about unmodeled and lower information risk (higher analyst consensus) for recog-
heterogeneity. Given that several recent and extensive surveys of nized fair values estimated using significant level 3 inputs.
this literature exist (e.g., Ryan, 2011, Section 4.5), we briefly sum- Collectively, these findings suggest that disclosures indicating
marize the four main findings in these studies that bear on the that banks more reliably estimate the fair value of their financial
hypotheses developed below. instruments are associated with enhanced value-relevance of these
First, for investment securities and other similarly liquid assets, estimates.
fair values are value-relevant and FVGL are returns-relevant, al-
though the returns-relevance results are more sensitive to model Hypotheses
specification and sample (Ahmed & Takeda, 1995; Barth, 1994; Car-
roll, Linsmeier, & Petroni, 2003; Danbolt & Rees, 2008). Second, We discuss banks’ MRM and CRM activities in the introduc-
inconsistent (essentially no) evidence has been generated as to tion and repeat this discussion here only insofar as it pertains di-
whether the fair values of loans and derivatives (other financial rectly to specific hypotheses. All of our hypotheses reflect the view
instruments) are value-relevant (Barth, Beaver, & Landsman, 1996; that risk modeling typically improves fair value estimation. We ac-
Eccher, Ramesh, & Thiagarajan, 1996; Nelson, 1996; Venkatacha- knowledge that risk modeling may instead deteriorate fair value
lam, 1996). Third, banks’ disclosed fair values of financial instru- estimation if it crowds out the appropriate use of judgment or de-
ments appear to be both noisy (i.e., unexplainable based on con- volves into a compliance exercise, as Mikes (2011) and Kaplan and
temporaneous economic events) and managed to make less sol- Mikes (2012) discuss occurred at specific banks.
vent banks appear more so. Surprisingly, the value-relevance of Our first and most general hypothesis is that banks’ MRM and
disclosed fair values does not appear to be affected by noise, but CRM enhance the returns-relevance of their FVGL, particularly for
it is reduced by discretionary behavior (Beaver & Venkatchalam, less liquid financial instruments that exhibit lower quality mar-
2003; Nissim, 2003). Fourth, fair values that are recognized (i.e., ket information (i.e., less cȁprice transparencydȁ) and thus greater
more prominent) are more value-relevant than those that are dis- need for risk modeling to estimate FVGL reliably. This hypothe-
closed (Ahmed, Kilic, & Lobo, 2006; Badertscher, Burks, & Eas- sis is supported by four prior theoretical and empirical literatures:
ton, 2014; Chambers, Linsmeier, Shakespeare, & Sougiannis, 2007; (1) the theoretical literature indicating that market prices respond
Dong, Ryan, & Zhang, 2014; Hirst, Hopkins, & Wahlen, 2004). Col- more strongly to more precise information (e.g., Holthausen & Ver-
lectively, these findings suggest that estimated fair values and recchia, 1988); (2) the empirical literature on the value-relevance
FVGL for which the signal-to-noise ratio is (perceived to be) higher of fair values and returns-relevance of FVGL by type of financial
have higher value-relevance and returns-relevance, respectively. instruments and by recognition versus disclosure discussed above;
We examine returns-relevance rather than value-relevance be- (3) the empirical literature that demonstrates that FAS 157 fair
cause the latter is both more extensively studied in the prior lit- value input level and voluntary fair value reliability disclosures af-
erature and more subject to omitted variables (but less subject fect this value-relevance discussed above; and (4) Bhat, Ryan, and
to measurement error). Barth, Beaver, and Landsman (2001) and Vyas’s (2014) findings that banks’ CRM is positively associated with
Holthausen and Watts (2001) discuss these methodological trade- the timeliness of their loan loss provisions. We formally state this
offs. hypothesis and all subsequent hypotheses in alternative form.
NIGL usually are highly liquid and easy to fair value. Hence, we ex- H5. The effect of CRM is stronger during the financial crisis than
pect Hypothesis H1 to hold most strongly for DISCGL, less strongly in other years.
for OCIGL, and not at all for NIGL.
Our second hypothesis is that banks’ MRM enhances the Measurement of variables and methodology
returns-relevance of their OCIGL, particularly for mortgage-backed
and asset-backed securities. This hypothesis has three interrelated Fair value gains and losses
bases. First, interest rate risk is the primary risk reflected in OCIGL,
which are generated by available-for-sale securities and cash-flow- We measure NIGL as trading revenue reported on line 5.c of
hedge derivatives. Banks’ most common types of available-for-sale Schedule HI and OCIGL as other comprehensive income reported
securities are U.S. Treasuries, other U.S. federal government se- on line 4 of Schedule HC-R of Y-9C filings. Unfortunately, both NIGL
curities, and agency-guaranteed mortgage-backed securities. Their and OCIGL include items other than FVGL; NIGL includes fee in-
most common type of cash-flow-hedge derivatives are interest rate come and realized gains and losses and OCIGL includes other com-
derivatives with highly creditworthy counterparties. These instru- prehensive income from pensions and foreign currency.8 For finan-
ments typically are near credit riskless. Moreover, to the limited cial assets (liabilities) for which fair values are disclosed, we cal-
extent that banks experience credit losses on these instruments, culate DISCGL as (minus) the change during the year of the differ-
they typically record these losses in net income under impairment ence between the fair value and carrying value of the instrument.9
accounting rules. We obtain disclosed fair and carrying values from SNL Financial.
Second, interest rate risk manifests through: (1) discounting, All FVGL components are after taxes calculated using a statutory
which more strongly affects longer duration positions; (2) pre- tax rate of 35% and divided by beginning-of-year market value of
payment of fixed-rate mortgages underlying mortgage-backed and equity.
other asset-backed securities; and (3) the exercise of other interest
rate options, which may be standalone cash-flow-hedge derivatives Unadjusted risk modeling measures
or embedded in available-for-sale securities (Ryan, 2011, Section
2.4). Banks must use considerably more sophisticated MRM (e.g., To construct our MRM and CRM measures, we had to decide
interest rate simulation) to determine the effects of prepayment which risk modeling activities to include in the measures and also
and other interest rate options than to determine discounting ef- the disclosure medium to examine for these activities. To make
fects. Relatedly, it is considerably more difficult for banks to hedge the first decision, we considered all of the market and credit risk
options than discounting effects. modeling activities (including related risk management processes)
Third, compared to banks’ other types of available-for-sale secu- mentioned in a 2001 public disclosure survey of large interna-
rities, mortgage-backed securities and other types of asset-backed tional banks conducted by the Basel Committee on Banking Su-
securities typically are both less liquid and subject to greater pervision.10 To accommodate the U.S. banks in our sample, we ex-
fair valuation difficulties, due to their embedded prepayment cluded risk modeling activities for which disclosures are required
options and complex structuring (e.g., waterfalls). Even agency- for U.S. banks or that are not provided by any sample bank in any
guaranteed mortgage-backed securities generally exhibit low price sample year. This yields five market risk modeling activities (inter-
transparency unless and until specific securities are identified in est rate gap analysis, interest rate sensitivity analysis, Value-at-Risk
to-be-announced trades (Vickrey & Wright, 2013).7 analysis, stress testing, and backtesting) and four credit risk mod-
eling activities (statistical credit risk measurement, credit scoring,
H2. MRM enhances the sensitivity of returns to OCIGL, especially internal credit risk rating, and stress testing). We describe these
OCIGL on less liquid and more difficult-to-fair-value mortgage- activities in detail in the Appendix cȁRisk modeling measures and
backed and asset-backed available-for-sale securities. chief risk officer indicatordȁ.
We choose banks’ annual Form 10-K filings as the sole disclo-
Our third hypothesis is that banks’ CRM primarily impacts the sure medium for three reasons. First, banks provide the most ex-
returns-relevance of their DISCGL. This hypothesis is motivated tensive disclosures of their risk modeling activities in these filings,
by the fact that banks assume credit risk primarily through their primarily due to SEC requirements. For example, SEC FRR 48 re-
funded loans and unfunded loan commitments for which esti- quires that banks make extensive disclosures of their market risk
mated fair values are disclosed (Ryan, 2011, Section 2.5). in these filings. Second, the extents of disclosure in Form 10-K fil-
ings and other media are positively correlated (Lang & Lundholm,
H3. CRM primarily enhances the sensitivity of returns to DISCGL. 1993). Third, our analysis of samples of banks’ financial analyst re-
ports and conference call transcripts yielded no mention of risk
Our fourth and fifth hypotheses are motivated by the intuition
modeling.
that banks’ risk modeling should be more important during peri-
We hand collected banks’ risk modeling disclosures from their
ods when the modeled risk is higher. In particular, MRM (CRM)
Form 10-K filings from 2001 to 2013. We electronically searched
should be more important during years when interest rates are
more variable (credit losses are unexpectedly higher).
8
In untabulated analysis, we tried to remove fee income from NIGL by subtract-
H4. The effect of MRM is stronger during years with above-median ing its average over the prior three or five years. This transformation of NIGL should
mostly remove fee income, which is much more persistent than fair value gains and
interest rate volatility than in other years.
losses. No inferences are affected by this alternative specification of NIGL.
9
Because provisions for credit losses on loans and loan commitments are
recorded in net income, these recognized accruals are not included in DISCGL. This
7 likely reduces the power of our tests of hypotheses H1, H3, and H5, because prior
Vickrey and Wright (2013) provide evidence that to-be-announced agency
mortgage-backed securities trades (which are accounted for as derivatives, not in- research shows that CRM is associated with these accruals (Bhat et al., 2014).
10
vestment securities, until settled) exhibit high volume and price transparency. They While these documents also consider operational risk modeling activities, we
also explain, however, that these trades exhibit various features (e.g., settlement do not deem this type of activity to be directly related to the estimation of fair
at a single date each month and the cheapest-to-deliver option) that reduce the values and FVGL for financial instruments. We acknowledge that Chernobai, Jorion,
price transparency these trades provide for existing pools of agency-guaranteed and Yu (2011) provide evidence that operational risk is positively correlated with
mortgage-backed securities. credit risk.
86 G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95
those filings for a large number of relevant words or phrases, such risk officer. CRO takes a value of one for these bank-year observa-
as cȁrisk management,dȁ cȁrisk model,dȁ cȁscoringdȁ and cȁrisk tions and zero otherwise.13
ratingdȁ. We read each identified disclosure to verify that the risk
modeling activity was disclosed. We construct an indicator for each
activity that takes a value of one if it is disclosed and zero other-
wise. Our unadjusted market risk modeling measure, MRM (credit First-stage models explaining risk modeling
risk modeling measure, CRM), is the simple average of the indica-
tors for the five market risk activities (four credit risk activities). In the two-stage approach, the first-stage models regress each
As averages of indicator variables, MRM and CRM take values from of the unadjusted MRM and CRM measures (collectively denoted
zero to one. XRM) on seven proxies for banks’ discipline over risk modeling,
Several caveats are in order regarding our risk modeling mea- technical sophistication, risk exposures, and risk tolerance, all of
sures. First, we observe banks’ disclosures of risk modeling activ- which we expect primarily to affect the extent of banks’ risk mod-
ities, not the activities themselves. Presumably the disclosure of eling activities rather than their disclosure of those activities. These
risk-modeling activities implies banks engage in the activities, be- proxies are: CRO, a proxy for banks’ discipline over risk mod-
cause there does not appear to be any reason for banks to lie in eling; log of total assets (SIZE) and a trading portfolio indica-
these disclosures, which generally are too terse to divulge mean- tor (TRADD), proxies for banks’ technical sophistication; 0–1 year
ingful proprietary information.11 However, the non-disclosure of an repricing gap (INTSEN), the proportion of assets that are commer-
activity does not imply that the bank does not engage in the activ- cial loans (COMMLOAN), and the standard deviation of fair value
ity. As discussed in the introduction and below, we employ a two- income deflated by beginning market value of equity over the past
stage approach to help ensure that our results are attributable to ten years (FVISTD), proxies for banks’ interest rate risk, credit risk,
banks’ risk modeling activities rather than their disclosure of those and overall risk, respectively; and tier 1 risk-based capital ratio
activities. Second, our risk modeling measures capture the exis- (TIER 1), a proxy for banks’ solvency or risk tolerance. We also in-
tence of particular activities more than the quality of those activi- clude year fixed effects to capture changes in the risk modeling
ties. Third, and relatedly, despite robustness checks to incorporate measures, particularly CRM, over time. This model is:
various different weighting approaches, simple averaging of the in-
dicators for risk modeling activities might not reflect the weights XRM = α + β1CRO + β2 SIZE + β3 T RADD + β4 INT SEN
investors assign to these activities. Lastly, despite careful reading + β5COMMLOAN + β6 FV IST D + β7 T IER1
of each disclosure, our coding of risk modeling activities might be
subjective. The large number of disclosures involved makes coding
+ year fixed effects + ε (1)
by multiple individuals costly, however.
We suppress time subscripts except when necessary for clarity.
We use the portions of the risk modeling measures explained by
as measures of
the seven proxies, denoted with hats, i.e., XRM,
Chief risk officer indicator risk modeling activities. We use the remaining portions of the risk
as measures of risk modeling
modeling measures, i.e., XRM − XRM,
We use an indicator variable for whether banks employ a chief disclosures.14
risk officer, CRO, as our measure of banks’ discipline over risk mod-
eling. We considered using standard proxies for corporate gover-
nance for this purpose, but these proxies do not appear to cor-
respond strongly, if at all, with the estimation of fair values and Models of the impact of risk modeling on the
FVGL.12 Consistent with this view, Aebi, Sabato, and Schmid (2012) returns-relevance of FVGL and hypotheses restated as
find that these standard proxies cȁare mostly insignificantly or coefficient restrictions
even negatively related to the banks’ performance during the cri-
sis,dȁ whereas banks that employ chief risk officers that report di- The returns models using the unadjusted risk modeling mea-
rectly to the board of directors exhibit superior performance dur- sures (i.e., in the one-stage approach) regress share returns for the
ing the crisis. Ellul and Yerramilli (2013) find similar results for 12 months ending four months after the fiscal year end (R)15 on
their risk management index, which depends strongly on the em- net income before FVGL recorded in net income (NIBNIGL), the
ployment and status of the chief risk officer. three types of FVGL (NIGL, OCIGL, and DISCGL), the risk modeling
CRO takes a value of one if the bank discloses that it employs a measure under consideration (either MRM or CRM), interactions
chief risk officer in its Form 10-K filing and zero otherwise. We between that risk modeling measure and each of NIBNIGL and the
searched filings to determine whether the words cȁchiefdȁ and
cȁriskdȁ appear within five words of each other and, if so, whether
the words cȁriskdȁ and either cȁofficerdȁ or cȁdirectordȁ appear
13
within five words of each other. For filings that met both filters, Using data available on banks’ top five officers from SNL Financial, we endeav-
ored to determine whether CROs were one of these officers, but this determination
we read the relevant passages to confirm the existence of a chief
cannot be made from these data.
14
In untabulated robustness analyses, we use an alternative first stage approach
that does not alter any inference in the second stage. Specifically, we include firm
attributes, that the disclosure literature surveyed by Healy and Palepu (2001) finds
11
In addition, disclosures of risk modeling are subject to internal controls, to CEO to be associated with disclosure, as additional explanatory variables in Eq. (1): an
and CFO certification of financial reports under Section 302 of the Sarbanes–Oxley indicator for issuance of debt or equity, i.e., capital market transactions that mo-
Act, and to audit (reading for consistency) by banks’ auditors if the disclosures tivate disclosure; an indicator for negative net income, a proxy for performance;
appear in the financial statement notes (MD&A) sections of financial reports. The the book-to-market ratio, a proxy for growth; number of analysts forecasting earn-
observed frequency of disclosures is fairly low, inconsistent with easily imitated ings and percentage institutional ownership, proxies for information environment.
cȁcheap talk.dȁ We use the portions of the risk modeling measures explained by these proxies as
12 measures of risk modeling disclosures.
More generally, Larcker, Richardson, and Tuna (2007) argue that construct valid-
15
ity issues with corporate governance measures cause mixed and hard-to-interpret We use this return window because some sample banks are non-accelerated
results in many studies, in part due to the absence of theory indicating which cor- filers that file their Form 10-K filings on the last day of the third month following
porate governance variables matter in what contexts. the end of the fiscal year.
G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95 87
three types of FVGL, and year fixed effects.16 These models are: required to file Y-9Cs increased from $150 million to $500 million
in 2006; we impose the latter size restriction for all bank-year ob-
R = a + b1 NIBNIGL + b2 NIGL + b3 OCIGL + b4 DISCGL servations to yield a more comparable sample through time. Some
+b5 (NIBNIGL ∗ XRM ) + b6 (NIGL ∗ XRM ) + b7 (OCIGL ∗ XRM ) bank holding companies that file Y-9Cs differ slightly from the cor-
+b8 (DISCGL ∗ XRM ) + b9 XRM + year fixed effects + e (2A) responding public companies that file Form 10-Ks.
Table 1, Panel A reports the construction of the final sample of
The returns models using the risk modeling activity and disclosure 238 unique banks and 2413 bank-year observations for the years
measures from the first-stage model Eq. (1) (i.e., in the two-stage 2002–2013. To mitigate the effects of outliers, we winsorize all
and XRM − XRM
approach) follow directly from substituting XRM continuous model variables at the 0.5% and 99.5% levels of their
for XRM in Eq. (2A). These models are: distributions.
Table 1
Sample selection and descriptive statistics.
#Bank-years #Banks
Panel B: Summary statistics for 238 banks from 2002 to 2013 (2413 bank-years)
R 0.060 0.326 −0.097 0.058 0.224
NIBNIGL 0.004 0.315 0.049 0.067 0.081
NIGL 0.001 0.007 0.000 0.000 0.000
OCIGL −0.001 0.036 −0.011 −0.001 0.009
DISCGL −0.001 0.224 −0.034 0.000 0.034
MRM 0.314 0.145 0.200 0.400 0.400
CRM 0.211 0.209 0.000 0.250 0.250
CRO 0.136 0.342 0.000 0.000 0.000
SIZE 14.730 1.531 13.655 14.342 15.388
TRADD 0.218 0.413 0.000 0.000 0.000
INTSEN 1.699 1.987 0.528 1.174 2.075
COMMLOAN 0.121 0.074 0.066 0.105 0.160
FVISTD 0.269 0.338 0.079 0.139 0.312
TIER1 12.328 2.923 10.310 11.870 13.880
Year 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
MRM components
INTGAP 135 149 147 155 154 148 143 142 131 113 110 104
IRSIMUL 131 146 159 177 176 176 176 182 170 157 156 150
VAR 8 7 8 8 9 8 7 7 7 8 10 10
MRSTRESS 1 1 1 3 3 5 5 5 5 9 9 9
MRBACKTEST 3 3 2 2 3 3 4 5 6 8 8 10
CRM components
MODEL 9 13 16 22 23 22 24 35 48 48 45 45
SCORING 13 17 18 17 19 21 18 22 23 24 27 25
RR 53 71 79 91 91 93 97 113 158 147 144 136
CRSTRESS 2 1 2 3 4 5 12 20 20 32 35 32
Notes: Panel A reports the sample selection process. Panel B reports summary statistics for the primary model variables, including net income before fair value gains
and losses recorded in net income (NIBNIGL), fair value gains and losses recorded in net income (NIGL), fair value gains and losses recorded in other comprehensive
income (OCIGL), disclosed fair value gains and losses (DISCGL), the risk modeling measures (MRM, and CRM), chief risk officer (CRO), and other bank characteristics.
Panel C reports the means of the risk modeling measures and CRO by year, as well as and the number of banks that disclose of the components of each risk modeling
measure by year. All variables are defined in the Appendix cȁDefinitions of variables and other acronymsdȁ except for the risk modeling components which are defined
in the Appendix cȁRisk modeling measures and chief risk officer indicatordȁ.
Empirical analysis models. These results indicate that banks’ discipline over fair value
estimation and technical sophistication explain their risk modeling.
First-stage regression results In the MRM model, the coefficient on INTSEN is significantly neg-
ative at the 5% level and the coefficient on FVISTD is significantly
Table 3 reports the estimation of Eq. (1), the first-stage models negative at the 10% level, suggesting that banks with higher inter-
used to explain the unadjusted risk modeling measures in terms est rate risk tend to seek out or tolerate rather than manage that
of risk-related explanatory variables and fixed year effects (untab- risk. In contrast, the coefficient on FVISTD is positive at the 10%
ulated). The first (second) column presents the results for the un- level in the CRM model. No other explanatory variable is signifi-
adjusted market (credit) risk modeling measure MRM (CRM). The cant in any model.
fit is considerably better for the CRM model, primarily because the
year fixed effects are more significant in this model due to CRM’s Primary regression results
strong upward trend over time reported in Table 1, Panel C.
The coefficient on CRO is significantly positive at the 5% level Table 4, Panel A reports the estimations of Eqs. (2A) and (2B),
in the MRM model and at the 10% level in the CRM model. The which we use to test hypotheses H1–H5 about whether and how
coefficient on SIZE is significantly positive at the 1% level in both banks’ risk modeling impacts the returns-relevance of their three
G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95 89
Table 2
Correlations.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
R (1) 0.485 0.069 −0.085 −0.026 0.009 0.071 0.012 −0.016 0.018 0.113 0.016 −0.104 0.195
0.000 0.001 0.000 0.194 0.675 0.001 0.572 0.435 0.372 0.000 0.440 0.000 0.000
NIBNIGL (2) 0.201 −0.076 −0.049 0.018 −0.050 −0.011 −0.019 −0.100 −0.052 −0.018 −0.048 −0.132 0.214
0.000 0.000 0.016 0.385 0.014 0.575 0.354 0.000 0.011 0.375 0.019 0.000 0.000
NIGL (3) 0.106 −0.068 −0.004 0.020 0.077 0.209 0.202 0.400 0.669 0.177 0.182 −0.098 −0.097
0.000 0.001 0.836 0.316 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
OCIGL (4) 0.044 −0.079 0.067 0.067 −0.003 −0.001 −0.020 −0.016 −0.005 0.038 0.004 0.016 −0.003
0.031 0.000 0.001 0.001 0.894 0.968 0.317 0.445 0.815 0.064 0.846 0.445 0.900
DISCGL (5) 0.017 −0.096 0.039 0.073 −0.001 0.033 0.018 0.013 0.031 0.037 0.018 0.000 0.003
0.416 0.000 0.054 0.000 0.963 0.104 0.383 0.512 0.128 0.066 0.382 0.984 0.881
MRM (6) 0.009 0.024 0.303 0.002 0.005 0.173 0.074 0.094 0.076 −0.011 0.051 −0.010 −0.045
0.648 0.239 0.000 0.941 0.793 0.000 0.000 0.000 0.000 0.603 0.012 0.611 0.028
CRM (7) 0.066 −0.029 0.254 0.003 0.038 0.209 0.229 0.330 0.240 0.144 0.030 −0.009 0.085
0.001 0.161 0.000 0.888 0.062 0.000 0.000 0.000 0.000 0.000 0.147 0.651 0.000
CRO (8) 0.013 −0.021 0.227 −0.009 0.020 0.131 0.261 0.364 0.237 0.141 0.043 0.011 0.004
0.510 0.302 0.000 0.666 0.320 0.000 0.000 0.000 0.000 0.000 0.034 0.592 0.839
SIZE (9) −0.004 0.011 0.459 0.007 0.002 0.231 0.409 0.452 0.499 0.184 0.201 −0.128 −0.102
0.826 0.597 0.000 0.727 0.921 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
TRADD (10) 0.020 0.012 0.388 0.001 0.021 0.136 0.263 0.237 0.593 0.177 0.199 −0.095 −0.137
0.315 0.566 0.000 0.953 0.304 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
INTSEN (11) 0.170 −0.360 0.176 0.084 0.066 −0.044 0.122 0.123 0.140 0.134 0.156 0.171 −0.114
0.000 0.000 0.000 0.000 0.001 0.031 0.000 0.000 0.000 0.000 0.000 0.000 0.000
COMMLOAN (12) 0.030 0.072 0.028 −0.007 −0.002 0.005 0.009 0.025 0.118 0.183 0.064 −0.053 −0.225
0.141 0.000 0.176 0.744 0.905 0.811 0.649 0.215 0.000 0.000 0.002 0.009 0.000
FVISTD (13) −0.150 −0.352 −0.057 0.026 −0.034 −0.067 −0.011 0.021 −0.070 −0.056 0.209 −0.098 −0.209
0.000 0.000 0.005 0.200 0.099 0.001 0.596 0.292 0.001 0.006 0.000 0.000 0.000
TIER1 (14) 0.192 0.252 −0.047 −0.020 0.022 −0.010 0.061 −0.011 −0.134 −0.142 −0.148 −0.188 −0.219
0.000 0.000 0.020 0.315 0.280 0.616 0.003 0.577 0.000 0.000 0.000 0.000 0.000
Notes: The table reports the Pearson (below the diagonal) and Spearman (above the diagonal) correlations of the primary model variables for 2413 bank-year obser-
vations for 238 banks from 2002 to 2013. Variables are defined in the Appendix cȁDefinitions of variables and other acronymsdȁ. p values are in italics below the
corresponding correlation.
Table 3 ket [credit] risk modeling measure MRM [CRM]. Column (3) [(5)]
First-stage regressions of risk modeling measures on risk management variables.
presents the results for Eq. (2B) using the market [credit] risk mod-
XRM MRM CRM eling activity and disclosure measures MRM and MRM − MRM
Variables (1) (2)
[CRM and CRM − CRM].
CRO 0.022∗∗ 0.022∗ In the benchmark model reported in column (1), the model fit
(0.020) (0.066) is good with an R 2 of 41.8%. Such a high R 2 for a returns model
SIZE 0.021∗∗∗ 0.046∗∗∗ obtains in large part from the substantial variation in bank indus-
(0.000) (0.000)
try performance across the sample years discussed earlier, which
TRADD 0.004 0.012
(0.652) (0.270) is captured by the fixed year effects. The coefficient on NIBNIGL is
INTSEN −0.004∗∗ −0.003 significantly positive at the 1% level but well below one at 0.249,
(0.021) (0.196) suggesting the presence of considerable noise in this variable. The
COMMLOAN −0.054 −0.016 coefficient on NIGL is significantly positive at the 5% level and well
(0.182) (0.757)
FVISTD −0.017∗ 0.022∗
above one at 3.297, indicating that it includes a permanent com-
(0.053) (0.059) ponent. This reflects the inclusion of some fee revenue in trading
TIER1 0.001 −0.001 revenue on the Y-9C filings and the persistence of FVGL from most
(0.347) (0.696) dealer and proprietary trading activities due to the incorporation
Constant −0.012 −0.357∗∗∗
of trading spreads (i.e., day-one profits). The coefficient on OCIGL
(0.781) (0.000)
Year effects Included Included is significantly positive at the 5% level and somewhat below one
Observations 2413 2413 at 0.742, consistent with it primarily capturing transitory income
R-squared 0.068 0.273 items. The coefficient on DISCGL is insignificant and close to zero,
Notes: This table reports the estimation of Eq. (1), which regresses of the risk mod- consistent with prior research documenting little or no returns rel-
eling measures (MRM and CRM, collectively denoted XRM) on risk management evance for FVGL on less liquid financial instruments discussed in
variables and year effects. Variables are defined in the Appendix cȁDefinitions of section ‘Financial report information about estimated fair values
variables and other acronymsdȁ. All variables are winsorized at the 0.5% and 99.5% of financial instruments, literature review, and hypothesis devel-
levels of their distribution. Standard errors are calculated clustering observations by
bank. P-values are in parentheses below the corresponding coefficients.
opment’.
∗
Denotes significance at the 10% level. The coefficients on the test variables reported in columns (2)–
∗∗
Denotes significance at the 5% level. (5) of the panel are with one explainable exception consistent with
∗∗∗
Denotes significance at the 1% level. our main hypotheses H1–H3. The coefficients on the interactions
of MRM and MRM with OCIGL in columns (2) and (3), respec-
tively, are both significantly positive at the 10% level, consistent
FVGL components. To link to prior research and provide a bench-
with Hypothesis H2. The coefficients on the interactions of MRM
mark for the main results, column (1) of the panel reports the
with DISCGL in these columns are both insignificant,
and MRM
results for the model including only NIBNIGL, the three FVGL
components, and the fixed year effects (untabulated). Column (2) however, inconsistent with Hypothesis H1. These insignificant coef-
[(4)] presents the results for Eq. (2A) using the unadjusted mar- ficients can be explained by low power resulting from much of the
90 G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95
overall sample period exhibiting stable interest rates and the large Table 4 (continued)
set of financial instruments (e.g., loans and deposits) reflected in Variables (1) (2)
DISCGL likely exhibiting considerable cross-hedging of interest rate
Panel B: Expanded regression of returns on FVGL, risk modeling measures, and
risk. This explanation is supported by results reported in Table 5 interactions including M/ABSDUM interactions
discussed below, which show that the coefficients on these interac- OCIGL 2.358∗∗∗ 0.215
tions are significantly positive during periods of high interest rate (0.002) (0.933)
volatility. OCIGL ∗ M/ABSDUM −1.678∗ −2.073
(0.084) (0.588)
The coefficients on the interactions of CRM and CRM with DIS-
OCIGL ∗ MRM 6.221
CGL in columns (4) and (5), respectively, are both significantly pos- (0.113)
itive at the 5% level, consistent with Hypotheses H1 and H3. The OCIGL ∗ MRM ∗ M/ABSDUM 6.927∗∗
coefficients on the interactions of CRM and CRM with OCIGL are (0.031)
OCIGL ∗ MRM 1.842
insignificant, as suggested by Hypothesis H3. These results sug- (0.814)
gest that DISCGL reflects banks’ primary assumption of credit risk ∗ M/ABSDUM
OCIGL ∗ MRM 7.220∗
through funded loans and unfunded loan commitments. (0.054)
Table 4, Panel A generally reports insignificant coefficients on ∗ M/ABSDUM
OCIGL ∗ (MRM − MRM) −5.873
the interactions of the risk modeling measures with the FVGL vari- (0.113)
∗ M/ABSDUM
OCIGL ∗ (MRM − MRM) 3.924
ables about which no hypotheses are made. The coefficients on the
(0.260)
CRM, and CRM
interactions of MRM, MRM, with NIGL are all in-
R-squared 0.427 0.439
significant. The coefficients on the interactions of the risk modeling
and CRM − CRM with all three Notes: Column 1 of Panel A reports the estimation of a restricted version of Eq. (2A)
disclosure measures MRM − MRM that regresses returns for the 12 months ending 4 months after the fiscal year end
types of FVGL are all insignificant. Hence, the act of disclosing risk (R) on net income before FVGL recorded in net income (NIBNIGL) and the three
types of FVGL (NIGL, OCIGL, and DISCGL) and includes fixed year effects. Column 2
(4) reports the estimation of Eq. (2A), which includes one of the unadjusted MRM
Table 4 and CRM measures described in the Appendix cȁRisk modeling measures and chief
Regression of returns on FVGL, risk modeling measures, and interactions. risk officer indicatordȁ and interactions involving that measure. Column 3 (5) re-
ports the estimation of Eq. (2B), which includes one of the MRM and CRM risk
XRM MRM MRM CRM CRM modeling activity measures (MRM or CRM, collectively denoted XRM),
the corre-
Variables (1) (2) (3) (4) (5) sponding risk modeling disclosure measure (MRM − MRM or CRM − CRM), and in-
teractions involving these measures. The MRM and CRM activity and disclosure
Panel A: Regression of returns on FVGL, risk modeling measures, and interactions
measures are the predicted values and residuals, respectively, from the estimation
NIBNIGL 0.249∗∗∗ 0.255∗∗∗ −0.206 0.235∗∗∗ 0.583
of the corresponding first-stage model reported in Table 3. Panel B reports the re-
(0.000) (0.000) (0.456) (0.000) (0.229)
sults of expanded versions of the regression reported in columns (2) and (3) of
NIGL 3.297∗∗ 10.695 ∗∗∗
32.875∗∗∗ 7.692 ∗∗
27.269∗∗
Panel A that include interactions of M/ABSDUM, an indicator for an above-median
(0.024) (0.001) (0.006) (0.043) (0.022)
proportion of available-for-sale securities that are mortgage-backed or asset-backed
OCIGL 0.742∗∗ 1.649∗∗∗ −0.311 0.501 −1.733
securities, with the variables involving OCIGL. Variables are defined in the Appendix
(0.021) (0.002) (0.884) (0.170) (0.427)
cȁDefinitions of variables and other acronymsdȁ. All variables are winsorized at the
DISCGL 0.043 0.075 −0.327 −0.025 −1.165∗∗
0.5% and 99.5% levels of their distribution. Standard errors are calculated clustering
(0.274) (0.453) (0.404) (0.654) (0.016)
observations by bank. P-values are in parentheses under the corresponding coeffi-
NIBNIGL ∗ XRM −0.008 0.095
cients.
(0.968) (0.680) ∗
Denotes significance at the 10% level.
NIGL ∗ XRM −11.562 −8.176 ∗∗
Denotes significance at the 5% level.
(0.103) (0.146) ∗∗∗
Denotes significance at the 1% level.
OCIGL ∗ XRM 3.194∗ 1.044
(0.059) (0.305)
DISCGL ∗ XRM −0.125 0.308∗∗
(0.678) (0.027)
XRM −0.026 0.011
(0.431) (0.657) modeling activities does not appear to affect the returns-relevance
NIBNIGL ∗ XRM 1.671∗ −0.503 of FVGL.
(0.081) (0.492)
In summary, the findings reported in Table 4, Panel A indi-
NIGL ∗ XRM −64.909 −25.949
(0.122) (0.143) cate that MRM activities enhance the returns-relevance of OCIGL,
OCIGL ∗ XRM 3.850∗ 3.686 which primarily reflect interest rate risk, consistent with Hypothe-
(0.064) (0.265) sis H2 but not Hypothesis H1. CRM activities enhance the returns-
DISCGL ∗ XRM 1.258 1.791∗∗ relevance of DISCGL, which reflect banks’ primary assumption of
(0.342) (0.014) credit risk through funded loans and unfunded loan commitments,
XRM −0.697∗∗∗ −0.253∗∗∗
consistent with Hypotheses H1 and H3.
(0.000) (0.003)
NIBNIGL ∗ (XRM − XRM) −0.139 −0.018
(0.629) (0.950) Regression results with interactions for potentially illiquid and
NIGL ∗ (XRM − XRM) −1.715 5.709
difficult-to-value available-for-sale securities
(0.731) (0.341)
OCIGL ∗ (XRM − XRM) −3.647 1.569
(0.125) (0.190) Hypothesis H2 is based on the assumption that banks do
DISCGL ∗ (XRM − XRM) −0.650 0.201 not determine OCIGL directly from market prices, obviating their
(0.147) (0.421) need to conduct MRM to make this determination. As discussed
(XRM − XRM) −0.012 0.036 in section ‘Financial report information about estimated fair val-
(0.735) (0.218)
ues of financial instruments, literature review, and hypothesis
Constant 0.297∗∗∗ 0.300∗∗∗ 0.522∗∗∗ 0.294∗∗∗ 0.467∗∗∗
(0.000) (0.000) (0.000) (0.000) (0.000)
development’, mortgage-backed and asset-backed securities typi-
Year effects Included Included Included Included Included cally are illiquid and include embedded interest-rate options and
Observations 2413 2413 2413 2413 2413 other interest-rate-related structuring that require MRM to accu-
R-squared 0.418 0.424 0.437 0.421 0.431 rately estimate the fair value of the securities. Reflecting this fact,
(Continued on next column) the hypothesis states it should hold more strongly for mortgage-
G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95 91
backed and asset-backed securities than for banks’ other primary omit fixed year effects in these models because both indicator vari-
types available-for-sale securities, such as U.S. Treasuries. To test ables partition on time.
this part of the hypothesis, we estimate expanded versions of Column (1) of Table 5 presents the results for the model in
Eqs. (2A) and (2B) that include interactions of an indicator vari- which UNSTABLE is interacted with the interactive variables in-
able for bank-year observations with above median percentage volving MRM. Consistent with Hypothesis H4, the coefficients on
of mortgage-backed and asset-backed securities in their available- OCIGL ∗ MRM ∗ UNSTABLE and DISCGL ∗ MRM ∗ UNSTABLE are sig-
for-sale securities portfolios in that year, denoted M/ABSDUM, nificantly positive at the 5% and 1% levels, respectively. The latter
with all variables involving OCIGL. We expect the coefficient on result is related to our explanation for the unexpectedly insignif-
the interaction of OCIGL with MRM to be more positive and icant coefficients on the interactions of DISCGL with MRM and
significant for observations for which M/ABSDUM takes a value in Table 4, Panel A discussed above. In contrast, the coeffi-
MRM
of one. cients on all other interactions involving UNSTABLE are insignifi-
To conserve space, Table 4, Panel B reports the estimated co- cant.
efficients only for the critical variables in these expanded models Column (2) of the table presents similar results for the model
and also the R 2 s; all omitted coefficients are essentially identi- where UNSTABLE is interacted with the interactive variables in-
cal to those reported in Panel A. Column (1) [(2)] of the panel re- and MRM-MRM.
volving MRM Again consistent with Hypoth-
ports the results for the expanded versions of Eq. (2A) [(2B)]. In esis H4, the coefficients on OCIGL ∗ MRM ∗ UNSTABLE and
column (1), the coefficient on the interaction of OCIGL and MRM is
DISCGL ∗ MRM ∗ UNSTABLE are both significantly positive at the
insignificantly positive, consistent with MRM not enhancing the es- 1% levels, while the coefficients on OCIGL ∗ (MRM − MRM) ∗ UN-
timation of the fair values of available-for-sale securities other than STABLE and DISCGL ∗ (MRM − MRM) ∗ UNSTABLE are both in-
mortgage-backed and asset-backed securities. The coefficient on significant. In contrast, the coefficients on all other interactions in-
the interaction of OCIGL and M/ABSDUM is significantly negative volving UNSTABLE are insignificant except for the significantly neg-
at the 10% level, consistent with fair valuation estimation posing ∗ UNSTABLE.
ative coefficient on NIBNIGL ∗ MRM
greater difficulties for mortgage-backed and asset-backed securities
Column (3) presents the results for the model where CRISIS
than for other available-for-sale securities. Consistent with Hypoth-
is interacted with the interactive variables involving CRM. Consis-
esis H2 that MRM enhances the fair value estimation of mortgage-
tent with Hypothesis H5, the coefficient on DISCGL ∗ CRM ∗ CRI-
backed and asset-backed securities, the coefficient on the interac-
SIS is significantly positive at the 10% level. In addition, the coef-
tion of OCIGL, MRM, and M/ABSDUM is significantly positive at the
ficients on NIGL ∗ CRM ∗ CRISIS and OCIGL ∗ CRM ∗ CRISIS are sig-
5% level.
nificantly positive, likely reflecting the pervasive effect of credit
Column (2) yields similar inferences. In particular, the co-
risk on banks’ various sources of net income during the financial
efficient on the interaction of OCIGL, MRM, and M/ABSDUM
crisis.
is significantly positive at the 10% level and the coefficient Column (4) presents similar results for the model where CRI-
on the interaction of OCIGL, MRM − MRM, and M/ABSDUM is
SIS is interacted with the interactive variables involving CRM
insignificant.
and CRM − CRM. Consistent with Hypothesis H5, the coefficient
on DISCGL ∗ CRM ∗ CRISIS is significantly positive at the 10%
Regression results with interactions for macroeconomic conditions level, while the coefficients on the interactions of CRISIS with
all variables involving CRM − CRM are insignificant. In contrast,
Hypothesis H4 predicts that MRM activities are more impor- the coefficients on all other interactions involving UNSTABLE are
tant in periods with more volatile interest rates. To test this hy- insignificant except for the significantly negative coefficient on
pothesis, we expand the models reported in columns (2) and (3) NIBNIGL ∗ CRM ∗ UNSTABLE and significantly positive coefficient
of Table 4, Panel A to include an indicator variable for years with on OCIGL ∗ CRM ∗ UNSTABLE.
above-median interest rate volatility that is interacted with the in- In summary, the unadjusted risk modeling measures and risk
and MRM − MRM
teractions of MRM, MRM, with NIBNIGL and the modeling activity measures generally are significant for economic
three types of FVGL. Similarly, Hypothesis H5 predicts that CRM conditions for which they are expected to be important and usually
activities are more important in periods with higher and more un- are insignificant otherwise. The risk modeling disclosure measures
certain credit losses. To test this hypothesis, we expand the mod- generally are insignificant regardless of the economic conditions.
els reported in columns (4) and (5) of Table 4, Panel A to include
an indicator variable for the financial crisis years that is interacted Changes specification
and CRM − CRM
with the interactions of CRM, CRM, with NIBNIGL
and the three types of FVGL. To help ensure that our primary second-stage results reported
We measure interest rate volatility during a year as the stan- in Table 4, Panel A are robust to omitted variables, and also as an
dard deviation of the daily yield on seven-year U.S. Treasuries. UN- alternative to the inclusion of year fixed effects as a means to cap-
STABLE takes a value of one for a year with above-median interest ture the trend in CRM over time reported in Table 1, Panel C, we
rate volatility and zero otherwise. CRISIS takes a value of one for estimate changes models. These models are identical to the mod-
the years 2007–2009, and zero otherwise, because the subprime els reported in columns (2)–(5) of Table 4, Panel A except that
crisis began in February 2007 (Ryan, 2008) and banks’ loan loss MRM − MRM,
MRM, MRM, CRM, CRM, and CRM − CRM are re-
provisions peaked in the third quarter of 2009. placed with the changes in the respective variables and the year
Table 5 presents the results of the estimations of the four ex- fixed effects are dropped.
panded models; columns (1) and (2) reporting the results for the Table 6 reports the estimation of these changes models, with
models including the indicator variable UNSTABLE and columns (3) columns (1) and (2) corresponding to columns (2) and (3), respec-
and (4) reporting the results for the models including the indicator tively, of Table 4, Panel A. The results in column (1) of Table 6
CRISIS. In the table, these indicator variables are generically de- are similar with those reported in column (2) of Table 4, Panel A.
noted DUMMY and the column headings specify which indicator The coefficient on the interaction of MRM with OCIGL remains
DUMMY represents. To conserve space, we discuss only the coef- significantly positive at the 10% level, consistent with Hypothesis
ficients on the interactive variables involving these indicators. We H2. The results in column (2) of Table 6 are similar to those re-
92 G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95
Table 5 (continued)
Table 5
XRM MRM CRM
Regression of returns on risk modeling measures, FVGL, and interactions impact of
DUMMY Unstable Crisis
macroeconomic conditions.
Variables (1) (2) (3) (4)
XRM MRM CRM
DUMMY Unstable Crisis ∗ DUMMY
(XRM − XRM) −0.042 −0.102
Variables (1) (2) (3) (4) (0.782) (0.254)
∗∗∗
DUMMY −0.022 −0.158 −0.244 −0.453∗∗∗
NIBNIGL 0.179∗∗ −0.808∗∗ 0.099 −0.686 (0.402) (0.176) (0.000) (0.003)
(0.015) (0.016) (0.140) (0.195) Constant 0.067∗∗∗ 0.379∗∗∗ 0.110∗∗∗ 0.413∗∗∗
NIGL 17.018∗∗∗ 43.237∗∗∗ 6.563∗ 25.126∗∗ (0.000) (0.000) (0.000) (0.000)
(0.000) (0.001) (0.073) (0.050) Year effects Excluded Excluded Excluded Excluded
OCIGL 0.995 −1.320 0.526 0.523 Observations 2413 2413 2413 2413
(0.126) (0.627) (0.114) (0.838) R-squared 0.086 0.117 0.205 0.232
DISCGL 0.133 −0.321 −0.086 −1.166∗∗
Notes: Columns (1) and (2) [(3) and (4)] of this table reports the results of expand-
(0.269) (0.560) (0.187) (0.039)
ing the regression reported in the columns (2) and (3) [(4) and (5)] of Table 4,
NIBNIGL ∗ XRM 0.965∗ 0.983∗∗∗
Panel A to include interactions of a dummy variable that takes a value of 1 for
(0.067) (0.000)
years with above-median unstable interest rates [the financial crisis years 2007–
NIGL ∗ XRM −16.638 −10.543
2009] with the interactions of NIBNIGL and the three types of FVGL with MRM,
(0.130) (0.158) and (MRM-MRM), [CRM, CRM and (CRM-CRM)] respectively. Variables are de-
MRM
OCIGL ∗ XRM −6.765 −2.383
fined in the Appendix cȁDefinitions of variables and other acronymsdȁ. All variables
(0.117) (0.234)
are winsorized at the 0.5% and 99.5% levels of their distribution. Standard errors are
DISCGL ∗ XRM 1.187∗ 0.158
calculating clustering observations by bank. P-values are in parentheses under the
(0.094) (0.522)
corresponding coefficients.
XRM −0.098∗ −0.010 ∗
Denotes significance at the 10% level.
(0.077) (0.690) ∗∗
Denotes significance at the 5% level.
NIBNIGL ∗ XRM ∗ DUMMY −0.879 −1.454 ∗∗∗
Denotes significance at the 1% level.
(0.199) (0.109)
NIGL ∗ XRM ∗ DUMMY −2.251 26.385∗∗∗
(0.492) (0.001)
OCIGL ∗ XRM ∗ DUMMY 5.582∗∗ 9.189∗∗∗
(0.025) (0.003)
DISCGL ∗ XRM ∗ DUMMY 1.080∗∗∗ 0.850∗ ported in column (3) of Table 4, Panel A. The coefficient on the
(0.003) (0.064) with OCIGL becomes more significantly pos-
XRM ∗ DUMMY 0.079 0.011
interaction of MRM
(0.361) (0.889) itive at the 5% level, consistent with Hypothesis H2, and the coeffi-
NIBNIGL ∗ XRM 5.207∗∗∗ 1.734∗∗ cient on the interaction of (MRM-MRM) with OCIGL remains in-
(0.000) (0.034) significant. Columns (3) and (4) of Table 6 correspond to columns
NIGL ∗ XRM −8.767 −2.431 (4) and (5), respectively, of Table 4, Panel A. The results in col-
(0.108) (0.079)
umn (3) of Table 6 are similar with those reported in column (4)
OCIGL ∗ XRM −2.030 −1.239
(0.816) (0.752) of Table 4, Panel A. The coefficient on the interaction of CRM
DISCGL ∗ XRM 0.215 1.670∗ with DISCGL remains significantly positive at the 5% level, con-
(0.907) (0.050) sistent with Hypotheses H1 and H3. The results in column (4) of
XRM −1.187∗∗∗ −0.477∗∗∗ Table 6 are similar with those reported in column (5) of Table 4,
(0.000) (0.000)
Panel A. The coefficient on the interaction of CRM with DISCGL
∗ DUMMY
NIBNIGL ∗ XRM −1.587∗∗ −0.689∗∗∗
remains significantly positive, albeit at the lower 10% level, consis-
(0.040) (0.000)
∗ DUMMY
NIGL ∗ XRM 9.819 12.825
tent with Hypotheses H1 and H3. The coefficient on the interaction
(0.352) (0.108) of (CRM − CRM) with DISCGL remains insignificant.
∗ DUMMY
OCIGL ∗ XRM 8.692∗∗∗ 4.100∗∗∗
(0.002) (0.000)
∗ DUMMY Summary and conclusion
DISCGL ∗ XRM 1.238∗∗∗ 0.271∗
(0.005) (0.074)
∗ DUMMY
XRM 0.584 0.323 We provide evidence that banks’ market and credit risk model-
(0.109) (0.162) ing (MRM and CRM, respectively) enhance the returns-relevance of
NIBNIGL ∗ (XRM − XRM) −0.855 0.308 banks’ estimated unrealized fair value gains and losses (FVGL) for
(0.590) (0.298) financial instruments. We employ both a one-stage approach that
NIGL ∗ (XRM − XRM) 8.155 0.823
uses unadjusted MRM and CRM measures developed from banks’
(0.251) (0.888)
financial report disclosures and a two-stage approach that parti-
OCIGL ∗ (XRM − XRM) 7.239 −1.184
(0.268) (0.340) tions these unadjusted measures into measures of banks’ risk mod-
DISCGL ∗ (XRM − XRM) −0.349 0.395 eling activities and disclosures of those activities. We predict and
(0.559) (0.191) find that banks’ MRM and CRM activities, but not disclosures of
(XRM − XRM) 0.047 0.033 these activities, enhance the returns-relevance of their FVGL, more
(0.717) (0.324) so for less liquid instruments for which FVGL typically are calcula-
∗ DUMMY
NIBNIGL ∗ (XRM − XRM) 0.734 −0.273
ble from disclosures rather than recorded in net income or other
(0.651) (0.634)
∗ DUMMY
NIGL ∗ (XRM − XRM) −31.947 8.593
comprehensive income. We further predict and find that banks’
(0.128) (0.833) MRM activities enhance the returns-relevance of FVGL recorded in
∗ DUMMY
OCIGL ∗ (XRM − XRM) −8.172 4.763 other comprehensive income, which primarily result from interest
(0.229) (0.266) rate movements, especially when these FVGL are for potentially
∗ DUMMY
DISCGL ∗ (XRM − XRM) −0.741 −0.661 illiquid and difficult-to-value mortgage-backed and asset-backed
(0.401) (0.259)
available-for-sale securities. We predict and find that banks’ CRM
(continued on next page) activities enhance the returns-relevance of their disclosed FVGL,
because banks assume credit risk primarily through their funded
G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95 93
Table 6 across our sample period. Finally, our results based on a U.S. bank
Regression of Returns on Risk Modeling Measures, FVGL, and Interactions
sample may not generalize to other industries or countries.
Changes Specification.
Despite these caveats, we believe our results have relevance for
XRM MRM CRM accounting standard setters, auditors, and other parties involved
Variables (1) (2) (3) (4) in requiring or evaluating estimates of the fair values of finan-
cial instruments and related disclosures in banks’ financial reports.
NIBNIGL 0.148∗∗∗ 0.189∗∗∗ 0.139∗∗∗ 0.174∗∗∗
Reflecting strong political pressure arose during the financial cri-
(0.000) (0.000) (0.000) (0.000)
NIGL 7.636∗∗∗ 6.936∗∗∗ 7.405∗∗∗ 7.106∗∗∗ sis, a number of the FASB’s and IASB’s recent proposals indicate
(0.000) (0.001) (0.001) (0.001) retrenchment in what they deem feasible or desirable regarding
OCIGL 0.099 0.073 0.032 0.086 the extent of fair value accounting for financial instruments.18 Our
(0.623) (0.713) (0.873) (0.662) findings suggest that the returns-relevance of FVGL for financial in-
DISCGL 0.033 0.028 0.033 0.024
(0.291) (0.334) (0.286) (0.416)
struments is enhanced by banks’ risk modeling.
NIBNIGL ∗ XRM −2.810 1.357∗∗∗
(0.628) (0.006)
NIGL ∗ XRM −168.167 1.402
(0.291) (0.935)
OCIGL ∗ XRM 10.703∗ 4.315∗
(0.052) (0.074) Risk modeling measures and chief risk officer indicator
DISCGL ∗ XRM −1.035 0.982∗∗
(0.666) (0.049) Risk modeling measures
XRM 0.297 −0.102∗
(0.410) (0.086)
NIBNIGL ∗ XRM 3.748 4.760∗∗ Our risk modeling measures are simple averages of indicators
(0.288) (0.027) for whether banks engage in each of the following risk model-
NIGL ∗ XRM −21.199 −9.753 ing activities (including related risk management processes) men-
(0.243) (0.428) tioned a survey developed by the Basel Committee on Banking Su-
OCIGL ∗ XRM 5.515∗∗ −2.080 pervision (2001):
(0.033) (0.930)
Market risk modeling:
DISCGL ∗ XRM −0.140 1.136∗
(0.975) (0.060)
XRM −1.789∗∗ −1.090∗∗
(0.033) (0.048) (1) INTGAP – Does the bank use maturity or repricing analysis?
NIBNIGL ∗ (XRM − XRM) 1.056 0.003 (2) IRSIMUL – Does the bank use interest rate simulation?
(0.721) (0.995) (3) VAR – Does the bank use Value at Risk for market risk?
NIGL ∗ (XRM − XRM) 37.275 −25.090 (4) MRSTRESS – Does the bank stress test its market risk models?
(0.498) (0.141)
(5) MRBACKTEST – Does the bank back test its market risk models?
OCIGL ∗ (XRM − XRM) 1.693 1.363
(0.879) (0.555)
DISCGL ∗ (XRM − XRM) −0.748 0.295
(0.630) (0.616) Credit risk modeling:
(XRM − XRM) −0.129 0.003
(0.588) (0.957)
Constant 0.056∗∗∗ 0.058∗∗∗ 0.057∗∗∗ 0.057∗∗∗ (6) MODEL – Does the bank use statistical credit risk modeling?
(0.000) (0.000) (0.000) (0.000) (7) SCORING – Does the bank use credit scoring models?
Observations 2214 2214 2214 2214
(8) RR – Does the bank have an internal credit risk rating process?
R-squared 0.041 0.050 0.050 0.051
(9) CRSTRESS – Does the bank use stress testing its credit risk
Notes: Columns (1) and (2) [(3) and (4)] of this table report the results models?
of a modification of the regression reported in columns (2) and (3) [(4)
and (5)] of Table 4, Panel A that includes the changes in MRM, MRM and
(MRM − MRM) [CRM, CRM and (CRM − CRM)] rather than the levels of these
variables. Variables are defined in the Appendix cȁDefinitions of variables and Our market risk modeling measure, MRM, is the simple average
other acronymsdȁ. All variables are winsorized at the 0.5% and 99.5% levels of the first five indicators. Our credit risk modeling measure, CRM,
of their distribution. Standard errors are calculated clustering observations by is the simple average of the next four indicators. As averages of
bank. P-values are in parentheses under the corresponding coefficients.
∗
indicator variables, both measures take values from zero to one.
Denotes significance at the 10% level.
∗∗
Denotes significance at the 5% level.
We identify whether banks engage in these activities from
∗∗∗
Denotes significance at the 1% level. hand-collected disclosures in their Form 10-K filing for the years
2001–2013. We first used search engines to locate words or
phrases such as cȁrisk management,dȁ cȁrisk model,dȁ cȁscoring,dȁ
and cȁrisk rating,dȁ and read each search result to determine
loans and unfunded loan commitments for which the fair values
whether one (or more) of the nine risk modeling activities is indi-
are disclosed rather than recognized. Finally, we show that the im-
cated. Each item is scored one if disclosed and zero otherwise. The
pact of MRM (CRM) activities is stronger during periods of higher
following table provides representative examples of disclosures for
interest rate volatility (the financial crisis).
each of the nine activities.
Our evidence is subject to the following caveats. First, we show
that banks’ MRM and CRM are associated with increased returns-
relevance for specific types of FVGL, not that they cause this in-
creased returns relevance. Correlated bank attributes, such as tech-
nical sophistication, may be the actual causes. Second, we do not
control for alternative financial report disclosures related to the 18
For example, see the FASB’s April 12, 2013 proposed Accounting Standards Up-
quality of their FVGL estimates, because these alternatives are sig- date, Financial Instruments—Overall (Subtopic 825-10): Recognition and Measure-
nificantly limited in terms of one or more of their information ment of Financial Assets and Financial Liabilities, which would allow firms consid-
quality, coverage across types of financial instruments, or coverage erable ability to classify financial instruments to avoid fair valuation.
94 G. Bhat, S.G. Ryan / Accounting, Organizations and Society 46 (2015) 81–95
The chief risk officer indicator, CRO, takes the value one if the
bank discloses in its Form 10-K filing that it employs a chief risk
officer and zero otherwise. We first search filings to determine
whether the words cȁchiefdȁ and cȁriskdȁ appear within five words
of each other. We then search the filings that meet the first filter
to determine whether the words cȁriskdȁ and either cȁofficerdȁ or
cȁdirectordȁ appear within five words of each other. We read the
relevant passages to confirm the existence of a chief risk officer for
the bank-years that meet both filters.