Embed Size (px)
Citation preview
Electronic copy available at: https://ssrn.com/abstract=3074825
Fair Value versus Amortized Cost Measurement and the Timeliness of Other-than-
Temporary Impairments: Evidence from the Insurance Industry
Urooj Khan
Columbia University
Stephen Ryan
New York University
Abhishek Varma
Illinois State University
November 20, 2017
Electronic copy available at: https://ssrn.com/abstract=3074825
Fair Value versus Amortized Cost Measurement and the Timeliness of Other-than-
Temporary Impairments: Evidence from the Insurance Industry
ABSTRACT: We investigate the impact of recurring fair value versus amortized cost
measurement on the timeliness of insurers’ OTT impairments of non-agency RMBS around the
2007–2009 financial crisis. Recurring fair value measurement requires firms to assess relevant
economic conditions quarterly, and thus to invest in information and control systems to make these
assessments. We expect these systems to discipline insurers’ OTT impairments. In contrast,
amortized cost measurement is a largely pre-determined accounting approach. Exploiting the
statutory accounting requirement that PC (life) insurers measure securities with NAIC designations
from 3 to 5 at fair value (amortized cost) and the statutory reporting requirement that insurers
provide security-level accounting information, we predict and find that PC insurers record timelier
OTT impairments of given non-agency RMBS with these designations than do life insurers. We
predict and find weaker evidence of spillover effects to the timeliness of PC insurers’ OTT
impairments of given non-agency RMBS with NAIC designations of 1 or 2.
Electronic copy available at: https://ssrn.com/abstract=3074825
1
1. INTRODUCTION
We investigate the impact of recurring fair value versus amortized cost measurement for
accounting recognition purposes1 on the timeliness of insurers’ other-than-temporary (OTT)
impairments of non-agency residential mortgage-backed securities (RMBS) from 2007 to 2011,
i.e., as a consequence of the 2007–2009 financial crisis. We examine OTT impairments of non-
agency RMBS during the crisis period because of the high frequency of credit downgrades of these
securities (Ellul et al. 2015), because financial institutions (including insurers) widely held these
securities, and because many parties criticized financial institutions for not recording timely write-
downs of financial assets (e.g., Dickey, King, and Shih 2008; Einhorn 2008). Supporting this
criticism, Vyas (2011) provides evidence that financial institutions recorded losses on financial
assets with lags relative to the deterioration in relevant credit indices during this period. The
timeliness of OTT impairments is important because prior research shows that timelier loss
recognition enhances financial institutions’ internal, regulatory, and market discipline and thereby
reduces the procyclicality of their investment and financing decisions (Stephanou 2010; Beatty
and Liao 2011; Bushman and Williams 2012, 2015; Bhat, Ryan, and Vyas 2017; Ryan 2017).
We examine OTT impairments by insurers rather than by other firms for three reasons.
First, investment securities are insurers’ primary financial asset. Second, insurers must report
detailed accounting information for each of their investment securities in their annual (and to a
lesser extent quarterly) statutory reports. Hence, we can observe the relative amounts and timing
of different insurers’ OTT impairments of a given security, which provides an unusually high level
1 For simplicity, throughout the paper we refer to recurring fair value measurement for accounting recognition (as
opposed to just disclosure) purposes as fair value measurement. Similarly, we refer to amortized cost measurement
for accounting recognition purposes as amortized cost measurement. Consistent with the broader recognition versus
disclosure literature, extant research on recognized versus disclosed fair values indicate the former are more value
relevant, and thus are likely measured with greater reliability, than are disclosed fair values (Ahmed, Kilic, and Lobo
2006; Müller, Riedl, and Sellhorn 2015).
2
of experimental control in our analysis. Third, under statutory accounting principles (SAP),
property-casualty (PC) insurers and life insurers account differently for investment securities that
are below investment grade but are not yet in or near default, i.e., securities with National
Association of Insurance Commissioners (NAIC) designations from 3 to 5.2 PC insurers measure
these securities at lower of cost or fair value, which generally equals fair value because insurers
rarely acquire non-investment grade securities, implying that the securities have been downgraded
after acquisition.3 In contrast, life insurers measure these securities at amortized cost. However,
PC insurers’ and life insurers’ investment securities are subject to the same SAP OTT impairment
rules. Hence, in principle SAP requires the two types of insurer to record OTT impairments of
given securities at the same times and, assuming the insurers have identical intents and abilities to
hold the securities, to write down the cost bases of the securities to the same amounts.
Reflecting PC insurers’ fair value measurement and life insurers’ amortized cost
measurement of non-agency RMBS with NAIC designations from 3 to 5 under SAP, we
hypothesize that PC insurers record timelier OTT impairments of these securities than do life
insurers, all else being equal. Our primary basis for this hypothesis is that, ignoring OTT
impairments of investment securities and other non-recurring write-downs of financial assets, as
well as interest revenue on certain risky assets, amortized cost is a pre-determined accounting
measurement approach that does not require reporting firms to pay attention to economic
conditions affecting their financial instruments. In contrast, recurring fair value measurement
2 NAIC designations are typically based on either credit analysis by the NAIC’s Securities Valuation Office or credit
ratings by Approved National Recognized Statistical Rating Organizations such as S&P, Moody’s, and Fitch. NAIC
designations range from 1 (highest credit quality) to 6 (lowest credit quality). NAIC designations 1 and 2 are for
investment-grade investment securities. NAIC designation 6 is for securities that are in or near default. 3 For this reason, we often refer to PC (life) insurers’ lower of cost or fair value measurement for non-agency RMBS
with NAIC designations below 2 (5) under SAP as fair value measurement.
3
requires reporting firms to assess these conditions no less frequently than quarterly.4 This ongoing
assessment requires reporting firms to invest in information systems that provide adequately
relevant and timely information to measure fair values, as well as control systems to ensure the
reliability of these measurements (Bhat and Ryan 2015). We expect these systems also discipline
reporting firms’ determinations of the amounts and timing of their OTT impairments of investment
securities.
A potential secondary basis for this hypothesis is that, under SAP lower of cost or fair value
accounting PC insurers record unrealized fair value losses on securities with NAIC designations
from 3 to 5 directly as reductions of statutory surplus. Hence, PC insurers’ risk-based capital ratios
generally will not decrease if they record OTT impairments of these securities, all else being equal,
implying that PC insurers have no capital-related incentives not to record OTT impairments of
these securities. In contrast, life insurers may have such incentives; their risk-based capital ratios
decrease with OTT impairments of securities with NAIC designations from 3 to 5 if their asset
valuation reserves (AVRs)—SAP-required income and statutory surplus smoothing mechanisms
that absorb both realized and unrealized credit losses on investments—are not large enough to
absorb their OTT impairments of securities. Life insurers’ AVRs are constructed to be large
enough to absorb their realized and unrealized credit losses in normal times, but this may not have
been the case for all life insurers during the severely adverse financial crisis period we examine.
Non-agency RMBS are a homogeneous class of security subject to common economic
conditions (e.g., house price depreciation) and for which fair value measurement involves similar
valuation inputs and models. Accordingly, we expect PC insurers’ fair value measurement of non-
agency RMBS with NAIC designations from 3 to 5 improves their understanding of non-agency
4 Reporting firms may engage third parties to make these assessments. For simplicity, we refer to reporting firms as
making these assessments throughout the paper.
4
RMBS with NAIC designations of 1 and 2, and we further hypothesize that PC insurers record
timelier OTT impairments of NAIC 1 and 2 designated non-agency RMBS than do life insurers.
We expect this spillover effect of fair value measurement to be weaker than the direct effect posited
in our first hypothesis, however, for two reasons. First, the spillover effect is indirect. Second,
OTT impairments of NAIC 1 and 2 designated securities reduce PC insurers’ statutory surplus and
risk-based capital ratios, creating capital-related frictions to these OTT impairments. Hence, the
potential secondary basis for our first hypothesis described in the prior paragraph does not apply
to these impairments.
We test the above hypotheses on a sample of non-agency RMBS: (1) that are held by at
least one PC insurer and at least one life insurer at the end of 2006, to enable comparison across
the two types of insurer; and (2) for which the insurers do not acquire any more of the securities
during 2007–2011, so that there essentially is only one vintage of each security. We eliminate
subsequent security-years in which the security is no longer held by at least one insurer of each
type. We also eliminate security-year observations involving insurers with either inadequate or
extremely high risk-based capital ratios, so that regulatory capital is neither so low as to yield
intense regulatory scrutiny nor so high as to largely insulate the insurer from economic and
accounting shocks. The sample includes 5,796 security-year observations for 1,590 non-agency
RMBS held by 352 PC insurers and 647 life insurers.
We measure the timeliness of an insurer’s OTT impairment of a given non-agency RMBS
in a given year as the insurer’s cumulative OTT impairment of the security through the end of that
year divided by its amortized cost of the security at the end of 2006—i.e., the last year end prior
5
to the financial crisis—minus the median of this ratio across all insurers holding the security.5 As
expected, we find that PC insurers record significantly timelier OTT impairments of non-agency
RMBS with NAIC designations from 3 to 5 than do life insurers. We also find significant but
weaker evidence that PC insurers record significantly timelier OTT impairments of non-agency
RMBS with NAIC designations 1 and 2 than do life insurers.
The main problem of inference confronting this study is that PC insurers and life insurers
have distinct business models (Ryan 2007, Chapter 13). Our security-level research design
addresses life insurers’ tendency to choose longer maturity and/or less liquid securities than do PC
insurers does not raise problems of inference for this study. However, life insurers typically hold
longer duration liabilities than do PC insurers and so may have longer intended holding periods
for given investment securities than do PC insurers. Hence, life insurers may expect to recover
more of the deterioration of the fair value of the securities over these longer holding periods. Under
this possibility, SAP may require PC insurers to record more frequent or larger OTT impairments
than it requires of life insurers.
We conduct two analyses to address this problem of inference. First, we conduct analyses
within each insurer type in which we compare the timeliness of OTT impairments of non-agency
RMBS with NAIC designations of 1 or 2, which both types of insurer measure at amortized cost,
to the timeliness of OTT impairments of non-agency RMBS with NAIC designations 3 to 5, which
PC insurers measure at fair value and life insurers measure at amortized cost. We expect that PC
insurers’ OTT impairments of non-agency RMBS with NAIC designations from 3 to 5 are timelier
than their OTT impairments of non-agency RMBS with NAIC designations of 1 or 2, whereas we
5 Our choice of this measure reflects our expectation that insurers had no incentive to overstate OTT impairments
during the financial crisis and its aftermath, in part owing to the severity of the crisis, and in part because reporting
firms can only reverse overstated OTT impairments indirectly through the gradual accrual of interest revenue.
6
expect no such difference for life insurers. Second, we compare the timeliness of OTT impairments
for the subsample of non-agency RMBS for which at least one PC insurer and one life insurer do
not sell all their holdings of the security in the following one or two years, reasonable horizons
over which insurers could feasibly evaluate their intent to hold securities. The results of both of
these analyses are consistent with those of the main analysis.
We conduct two additional analyses to ensure the robustness of our findings. First, we
compare the timeliness of OTT impairments of the subsample of non-agency RMBS whose NAIC
designations were downgraded after 2006, indicating significant deteriorations in their credit
quality. Consistent with our main analysis, we find that PC insurers record timelier OTT
impairments of given downgraded non-agency RMBS, particularly those with post-impairment
NAIC designations from 3 to 5, than do life insurers.
Second, we employ an alternate measure of the timeliness of OTT impairments. Following
Vyas (2011), we calculate the timeliness of an insurer’s OTT impairment of a non-agency RMBS
in a given year as the cumulative OTT impairment recorded by the insurer through the end of the
year divided by the cumulative OTT impairment recorded by the insurer through the end of 2011,
again subtracting the median of this measure across all insurers holding the security. The results
of this analysis are also consistent with those of the main analysis.
To the best of our knowledge, our study is the first to show that fair value measurement
leads to timelier recognition of any type of non-recurring write-downs of financial assets than does
amortized cost measurement. This finding is akin to prior findings that fair value measurement
leads to less gains trading in investment securities than does amortized cost measurement (Ellul et
al. 2015), a longstanding concern of policymakers (Wyatt 1991).
Our findings speak to the debate about the role of fair value accounting during the financial
crisis. To date, this debate has centered on the use of fair value accounting information by parties
7
external to the reporting firm, in particular, whether fair value accounting induced significant
adverse feedback effects that exacerbated the crisis versus simply summarized the effects of the
crisis to users of financial reports (Ryan 2008; Laux and Leuz 2009, 2010). Our results are
consistent with fair value measurement also having implications internal to firms, specifically,
with it leading firms to invest in information and control systems that, as a side benefit, increase
the timeliness of their OTT impairments of securities. We conjecture that these systems may also
improve the overall transparency of firms’ financial reporting, e.g., MD&A and footnote
disclosures, a promising topic for future research in our view.
Relatedly, our findings complement those of prior studies that provide evidence that firms
exploit their discretion over fair value measurements to achieve financial reporting objectives,
thereby reducing transparency. For example, Hanley, Jagolinzer, and Nikolova (2017) provide
evidence of greater inflation of fair value measurements of a given investment security by insurers
that base these measurements on lower-level inputs than the consensus insurer, by insurers that
self-estimate the measurements rather than obtain them from a third-party, and by public insurers
than by private insurers. Our results indicate that fair value measurement can enhance the overall
quality of financial statements by improving the timeliness of OTT impairments of investment
securities.
The remainder of the paper is organized as follows. Section 2 discusses the institutional
background and develops our empirical predictions. Section 3 reports the details of the sample
construction and the empirical tests. Sections 4 and 5 present the results of the primary tests and
of additional analyses, respectively. Section 6 concludes.
8
2. INSTITUTIONAL BACKGROUND AND HYPOTHESIS DEVELOPMENT
2.1. Background on Insurance Regulation and Statutory Reporting
Under the McCarran-Ferguson Act of 1945, insurers are regulated at the state level.
Insurers are required to provide the NAIC and their state regulators with highly detailed and
standardized annual and quarterly statutory financial statements prepared in accordance with SAP.
While SAP developed at the state level, beginning in 2001 the NAIC has codified SAP in its
Accounting Practices and Procedures Manual. This manual does not override state-level variation
in SAP, but it has led to a high degree of consistency in SAP across states. The requirements of
SAP and U.S. generally accepted accounting principles (GAAP) overlap in many respects,
although SAP tends to be more conservative owing to insurance regulators’ primary focus on
insurer solvency.
Since 1993, insurers are subject to risk-based capital requirements that are based on SAP
measurements of assets, liabilities, and capital. An insurer’s minimum risk-based capital
requirement is largely built up from the sum of the requirements for its individual exposures,
although certain covariances across exposures and insurer-level risks are also incorporated. The
risk-based capital requirements for fixed-income securities are determined by their NAIC
designations. Higher designations indicate greater credit risk and thus have higher risk-based
capital requirements. The risk-based capital requirements for securities with NAIC designations
from 3 to 5 differ for PC and life insurers to compensate for the different required accounting for
these securities by the two types of insurers under SAP.
2.2. Statutory Accounting Principles for Non-Agency RMBS
Statement of Statutory Accounting Principles No. 43R, Loan Backed and Structured
Securities, (SSAP 43R) is the SAP standard governing non-agency RMBS beginning in the third
quarter of 2009. SSAP 43R superseded the identically titled prior standard SSAP 43. Ignoring
9
OTT impairments, the two SAP standards contain identical requirements regarding the balance
sheet measurements and income statement classifications for non-agency RMBS. Both standards
require: (1) all insurers to measure investment-grade securities (NAIC designation 1 or 2) at
amortized cost; (2) PC insurers to measure non-investment grade securities that are not yet in or
near default (NAIC designation from 3 to 5) at the lower of cost or fair value and life insurers to
measure these securities at amortized cost; (3) all insurers to measure securities in or near default
(NAIC designation 6) at lower of cost or fair value; (4) PC insurers to record both realized and
unrealized gains and losses on securities in net income and statutory surplus; and (5) life insurers
to record realized and unrealized gains and losses on securities in asset valuation reserves (AVRs).
Life insurers’ AVRs are SAP-required income and statutory surplus smoothing mechanisms that
absorb both realized and unrealized credit losses on investments as long as the AVRs remain within
specified bounds.6
The main differences between SSAP 43R and SSAP 43 pertain to the criteria under which
securities are OTT impaired and the accounting for OTT impairments. SSAP 43R’s OTT
impairment requirements are very similar to those of FASB Staff Position FAS 115-2 and FAS
124-2, Recognition and Presentation of Other-Than-Temporary Impairments, an amendment of
Statement of Financial Accounting Standards No. 115, Accounting for Certain Investments in Debt
6 Life insurers’ AVRs are liabilities that first reduce life insurers’ unassigned surplus but then are added back and have
no effect on statutory surplus. Life insurers first determine AVRs for each asset class, in particular, distinguishing the
default component associated with fixed income assets, such as non-agency RMBS, from the equity component
associated with common stock and real estate assets. Life insurers then sum these component AVRs to the insurer
level. Each year a life insurer adds a basic contribution that equals its estimated annual after-tax losses based on 1992
baseline assumptions to its AVR. The insurer may also add a voluntary contribution. In addition, if the insurer’s current
AVR is less than the target AVR—which is the 85th percentile of the modeled credit loss distribution for each asset
class—the insurer adds 20% of the difference to the AVR each period. These contributions reduce insurers’ income
and statutory surplus. As long as an insurer’s AVR remains positive and below the specified maximum, the insurer’s
realized and unrealized gains (losses) add to (subtract from) its AVR, and so have no immediate effect on its income
and statutory surplus. See Bennett (2013) and Berry-Stölzle, Nini, and Wende (2014, footnote 15) for further details
about AVRs.
10
and Equity Securities, (“FSP FAS 115-2 & FAS 124-2”) with respect to whether securities are
OTT impaired and the measurement of OTT credit losses.7
Simplifying somewhat, under SSAP 43R securities are OTT impaired when the holder does
not expect to receive the cash flows expected to be collected at acquisition or the time of the most
recent prior OTT impairment. If the holder does not intend to sell and more likely than not will not
have to sell OTT-impaired securities prior to recovery of their amortized cost bases, these
amortized cost bases are written down to the present value of the cash flows expected to be
collected. These present values exceed the securities’ fair values if the effective interest rates are
less than the current relevant market rates for the securities. If the holder intends to sell or more
likely than not will have to sell OTT impaired securities prior to recovery of their amortized cost
bases, these amortized cost bases are written down to fair value. PC insurers record write-downs
of the amortized cost bases of securities, i.e., OTT credit losses, as reductions of net income and
statutory surplus. Life insurers record these losses in their AVRs, as long as these reserves remain
positive. Life insurers’ AVRs are constructed to be large enough to absorb their OTT impairments
of investment securities and other realized and unrealized losses on investment assets in normal
times; this may not have been the case for all life insurers during the severely adverse financial
crisis period we examine.8
7 Unlike FSP FAS 115-2 and FAS 124-2, SSAP 43R does not require insurers to record OTT non-credit losses, i.e.,
decreases of fair value below the present value of the cash flows expected to be collected, on OTT-impaired investment
securities that they do not intend to sell and more likely than not will not have to sell prior to recovery of the securities’
amortized cost bases. However, like SSAP 43, SSAP 43R effectively requires PC (life) insurers to record these non-
credit losses as unrealized losses under lower of cost or fair value accounting for investment securities with NAIC
designations 3-6 (6). The cumulative unrealized losses on these securities under lower of cost or fair value accounting
equal the cumulative non-credit losses on the securities in periods with OTT impairments, but not in other periods. 8 As discussed in footnote 6, the target AVR is the 85th percentile of the modeled credit loss distribution for each asset
class. Berry-Stölzle, Nini, and Wende (2014) state that “In aggregate, the AVR was reduced substantially during
2008.”
11
In contrast, SSAP 43 required a very weak OTT impairment approach based on
undiscounted cash flows. A security was OTT impaired if the undiscounted sum of expected cash
flows was less than the security’s amortized cost basis. The amortized cost basis of an OTT-
impaired security was written down to the undiscounted sum of the expected cash flows.
All mentions of SAP in this paper refer to SSAP 43 up to 2009Q2 and to SSAP 43R
beginning in 2009Q3. These two standards are consistent with the SAP standards governing other
types of investment securities during the corresponding time periods.
2.3. Fair Value versus Amortized Cost Measurement and the Timeliness of OTT Impairments
Ignoring sales, the amortized cost basis of a security equals its acquisition cost minus
cumulative principal repayments, plus cumulative interest revenue, minus cumulative interest
receipts, and minus cumulative OTT impairments. With the exception of the calculation of non-
recurring OTT impairments and other write-downs of financial assets,9 as well as interest revenue
on certain risky assets,10 amortized cost is a pre-determined accounting measurement basis that
does not incorporate current information in a timely fashion.
In contrast, glossing over the subtleties of fair value measurement guidance, in principle
the fair value of a security should equal the holder’s estimate of the expected present value of the
future cash flows, where the expectations of future cash flows and security-specific discount rates
are based on current information about the security’s performance and relevant economic
conditions (Barth 2014). In cases where fair values are not directly observed (i.e., are based on
9 The primary other types of non-recurring write-downs of financial assets are: (1) write-downs of heterogeneous loans
held for investment to the present value of expected cash flows under FAS 114; and (2) lower of cost or fair value
accounting for loans held for sale, with write-downs to fair value being non-reversible, under FAS 65 and SOP 01-6. 10 EITF 99-20 governs the GAAP accounting for interest revenue on beneficial interests in securitized financial assets,
such as non-agency RMBS, if these interests are not of high credit quality. EITF 99-20 requires firms that hold these
securities to revise estimates of future cash flows each period and, if these revisions do not yield OTT impairments,
to recalculate the effective interest rates used to calculate interest revenue. SSAP 43R, paragraphs 20-23, incorporate
EITF 99-20’s requirements into SAP.
12
Level 2 or Level 3 inputs as specified in FAS 157’s fair value hierarchy), in order to accurately
and reliably estimate fair values the reporting firm must invest in information systems to enable
adequately timely identification and use of information about security performance and relevant
economic conditions (Bhat and Ryan 2015). After acquisition, reporting firms invariably fair value
non-agency RMBS using Level 2 or Level 3 inputs owing to the inactivity of the secondary markets
for these securities (Hu 2001, pp. 117-119; Bond Market Association and American Securitization
Forum 2006, pp. 41-44).
In addition, firms must have control systems in place to ensure the reliability of their fair
value measurements. A few examples of such controls are pricing and valuation committees that
include senior management personnel independent of firms’ trading and investing functions and
that evaluate fair value measurements and fair valuation methodologies; comparisons of fair value
measurements of securities to the prices received when the securities are sold and other forms of
backtesting; and comparisons of self-estimated fair value measurements with quotes from
independent pricing services or brokers.
Bhat and Ryan (2015) provide evidence suggesting that firms’ investments in risk
modeling systems improve the quality of their fair value estimates. Specifically, they show that
banks’ market risk and credit risk modeling are associated with enhanced returns relevance of
unrealized fair value gains and losses.
Based on the discussion above, we expect that PC insurers required to measure a given
non-agency RMBS with NAIC designation from 3 to 5 at fair value have access to timelier and
more reliable information about the security’s current performance and relevant economic
conditions than do life insurers required to measure the security at amortized cost. We expect this
enhanced information increases PC insurers’ awareness of the proper amount and timing of OTT
impairments of the security. Such awareness was particularly important during the severely
13
adverse conditions that arose during or as a direct consequence of the financial crisis, which
yielded high economic losses on non-agency RMBS.11 In addition, as discussed in the introduction,
PC insurers have no capital-related incentives not to record OTT impairments of these securities,
whereas life insurers may have such incentives if their AVRs are not large enough to absorb the
losses. We formally state this expectation in alternative form:
H1 (Direct Effect of Fair Value Measurement of Given Securities): PC insurers recorded
timelier OTT impairments of given non-agency RMBS with NAIC designations from 3 to 5
during and shortly after the financial crisis than did life insurers.
Non-agency RMBS are a homogeneous class of security subject to common economic
conditions (e.g., house price depreciation) and for which fair value measurement involves the same
types of valuation inputs and models. Hence, we expect PC insurers’ fair value measurement of
non-agency RMBS with NAIC designations from 3 to 5 improves their understanding of non-
agency RMBS with NAIC designations of 1 and 2. For example, PC insurers should have better
understandings than do life insurers of the likelihoods that given non-agency RMBS with NAIC
designations of 1 and 2 will subsequently be downgraded to NAIC designations of 3 or worse,
requiring PC insurers to measure the securities at fair value. Accordingly, we further hypothesize
that PC insurers record timelier OTT impairments of NAIC 1 and 2 designated non-agency RMBS
than do life insurers.12 We expect this spillover effect to be somewhat weaker than the direct effect
11 We expect the financial crisis and the resulting downgrades of non-agency RMBS led life insurers to improve their
information and control systems for the securities to some extent. However, we do not expect any such improvements
to cause life insurers’ systems to catch up to the quality of those of PC insurers. 12 If insurers are averse to recording OTT impairments in large chunks, say due to the non-reversibility of these
impairments, then PC insurers’ fair value measurement of non-agency RMBS with NAIC designations from 3 to 5
could also yield spillover effects on the relative timeliness of PC insurers’ and life insurers’ OTT impairments of non-
agency RMBS with NAIC designations of 6, even though both types of insurer measure these securities at lower of
cost or fair value. We do not propose such a hypothesis, however, primarily because these securities constitute only
183 (3.2%) of our 5,796 security-year observations, so the statistical power of these analyses is low. Reinforcing this
lack of power, securities with NAIC designations of 6 typically are very substantially OTT impaired by both types of
insurer; for example, by the end of 2010 our sample PC (life) insurers had recorded average cumulative OTT
impairments of 30 (25) percent on securities with NAIC designations of 6. For completeness, however, we report
empirical analyses for these securities.
14
postulated in Hypothesis H1, however, for two reasons. First, this spillover effect is indirect, as
both types of insurer measure NAIC 1 and 2 designated non-agency RMBS at amortized cost.
Second, OTT impairments of NAIC 1 and 2 designated non-agency RMBS reduce PC insurers’
statutory surplus and risk-based capital ratios, creating capital-related frictions to their OTT
impairments of these securities.
H2 (Spillover Effect of Fair Value Measurement of Similar Securities): PC insurers
recorded timelier OTT impairments of given non-agency RMBS with NAIC designations of
1 or 2 during and shortly after the financial crisis than did life insurers. This spillover
effect is weaker than the direct effect posited in Hypothesis H1.
Prior studies show that insurers exploit their discretion over fair value measurements to
achieve financial reporting objectives (Vyas 2011; Ellul et al. 2015; Hanley et al. 2017). Therefore,
in our empirical tests of both hypotheses we control for insurers’ financial reporting incentives
(such as regulatory capital and leverage) and other factors that may influence the timeliness of
insurers’ OTT impairments.
3. DATA AND RESEARCH DESIGN
3.1. Data Sources and Sample Construction
We construct a sample of non-agency RMBS held by insurers during 2007–2011, a period
during which most insurers holding these securities recorded OTT impairments of the securities.
We obtain insurers’ annual financial statement and security-level data indirectly from the annual
statutory filings of insurers and directly from S&P Global Market Intelligence. Insurers prepare
these filings in accordance with SAP, and they present the filings in the format of standardized
“blanks” created by the NAIC. Schedule D, Part 1 of these filings reports the actual (i.e.,
acquisition) cost, fair value, and a detailed T-account roll-forward during the year for each fixed-
income security, including non-agency RMBS, held by the insurer at the end of the year. Other
15
parts of Schedule D provide information about each fixed-income security: acquired during the
year (Part 3); sold, redeemed, or otherwise disposed of during the year (Part 4); and both acquired
and sold during the year (Part 5). Parts 3-5 (but not Part 1) of Schedule D are also provided in
insurers’ quarterly statutory filings.
Table 1 summarizes the sample construction. We begin with the entire universe of 5,552
insurers (3,860 PC and 1,692 life) available on S&P Global Market Intelligence during 2007-2011,
regardless of whether or not they hold non-agency RMBS. We then eliminate insurer-years that
do not hold non-agency RMBS, which yields 1,723 insurers (1,188 PC and 535 life) that hold
22,567 non-agency RMBS, for a total of 172,398 security-insurer-year observations. We impose
three primary filters on these observations to yield well-specified hypothesis tests.
First, because we calculate the timeliness of the cumulative OTT impairment of a security
by dividing that impairment by the security’s actual cost at the end of 2006, we exclude non-
agency RMBS that the sample insurers entirely or partly acquired after 2006. This filter yields a
43 percent decrease in the number of non-agency RMBS, a 21 percent decrease in the number of
insurers, and a 45 percent decrease in the number of security-insurer-year observations.
Second, following prior studies (e.g., Ellul et al. 2015), we exclude insurer-years with risk-
based capital ratios below 200 or above 2000 percent. Insurers in the former range face intense
regulatory scrutiny that likely dominates their economic and accounting behavior. Insurers in the
latter range are highly insulated from economic and accounting shocks. This filter yields relatively
small decreases in the numbers of non-agency RMBS, insurers, and security-insurer-year
observations.
Lastly, because we assess the relative timeliness of OTT impairments by PC insurers versus
life insurers, we exclude security-year observations in which the security is not held by both types
of insurer. Owing to the typically highly concentrated holdings of a given non-agency RMBS (an
16
individual insurer often holds an entire tranche created in a non-agency RMBS securitization), this
filter yields an 87 percent decrease in the number of non-agency RMBS, a 15 percent decrease in
the number of insurers, and a 73 percent decrease in the number of security-insurer-year
observations. This filter provides a substantial methodological side-benefit that compensates for
this sizable loss of observations, however, by enabling us to control for the influence of all security-
specific factors on the relative timeliness of OTT impairments by the two types of insurer.
The final sample includes 1,590 non-agency RMBS securities held by 1000 insurers (648
PC and 352 life), for 24,566 security-insurer-year observations. In total, the three filters eliminate
93 percent of the number of non-agency RMBS, 42 percent of the number of insurers (34 percent
of the PC insurers and 46 percent of the life insurers), and 86 percent of the number of security-
insurer-year observations. The retention of significant numbers of both PC insurers and life
insurers preserves the power of our tests of the relative timeliness of OTT impairments of non-
agency RMBS by the two types of insurer, as well as the generalizability of our findings.
3.2. Primary Measure of the Timeliness of OTT Impairments
We measure the timeliness of insurers’ OTT impairments of non-agency RMBS during our
sample period as follows. First, we divide insurer i’s cumulative OTT impairment recorded on
security s from the beginning of 2007 to the end of year t, denoted CumImpi,s,t, by the insurer’s
actual cost of the security at the end of 2006, denoted Costi,s. (Definitions of all variables in the
paper are provided in Appendix A.) Costi,s is unaffected by insurer i’s accounting measurement
approach for security s and is largely unaffected by the timing of insurer i’s purchase of security s
prior to the end of 2006, because the market values of non-agency RMBS securities generally did
not appreciably vary from par prior to the subprime crisis beginning in February 2007 (Ryan 2008).
Hence, CumImpi,s,t/Costi,s should be comparable across insurers and securities. Second, to provide
a benchmark for CumImpi,s,t/Costi,s, we calculate the median of this ratio across all insurers holding
17
security s at the end of year t, denoted Median(CumImp.,s,t/Cost.,s). Recall that for a security-year
to be included in the sample, the security must be held by both PC and life insurers in that year.
Third, we calculate the timeliness of insurer i’s cumulative OTT impairment of non-agency RMBS
s as of the end of year t, denoted TMi,s,t, as one hundred times CumImpi,s,t/Costi,s minus
Median(CumImp.,s,t/Cost.,s). By construction, TMi,s,t equals zero for securities not impaired by any
insurer.
Table 2 presents descriptive information about the sample as of the end of each year,
including the number of securities, the number of sample insurers, the average number of insurers
holding a given security, the number of OTT-impaired securities, and the mean and variation
across insurers of the cumulative OTT impairments of the securities. The number of securities
monotonically declines from 1,563 in 2007 to 794 in 2011 as the sample insurers dispose of
securities (e.g., through sale or litigation settlements) or leave the sample (e.g., through failure or
merger), with cascade effects when either of these effects causes a security to no longer be held by
both a PC insurer and a life insurer. The number of insurers declines monotonically from 948 in
2007 to 604 in 2011. The average number of insurers holding a given security falls slightly from
4.52 in 2007 to 3.88 in 2011, reflecting the decrease in the number of sample insurers over time.
We define a security as OTT impaired as of the end of a year if at least one sample insurer
has recorded a positive cumulative OTT impairment of the security from 2007 to the end of that
year. The number of OTT-impaired securities rises sharply from 113 (7.7 percent of the sample)
in 2007 to 409 (36.1 percent of the sample) in 2009, reflecting the worsening of the financial crisis,
but then falls to 363 (45.7 percent of the sample) in 2011, reflecting sample attribution, despite the
increasing percentage of OTT-impaired securities. The average number of insurers holding a given
OTT-impaired security falls monotonically from 5.13 in 2007 to 4.00 in 2011, reflecting sample
attrition.
18
To give a sense for the typical magnitudes of OTT impairments, Table 2 reports two
measures of these magnitudes in the sample in each year. The aggregate cumulative OTT
impairment of securities in a year, denoted Agg CumImp % on Imp Secs, equals the sum of
CumImpi,s,t across insurers and securities divided by the sum of Costi,s across insurers and securities
in the year. The average cumulative OTT impairment of securities in a year, denoted Avg CumImp
% on Imp Secs, equals the mean of CumImpi,s,t/Costi,s across insurers and securities in the year.
While the aggregate measure is somewhat lower than the average measure, the two measures
follow similar upward paths. For example, the average measures increases monotonically from
4.76 percent in 2007 to 17.51 percent in 2011.
To give a sense for the variability of OTT impairments in the sample in each year, Table
2 reports two measures of this variability in the sample in each year. Both measures are based on
the standard deviation of CumImpi,s,t/Costi,s across insurers for each security in a year, denoted
StdDevs,t. The first measure, denoted Avg Std Dev Imp Sec, is the mean of Std Devs,t across
securities in the year. The second measure, denoted IQR Std Dev Imp Sec, is the interquartile
range of of StdDevs,t across securities in the year. Both measures increase from 2007 to their
peaks in 2008 and then fall monotonically to 2011, indicating that insurers’ average percentage
cumulative OTT impairments first diverge as the crisis develops and then converge as the crisis
resolves.
3.3. Empirical Model
In our main tests of Hypotheses H1 and H2, we estimate annual cross-sectional regressions
of our insurer-security-year level measure of the timeliness of OTT impairments, TMi,s,t, on an
indicator13 for property casualty insurer, PCi; indicators for securities with NAIC designations of
13 All indicator variables in the paper take a value of one if the condition is present and zero otherwise.
19
1 and 2 (NAIC_1_2s,t), from 3 to 5 (NAIC_3_5s,t), and of 6 (NAIC_6s,t); interactions of PC and the
NAIC designation indicators; and insurer- and security-specific control variables collectively
denoted Xc:
TMi,s,t = α + γ3-5NAIC_3_5s,t + γ6NAIC_6s,t + β1-2 (PCi × NAIC_1_2s,t)
+ β3-5 (PCi × NAIC_3_5s,t) + β6 (PCi × NAIC_6s,t) + ∑ 𝜔𝑐𝑁𝑐=1 𝑋𝑐 (1)
+ μstate + εi,s,t.
NAIC_1_2s,t is omitted due to the inclusion of an intercept in equation (1). Hypothesis H1 posits
that the coefficient β3-5 on PCi × NAIC_3_5s,t is positive. Hypothesis H2 posits that the coefficient
β1-2 on PCi × NAIC_1_2s,t is positive but less than β3-5. In both hypotheses, PC insurers record
timelier OTT impairments of given non-agency RMBS with the specified NAIC designations than
do life insurers. For reasons discussed in footnote 12, we include PCi × NAIC_6s,t in equation (1),
but we make no hypotheses about the coefficient β6 on this variable.
Many prior studies, most examining banks but some examining insurers, provide evidence
that financial institutions exploit the discretion or ambiguity in financial or regulatory reporting
rules governing fair value measurement, OTT impairment, loan and claim loss provisions, and
other accounting areas to meet income, capital, and tax objectives; see Beatty and Liao (2014) for
a review of the extensive banking literature and Hanley et al. (2017) for a recent study of insurers’
fair value measurement of investment securities.14 Based on these studies, the control variables Xc
include the following insurer characteristics related to their financial condition, the risk of their
investment securities, and their financial reporting incentives. Sizei,t denotes the natural logarithm
of total assets. ROAEi,t denotes one hundred times net income divided by average equity.
Low_RBCRatioi,t denotes an indicator for risk-based capital ratio below the median for the same
14 Following the seminal paper by Petroni (1992), most studies examining accounting discretion by insurers exploit
PC insurers’ unique claim loss reserve triangles required to be reported in their financial and statutory filings. The
triangles in financial filings (statutory filings) enable calculation of (report) revisions of these reserves for the
insurance coverage written in individual accident years.
20
type of insurer in the year. Leveragei,t denotes one hundred times total liabilities divided by total
assets. Liquidi,t denotes one hundred times cash and cash equivalents divided by total assets. Riskyi,t
denotes one hundred times investment securities with NAIC designations from 3 to 6 divided by
total investment securities. Publici,t denotes an indicator for publicly traded.15 All insurer
characteristics are lagged by one year to ensure that insurers’ OTT impairments are not
mechanically related to these characteristics. Given the distinct business models of PC insurers
and life insurers (Ryan 2007, Chapter 13), for the continuous insurer characteristics (Sizei,t,
ROAEi,t, Leveragei,t, Liquidi,t, and Riskyi,t) we subtract the mean of the variable for the same type
of insurer (PC or life) in the same year; this mean adjustment by insurer type does not affect any
of our inferences. We calculate these means using the universe of insurers with data available on
S&P Global Market Intelligence (i.e., not restricting these insurers to be in the final sample or even
to hold non-agency RMBS), so the means of the industry-adjusted variables for the sample
observations generally are not zero.
We also control for two variables that pertain to a given security in a year. PartialSalei,s,t
denotes an indicator for a given insurer partially selling a given security during the year, implying
that the insurer has first-hand recent experience of the security’s market value. Hldrss,t denotes the
number of sample insurers that hold a given security in a given year, which captures the liquidity
of the security and the reliability of our OTT impairment timeliness measure, TMi,s,t.
Equation (1) also include fixed effects for the insurer’s state of domicile, denoted μstate, as
each state has its own insurance regulator and regulatory environment. We estimate equation (1)
using pooled OLS and calculate standard errors clustering observations by insurer.
15 Fifteen mutual or private insurers (1.5% of the sample) become public during our sample period.
21
4. EMPIRICAL RESULTS
4.1. Univariate Tests of Differences of the Mean of TMi,s,t across Types of Insurers by Year
Panel A of Table 3 reports the number of unique securities and the number of total insurer-
security observations for life insurers (left five columns) and for PC insurers (right five columns)
in each of the five sample years from 2007 to 2011. This panel reports consistent but less
aggregated data about the sample than is reported in Table 2. Although the sample contains more
PC insurers (648) than life insurers (352), the number of insurer-security-year observations are
fairly close to equal for the two types of insurer in each year. This panel provides a reference point
for corresponding data in the remaining panels in the table.
Using the same columnar structure as in Panel A, Panels B through E of Table 3 report the
number of unique securities, the number of total insurer-security observations, and the mean and
standard deviation of TMi,s,t for various subsamples of OTT-impaired non-agency RMBS for the
two types of insurer in the five sample years. Panel B reports these data for all OTT-impaired
securities, while Panels C, D, and E report these data for OTT-impaired securities with NAIC
designations from 3 to 5, of 1 or 2, and of 6, respectively. Statistically significant differences in
the mean of TMi,s,t across the two types of insurers in a given year for each subsample are indicated
by asterisks in the columns for PC insurers.
In Panel B for the subsample of all OTT-impaired securities, the mean of TMi,s,t is
significantly higher at the 1 percent level in each year for PC insurers than for life insurers. These
means differences are consistent with both Hypothesis H1 and Hypothesis H2, although only one
of the hypotheses holding could yield these results.
In Panel C for the subsample of all OTT-impaired securities with NAIC designations from
3 to 5, the mean of TMi,s,t is insignificantly different in 2007, significantly higher at the 5 percent
level in 2008, and significantly higher at the 1 percent level in 2009-2011 for PC insurers than for
22
life insurers. The significant differences of the means in 2008–2011 are consistent with Hypothesis
H1. While not posited in Hypothesis H1, the relatively low significance levels in 2007 and to a
lesser extent 2008 likely are driven by the small number of unique OTT-impaired securities with
NAIC designations from 3 to 5 in those years, which limits the power of the tests of means
differences in these years. The number of unique OTT-impaired securities with NAIC designations
from 3 to 5 peaks in 2009, and the corresponding number of insurer/OTT-impaired security
observations peaks in 2010 for both life insurers and PC insurers.
In Panel D for the subsample of all OTT-impaired securities with NAIC designations of 1
or 2, in each sample year the mean of TMi,s,t is significantly higher at the 1 percent level for PC
insurers than for life insurers, consistent with Hypothesis H2. The higher statistical significance in
this panel than in Panel C might appear to be, but is not, inconsistent with the second part of
Hypothesis H2 that the predicted spillover effects of fair value measurement are weaker than the
direct effects predicted in Hypothesis H1. In fact, this higher statistical significance is attributable
to the larger numbers of unique OTT-impaired securities and insurer/OTT-impaired security
observations with NAIC designations of 1 or 2 than of 3 to 5 in each sample year, particularly in
2007 and 2008. Relatedly, the number of unique OTT-impaired securities with NAIC designations
of 1 or 2 peaks in 2008 (not 2009, as for OTT-impaired securities with NAIC designations from 3
to 5), and the number of insurer/OTT-impaired security observations peaks in 2008 (not 2010), for
both life insurers and PC insurers. The differential timing of these peaks reflects securities
migrating to riskier NAIC designations as the effects of the financial crisis manifest.
Consistent with these points, and providing further support for Hypothesis H2, in
untabulated analysis we calculated the difference across PC insurers and life insurers in the means
of TMi,s,t for OTT-impaired securities with NAIC designations from 3 to 5 versus with NAIC
designations of 1 or 2 in each year. These differences are statistically significantly positive at the
23
5 percent level or better in all years except 2011, for which the difference is positive but
insignificant.
Panel E for the subsample of all OTT-impaired securities with NAIC designations of 6 is
included for completeness. For reasons discussed in footnote 12, the differences in the means of
TMi,s,t are insignificant for this subsample.
Table 4, Panel A reports summary statistics—mean, median, standard deviation, mean for
PC insurers, mean for life insurers, and the magnitude and significance of the difference of the
means for PC insurers versus life insurers—for the equation (1) variables for the pooled sample of
24,566 insurer-security-year observations (12,723 for PC insurers and 11,843 for life insurers)
from 2007–2011.16 The panel omits the medians and standard deviations of the indicator variables,
as these statistics follow directly from the corresponding means. The panel also omits the
differences of the means of the continuous insurer characteristic control variables for PC insurers
versus life insurers for reasons discussed below.
The mean of TMi,s,t is 0.60, i.e., the sample insurers on average record 0.60 percent larger
cumulative OTT impairments divided by actual cost than the median insurer holding a given
security in a given year.17 Consistent with the differences in the yearly means of TMi,s,t for PC
insurers and life insurers reported in Table 3, the mean of TMi,s,t is significantly higher at the 1
percent level for PC insurers (1.16) than for life insurers (-0.01). Non-agency RMBS with NAIC
designations of 1 or 2 (from 3 to 5) [of 6] constitute 85 percent (12 percent) [3 percent] of the
16 The means of the equation (1) variables are calculated across insurer-security-year observations regardless of
whether the variables are defined at that level or at more aggregate levels. For example, for the insurer characteristic
control variables, an insurer that holds multiple non-agency RMBS in a given year will be represented multiple times
in that year. Similarly, for the security characteristic control variables, a non-agency RMBS that is held by more than
one insurer in a year will be represented multiple times in that year. 17 The right skewness of TMi,s,t is partly an artifact of (1) its mean being calculated across insurer-security-year
observations, (2) PC insurers recording timelier OTT impairments of given securities than do life insurers, and (3)
there being more PC insurer-security-year observations than life insurer-security-year observations.
24
sample observations. These percentages are very similar for PC insurers and life insurers, although
the differences for NAIC_1_2s,t and NAIC_3_5s,t across the two types of insurer are significant at
the 1 percent level and indicate that PC insurers hold slightly less risky profiles of securities than
do life insurers. Public insurers constitute 44 percent of the sample, with life insurers being slightly
but significantly more likely at the 1 percent level to be public (45 percent) than are PC insurers
(43 percent). Three quarters of the insurer-security-year observations experience partial sales, with
PC insurers being slightly but significantly more likely at the 1 percent level to engage in partial
sales of securities (77 percent) than are life insurers (73 percent). The mean of Hldrss,t is 7.39 and
is significantly higher at the 1 percent level for PC insurers (7.82) than for life insurers (6.94).
We turn to the continuous insurer characteristic control variables (Sizei,t, ROAEi,t,
Leveragei,t, Liquidi,t, and Riskyi,t). Recall that these variables are adjusted for their means for the
same type of insurer in the year. Hence, the differences of the means of these variables for PC
insurers versus life insurers are not readily interpretable, and so the panel omits these differences.
The panel reports the means for PC insurers and life insurers, however, which we discuss when
they yield different insights than the overall mean across both types of insurer.
Compared to the universe of insurers, the sample insurers are on average more leveraged
(mean Leveragei,t = 8.64), larger (mean Sizei,t = 2.35), less liquid (mean Liquidi,t = -7.80), and hold
less risky securities (mean Riskyi,t = -0.66). Compared to this universe, the sample insurers on
average have similar profitability (mean ROAEi,t = -0.03), although the sample PC insurers are
more profitable (mean ROAEi,t = 3.33) and the sample life insurers are less profitable (mean
ROAEi,t = -3.64).
Table 4, Panel B reports the Spearman correlations of the equation (1) variables, excepting
the NAIC designation indicators, for the pooled sample. The correlations of the OTT impairment
timeliness measure, TMi,s,t, with the other variables typically are relatively low in absolute
25
magnitude (a maximum of 0.08) but in many cases are significantly different from zero.
Specifically, TMi,s,t is significantly positively correlated with PCi, consistent with PC insurers
recording timelier OTT impairments of given non-agency RMBS than do life insurers. TMi,s,t is
significantly positively correlated with ROAEi,t at the 5 percent level and significantly negatively
correlated with Leverage, Risky, and Liquid at the 1 percent level, consistent with healthier insurers
recording timelier OTT impairments. The significantly negative correlation of TMi,s,t with Sizei,t at
the 1 percent level appears to conflict with this conclusion; however, this negative correlation is
consistent with the significantly negative correlation of Sizei,t with PCi. TMi,s,t is significantly
negatively correlated with Publici,t at the 10 percent level, consistent with insurers with stronger
reporting incentives recording less timely OTT impairments. TMi,s,t is significantly positively
correlated with Hldrsi,s,t at the 1 percent level, consistent with insurers holding more liquid
securities recording timelier OTT impairments.
The control variables in many cases have correlations with high absolute magnitudes. For
example, PCi is highly and significantly negatively correlated with Leveragei,t (-0.36), Liquids,t
(-0.57), and Riskyi,t (-0.20). Leveragei,t is highly and significantly positively correlated with
Low_RBCRatioi,t (0.23), Publici,t (0.20), and Sizei,t (0.51). Publici,t is highly and significantly
positively correlated with Sizei,t (0.36), and Sizei,t is highly and significantly correlated with Riskyi,t
(0.32). Owing to these high correlations, in the estimations of equation (1) the control variables
often are insignificant despite their significant correlations with TMi,s,t.
4.2. Relative Timeliness of OTT Impairments by PC Insurers versus Life Insurers
As a preliminary before testing Hypotheses H1 and H2, we discuss the annual cross-
sectional estimations of a nested version of equation (1) that does not distinguish security-year
observations by their NAIC designations. The results of these estimations are reported in Table 5,
Panel A. With the exception of a few variable-years, the control variables are insignificant. The
26
coefficient on PCi is significantly positive at the 1 percent level in all five sample years. Hence,
on average PC insurers record timelier OTT impairments of given non-agency RMBS than do life
insurers in each of these years.
Panel B of Table 5 reports the estimation of full equation (1). We use the results of this
estimation to test both hypotheses. The control variables remain with a few exceptions
insignificant. Hypothesis H1 posits that the coefficient β3-5 on PCi × NAIC_3_5s,t is positive, i.e.,
that PC insurers record timelier OTT impairments of non-agency RMBS with NAIC designations
from 3 to 5, which they measure at fair value, than do life insurers, who measure these securities
at amortized cost. Consistent with this hypothesis, and also with the analysis reported in Table 3,
this coefficient is significantly positive at the 5 percent level in 2008 and at the 1 percent level in
2009, 2010, and 2011. Consistent with the analysis in Table 3, the coefficient β3-5 is positive but
insignificant in 2007, reflecting the few unique securities with NAIC designations from 3 to 5 and
corresponding total insurer-security observations in that year.
To give a sense for the economic significance of this finding, the coefficient β3-5 peaks at
6.28 in 2008, indicating that by that year life insurers had cumulatively recorded 6.28 percent
smaller OTT impairments as a percentage of the acquisition cost of non-agency RMBS with NAIC
designations from 3 to 5 than had PC insurers. Although β3-5 decreases gradually in subsequent
years, the number of OTT-impaired non-agency RMBS with NAIC designations from 3 to 5 in our
sample increases sharply from 2008 to 2009 (see Table 3, Panel C). The latter effect dominates the
former, causing the incremental dollar amount of life insurers’ smaller cumulative OTT
impairments of these securities to peak in 2009 at $67 million, 68 percent of life insurers’ reported
27
OTT impairments of the securities of $99 million in 2009, and 5 percent of their acquisition cost
of $1.4 billion for the securities.18
The first part of Hypothesis H2 posits that the coefficient β1-2 on PCi × NAIC_1_2s,t is
positive, i.e., that PC insurers record timelier OTT impairments of non-agency RMBS with NAIC
designations of 1 and 2 than do life insurers, reflecting an information spillover from PC insurers’
fair value measurement of non-agency RMBS with NAIC designations from 3 to 5. Largely
consistent with this part of Hypothesis H2, and also with the analysis reported in Table 3, the
coefficient β1-2 is significantly positive at the 1 percent level in 2007, 2008, and 2009 and at the 5
percent level in 2011, but insignificant in 2010.
To give a sense for the economic significance of this finding, the coefficient β1-2 peaks at
1.15 in 2008, indicating that by that year life insurers had cumulatively recorded 1.15 percent
smaller OTT impairments as a percentage of the acquisition cost of non-agency RMBS with NAIC
designations of 1 or 2 than had PC insurers. The incremental dollar amount of the cumulatively
smaller OTT impairments of these securities for life insurers also peaks in 2008 at $132 million,
107 percent of life insurers’ cumulative OTT impairments of the securities of $123 million in
2008, and over 1 percent of their acquisition cost of $11.5 billion for the securities.
The second part of Hypothesis H2 posits that the coefficient β1-2 on PCi × NAIC_1_2s,t is
smaller than the coefficient β3-5 on PCi × NAIC_3_5s,t, i.e., that the spillover effect of fair value
measurement predicted in the first part of Hypothesis H2 is weaker than the direct effect of fair
value predicted in Hypothesis H1. The difference of β3-5 and β1-2 in each year and Chi-square tests
of the significance of these coefficient differences are reported at the bottom of Table 5, Panel B.
18 We considered calculating economic significance at the insurer level under the assumption that the estimated
coefficients for our (highly restricted) sample of non-agency RMBS generalized to all non-agency RMBS or even to
all debt investment securities. As it is difficult to know whether this assumption is appropriate, particularly for all debt
securities, we decided against making these calculations.
28
Largely consistent with the second part of H2, β3-5 − β1-2 is significantly positive at the 5 percent
level in 2008 and at the 1 percent level in 2009 through 2011, but is insignificant in 2007.
In summary, the results of the tests across the two types of insurer reported in Table 5 are
largely consistent with PC insurers’ (life insurers’) fair value (amortized cost) measurement of
given non-agency RMBS with NAIC designations from 3 to 5 leading PC insurers to record
timelier OTT impairments of both these securities and non-agency RMBS with NAIC designations
of 1 and 2 than do life insurers.
5. SUPPLEMENTAL ANALYSES AND ROBUSTNESS TESTS
5.1. Relative Timeliness of OTT Impairments of Non-agency RMBS with NAIC Designations from
3 to 5 versus of 1 or 2 within Insurer Types
As discussed in the introduction, the main problem of inference confronting this study is
that PC and life insurers have distinct business models (Ryan 2007, Chapter 13). Given our
security-level research design, which addresses life insurers’ tendency to choose longer maturity
and/or less liquid securities than do PC insurers, the primary remaining concern is that life insurers
typically hold longer duration liabilities than do PC insurers and so may have longer intended
holding periods for given investment securities than do PC insurers. Hence, life insurers may
expect to recover more of the deterioration of the fair value of the securities over these longer
holding periods. Under this possibility, SAP may require PC insurers to record more frequent or
larger OTT impairments than it requires of life insurers.
We address this problem of inference in this section and the following section. In this
section, we do so by conducting analyses within each insurer type. Specifically, we compare the
timeliness of OTT impairments of non-agency RMBS with NAIC designations of 1 or 2, which
both types of insurer measure at amortized cost, to the timeliness of OTT impairments of non-
agency RMBS with NAIC designations 3 to 5, which PC insurers measure at fair value and life
29
insurers measure at amortized cost. For essentially the same reasons that underlie the development
of Hypotheses H1 and H2 stated in Section 2.3, we expect that PC insurers’ OTT impairments of
non-agency RMBS with NAIC designations from 3 to 5 are timelier than their OTT impairments
of non-agency RMBS with NAIC designations of 1 or 2. In contrast, we expect no such difference
for life insurers.
To test this expectation for PC insurers, the bottom of Table 5, Panel B reports Chi-square
tests of the sum of the coefficients in equation (1) that capture the timeliness of PC insurers’ OTT
impairments of non-agency RMBS with NAIC designations from 3 to 5 (i.e., the coefficient γ3-5
on NAIC_3_5 plus the coefficient β3-5 on PCi×NAIC_3_5) minus the coefficient that captures the
timeliness of PC insurers’ OTT impairments of non-agency RMBS with NAIC designations of 1
or 2 (i.e., the coefficient β1-2 on PCi×NAIC_1_2). These coefficient differences are insignificant
in 2007 but significantly positive at the 5 percent level in 2008 and the 1 percent level in 2009
through 2011.19 Hence, as expected we find that PC insurers record significantly timelier OTT
impairments of non-agency RMBS with NAIC designations from 3 to 5 than of non-agency RMBS
with NAIC designations of 1 or 2 in all sample years except 2007. To give a sense for the economic
significance of these differences, PC insurers’ cumulative OTT impairments through 2008 scaled
by acquisition cost were 3.34 percent larger for non-agency RMBS with NAIC designations from
3 to 5 than for non-agency RMBS with NAIC designations of 1 and 2.
The coefficient γ3-5 on NAIC_3_5 captures the timeliness of life insurers’ the timeliness of
life insurers’ OTT impairments of non-agency RMBS with NAIC designations from 3 to 5 relative
to the timeliness of their OTT impairments of non-agency RMBS with NAIC designations of 1 or
2. In contrast to the result for PC insurers, γ3-5 is insignificant in all sample years except 2009,
19 As discussed in Section 4.1, this insignificance in 2007 is likely attributable to the relatively few non-agency RMBS
with NAIC designations from 3 to 5 held by PC insurers in that year, which reduces the power of the test.
30
when the coefficient is significantly negative at the 10 percent level. Hence, with this marginally
significant and wrong-signed exception, for life insurers as expected we find statistically
insignificant differences in the timeliness of OTT impairments of non-agency RMBS with NAIC
designations from 3 to 5 versus NAIC designations of 1 or 2.
In summary, this supplemental evidence is inconsistent with the distinct business models
of PC and life insurers explaining our findings. Consistent with Hypotheses H1 and H2, we find
that PC insurers record timelier OTT impairments of non-agency RMBS with NAIC designations
from 3 to 5, which they measure at fair value, than of non-agency RMBS with NAIC designations
of 1 or 2, which they measure at amortized cost. In contrast, we find that the timeliness of life
insurers’ OTT impairments does not vary for non-agency RMBS with NAIC designations from 3
to 5 versus NAIC designations of 1 or 2, both of which they measure at amortized cost.
5.2. Relative Timeliness of OTT Impairments by PC Insurers versus Life Insurers, Requiring a
Minimum Investment Horizon
In this section, we alternatively address the main problem of inference by comparing the
timeliness of OTT impairments for the subsample of non-agency RMBS for which at least one PC
insurer and one life insurer do not sell all their holdings of the security in the following one or two
years, reasonable horizons over which insurers could feasibly evaluate their intent to hold
securities. Table 6 reports the results for the one-year horizon; depending on the year, requiring
this horizon reduces the sample size by about 10 to 20 percent compared to the sample in the
primary results reported in Table 5. The untabulated results for the two-year horizon are similar
despite the fact that requiring this horizon reduces the sample size by about 30 to 35 percent
depending on the year. As in Panel B of Table 5, Table 6 reports the estimation of the full equation
(1). As the coefficients on the control variables are similar to those reported in Table 5, to conserve
space Table 6 reports only the coefficients on the variables of interest.
31
Consistent with Hypothesis H1, the coefficient β3-5 on PCi × NAIC_3_5s,t in the estimation
of the full equation (1) reported in Table 6 is significantly positive at the 5 percent level in 2008
and at the 1 percent level in 2009 through 2011. Consistent with the first part of Hypothesis H2,
the coefficient β1-2 on PCi × NAIC_1_2s,t in the estimation of the full equation (1) reported in Panel
B of Table 6 is positive and significant at the 5 percent level in 2007, at the 1 percent level in 2008,
2009, and 2011, and at the 10 percent level in 2010. The difference of β3-5 and β1-2 in each year and
Chi-square tests of the significance of these coefficient differences are reported at the bottom of
Table 6. Largely consistent with the second part of Hypothesis H2, the incremental timeliness of
OTT impairments by PC insurers versus life insurers is larger for non-agency RMBS with NAIC
designations from 3 to 5 than for those with NAIC designations of 1 or 2; β3-5 − β1-2 is significantly
positive at the 10 percent level in 2007, the 5 percent level in 2008 and the 1 percent level in 2009
and 2010, but is insignificant in 2011.
In summary, this supplemental evidence is inconsistent with the longer investment horizon
of life insurers than of PC insurers explaining our findings. Consistent with Hypotheses H1 and
H2, we find that PC insurers record timelier OTT impairments of non-agency RMBS with NAIC
designations both from 3 to 5 and of 1 and 2 than do life insurers.
5.3. Analysis of Credit Downgrade Subsample
The idea motivating our primary tests reported in Table 5 is that PC insurers’ recurring fair
value measurement of non-agency RMBS with NAIC designations from 3 to 5 leads them to
develop superior information and control systems to those of life insurers, who measure the
securities at amortized cost. A possibility that would work against this idea is that downgrades of
non-agency RMBS to lower (non-investment grade) NAIC designations may induce life insurers
either to develop or to obtain access to third party systems for these securities. Such downgrades
indicate (significant) deteriorations in the credit quality of securities and thus (significantly)
32
increased likelihoods that holders should record OTT impairments of the securities or accrue
interest revenue under EITF 99-20. However, even were this possibility true, we expect that PC
insurers’ longer-standing and more comprehensive systems would provide more discipline over
their OTT impairments of the securities. To rule out this possibility, in this section we conduct
analyses to ensure the robustness of our primary findings to this possibility by comparing the
timeliness of PC insurers’ and life insurers’ OTT impairments of the subsample of non-agency
RMBS whose NAIC designations experienced downgrades after 2006.
Table 7, Panel A reports the total number of cumulative downgrades of the sample non-
agency RMBS from 2006 to each sample year. Tracking the increasing severity and cumulative
adverse effects of the financial crisis, this number first rises sharply from 20 securities (1.3 percent
of the sample) in 2007, to 176 securities (12.9 percent of the sample) in 2008, and to 451 securities
(39.8 percent of the sample) in 2009. This number then falls more gradually to 308 (38.8 percent
of the sample) in 2011, owing to sample attrition discussed in Section 3.1.
Table 7, Panel B reports the percentages of the total number of cumulative downgrades up
to each sample year that are attributable to downgrades from each NAIC designation to each lower
NAIC designation. Because insurers mostly held NAIC designation 1 non-agency RMBS in 2006,
the vast majority of these cumulative downgrades are from that designation. We do not discuss
2007 because of the few downgrades that occurred in that year. In 2008, i.e., relatively early in the
crisis, 52.3 percent of total downgrades were from NAIC designation 1 to NAIC designation 2.
This percentage monotonically diminishes to 24.4 percent in 2011. As a consequence, the
percentage of total downgrades from NAIC designation 1 to the non-investment grade NAIC
designations from 3 to 6 increases from 41.5 percent of total downgrades in 2008 to 70.1 percent
of total downgrades in 2011.
33
Table 8 reports the estimation of the full equation (1) for the downgrade subsample. As the
coefficients on the control variables are similar to those reported in Table 5, to conserve space
Table 8 reports only the coefficients of the variables of interest.
Consistent with Hypothesis H1 and also with the analysis reported in Panel B of Table 5,
the coefficient β3-5 on PCi × NAIC_3_5s,t is positive at the 5 percent level in 2008 and at the 1
percent level in 2009 through 2011. Also consistent with the prior analysis, β3-5 is insignificant in
2007, owing to the relatively few total insurer-downgraded security observations in 2007. Hence,
on average PC insurers record timelier OTT impairments of given downgraded non-agency RMBS
with NAIC designations from 3 to 5 than do life insurers in the last four years of the sample period.
Only weakly consistent with the first part of Hypothesis H2, the coefficient β1-2 on PCi ×
NAIC_1_2s,t is positive and significant at the 10 percent level in 2008 and at the 5 percent level in
2009. The difference of β3-5 and β1-2 in each year and Chi-square tests of the significance of these
differences are reported at the bottom of Table 8. Consistent with the second part of Hypothesis
H2, β3-5 − β1-2 is significantly positive at the 5 percent level in 2007 and at the 1 percent level in
2009 through 2011.
In summary, our primary results are robust to restricting the sample to non-agency RMBS
that have experienced downgrades. Consistent with Hypotheses H1 and H2, we find that PC
insurers record timelier OTT impairments of downgraded non-agency RMBS with NAIC
designations both from 3 to 5 and of 1 and 2 than do life insurers. Hence, life insurers do not
appear to invest sufficiently in systems to record OTT impairments of downgraded non-agency
RMBS on as timely a basis as do PC insurers.
5.4. Alternative Measure of the Timeliness of OTT Impairments
In this section, we examine the robustness of our findings to the use of an alternative
measure of the timeliness of OTT impairments, denoted TM_Alti,s,t, which we calculate in two
34
steps. First, similar to Vyas (2011), we divide the cumulative OTT impairment recorded by an
insurer on a given non-agency RMBS by the end of each sample year from 2007 to 201020 by the
cumulative OTT impairment recorded by the insurer on the security by the end of 2011;21 this step
is based on the assumption that insurers fully record OTT impairments of their non-agency RMBS
attributable to the financial crisis by the end of 2011. Second, we subtract from this cumulative
OTT impairment ratio the median of the ratio for all insurers holding that security in that sample
year, so that TM_Alti,s,t captures the timeliness of insurers’ OTT impairments relative to those of
other insurers holding the same security.
Table 9 reports the results of the estimation of the full equation (1) replacing the dependent
variable TMi,s,t in equation (1) with TM_Alti,s,t. To conserve space, Table 9 reports only the
coefficients of the variables of interest.
Consistent with Hypothesis H1, the coefficient β3-5 on PCi × NAIC_3_5s,t is significantly
positive at the 1 percent level in 2008 through 2010 but insignificant in 2007. Hence, using the
alternative measure of the timeliness of OTT impairments, PC insurers generally record timelier
OTT impairments of non-agency RMBS with NAIC designations from 3 to 5 than do life insurers.
To give a sense for the economic significance of this finding, the coefficient of 19.34 on the
PCi×NAIC_3_5s,t in 2008 indicates that by that year PC insurers cumulatively recorded 19.34
percent more of their cumulative OTT impairments through 2011 than did life insurers. Although,
by the construction of TM_Alti,s,t, life insurers must catch up on this OTT impairment timeliness
measure by the end of 2011, PC insurers record consistently timelier OTT impairments through
the end of 2010.
20 We exclude 2011, because TM_Alti,s,t equals 1 by construction in that year. 21 To be able to calculate TM_Alti,s,t, we require (1) an insurer to hold the security throughout the 2007-2011 sample
period and (2) the insurer to record at least one OTT impairment on the security during that period.
35
Only very weakly consistent with the first part of Hypothesis H2, the coefficient β1-2 on
PCi × NAIC_1_2s,t is significantly positive only at the 10 percent level in 2007. The difference of
β3-5 and β1-2 in each year and Chi-square test of the significance of this difference is reported at the
bottom of Table 9. Consistent with the second part of Hypothesis H2, β3-5 − β1-2 is significantly
positive at the 1 percent level in 2008 through 2010, although this difference is negative and
significant at the 10 percent level in 2007.
In summary, the use of the alternative measure of the timeliness of OTT impairments
generally does not affect our inferences regarding Hypothesis H1 or the second part of Hypothesis
H2, but it weakens our inferences regarding the first part of Hypothesis H2. In particular, we
continue to find that PC insurers record timelier OTT impairments of non-agency RMBS with
NAIC designations from 3 to 5 than do life insurers, and this incremental timeliness is significantly
greater than for OTT impairments of non-agency RMBS with NAIC designations from 3 to 5 than
with NAIC designations of 1 or 2.
6. CONCLUSION
Some accounting rules require reporting firms to measure their financial instruments at fair
value on a recurring basis, while others require amortized cost measurement. Recurring fair value
measurement necessitates that firms evaluate economic conditions affecting their financial
instruments each quarter. To make these evaluations in an adequately timely and reliable fashion,
firms must invest in information and control systems. In contrast, amortized cost is a largely pre-
determined accounting measurement approach that typically does not require reporting firms to
pay attention to economic conditions affecting their financial instruments. OTT impairments of
non-trading investment securities constitute the most significant exception to the pre-determined
nature of amortized cost measurement; the accounting rules governing these impairments are
36
intended to ensure that sufficiently bad news is incorporated into firms’ recognized measurements
of the securities. Given the pre-determined nature of amortized cost measurement, however, the
question arises whether reporting firms have adequate information and control systems in place to
record timely OTT impairments of the securities.
In this study, we investigate this question in the context of statutory accounting by insurers,
owing to three facts: (1) statutory accounting principles require PC insurers to measure investment
securities with NAIC designations from 3 to 5 at fair value, but life insurers to measure these
securities at amortized cost; (2) the two types of insurer are subject to the same OTT impairment
accounting rules; and (3) statutory reporting rules require both types of insurer to provide detailed
security-level accounting information in their annual and quarterly statutory filings. We exploit
these facts to develop a sample of non-agency RMBS held by at least one PC insurer and one life
insurer during our 2007-2011 sample period, and to observe the relative amounts and timing of
different PC and life insurers’ OTT impairments of given non-agency RMBS during this period.
We hypothesize and provide evidence that requirements for PC insurers to measure non-
agency RMBS with NAIC designations from 3 to 5 at fair value lead them to record timelier OTT
impairments of given securities than did life insurers, who measure these securities at amortized
cost. We further hypothesize and provide evidence that PC insurers’ fair value measurement of
non-agency RMBS with NAIC designations from 3 to 5 improves their understanding of similar
non-agency RMBS with NAIC designations of 1 and 2, leading PC insurers to record timelier OTT
impairments of NAIC 1 and 2 designated non-agency RMBS than did life insurers during our
sample period. Lastly, we hypothesize and provide evidence that this spillover effect of fair value
measurement of similar securities is weaker than the direct effect of fair value measurement of the
identical security.
37
The main problem of inference confronting this study is that life insurers may have longer
intended holding periods for given investment securities than do PC insurers. Under this
possibility, life insurers may expect to recover more of the deterioration of the fair value of the
securities over their longer holding periods, so SAP may require PC insurers to record more
frequent or larger OTT impairments than SAP requires of life insurers.
We address this problem of inference by conducting two supplemental analyses. First, we
conduct analyses within each insurer type. We predict and find that PC insurers’ OTT impairments
of non-agency RMBS with NAIC designations from 3 to 5 were timelier than their OTT
impairments of non-agency RMBS with NAIC designations of 1 or 2 during our sample period,
whereas we predict and find no such difference for life insurers. Second, we restrict the sample to
non-agency RMBS for which at least one PC insurer and one life insurer did not sell all their
holdings of the security in the following one or two years, reasonable horizons over which insurers
could feasibly evaluate their intent to hold securities. We find results consistent with those of the
main analysis. We also show that our results are robust to restricting the sample to downgraded
securities and to the use of an alternative measure of the timeliness OTT impairments attributable
to Vyas (2011).
To the best of our knowledge, our study is the first to show that fair value measurement
leads to timelier recognition of any type of non-recurring write-downs of financial assets than does
amortized cost measurement. This finding is consistent with the requirement for firms to measure
investment securities at fair value leading them to invest in information and control systems that
increase the timeliness of their OTT impairments of the securities. This finding suggests that these
systems enhance the overall quality of financial statements, not just the quality of fair value
estimates. In this vein, we conjecture that these systems may also improve the overall transparency
38
of firms’ financial reporting, e.g., MD&A and footnote disclosures, a promising topic for future
research in our view.
39
APPENDIX A
The table defines all the variables in equation (1) in alphabetical order.
Variable Definition
Hldrss,t Number of insurers holding security s in year t.
Leveragei,t One hundred times total liabilities for insurer i in year t divided by total assets for insurer
i in year t minus the mean of this variable for all insurers of the same type (PC or life)
in year t.
Liquidi,t One hundred times cash equivalents for insurer i in year t divided by total assets for
insurer i in year t minus the mean of this variable for all insurers of the same type (PC
or life) in year t.
Low_RBCRatioi,t Indicator variable that takes a value of 1 if the risk based capital ratio for insurer i in year
t is below the median for all insurers of the same type (PC or Life) in year t and 0
otherwise.
NAIC_1_2s,t Indicator variable that takes a value of 1 if security s has a NAIC designation of 1 or 2
in year t and 0 otherwise.
NAIC_3_5s,t Indicator variable that takes a value of 1 if security s has a NAIC designation from 3 to
5 in year t and 0 otherwise.
NAIC_6s,t Indicator variable that takes a value of 1 if security s has a NAIC designation of 6 in year
t and 0 otherwise.
PartialSalei,s,t Indicator variable that takes a value of 1 if insurer i partially sells security s during year
t and 0 otherwise.
PCi Indicator variable that takes a value 1 for a PC insurer and 0 for a life insurer.
Publici Indicator variable that takes a value of 1 for a publicly listed insurer and 0 otherwise.
Riskyi,t One hundred times insurer i’s investment securities with (non-investment grade) NAIC
designations from 3 to 6 in year t divided by insurer i’s total investment securities in
year t minus the mean of this variable for all insurers of the same type (PC or life) in
year t.
ROAEi,t One hundred times net income for insurer i in year t divided by average equity for insurer
i in year t minus the mean of this variable for all insurers of the same type (PC or life)
in year t.
Sizei,t Natural logarithm of total assets for insurer i in year t minus the mean of this variable
for all insurers of the same type (PC or life) in year t.
TMi,s,t One hundred times insurer i’s cumulative OTT impairment recorded on security s from
2007 to the end of year t divided by insurer i’s actual cost of the security at the end of
2006 minus the median of this variable across all insurers holding security s at the end
of year t.
TM_Alti,s,t One hundred times insurer I’s cumulative OTT impairment on security s from 2007 to
the end of year t divided by the cumulative OTT impairment recorded by insurer i on
the security from 2007 by the end of 2011 minus the median of this variable for all
insurers holding the security in year t.
40
REFERENCES
Ahmed, A., E. Kilic, and G. Lobo. 2006. Does recognition versus disclosure matter? Evidence
from value‐relevance of banks’ recognized and disclosed derivative financial instruments.
The Accounting Review 81 (3): 567-588.
Barth, M. 2014. Measurement in financial reporting: The need for concepts. Accounting
Horizons 28 (2): 331-52.
Beatty, A., and S. Liao. 2011. Do delays in expected loss recognition affect banks’ willingness to
lend? Journal of Accounting and Economics 51 (1): 1-20.
Beatty, A., and S. Liao. 2014. Financial accounting in the banking industry: A review of the
empirical literature. Journal of Accounting & Economics 58: 339-383.
Bennett, N. 2013. Life risk-based capital and the asset valuation reserve. Presentation to the
NAIC’s IRBC Working Group. December 5. Available at
http://www.actuary.org/files/RBC_AVR_Presentation_to_NAIC_IRBCWG_12-5-13.pdf.
Berry-Stölzle, T., G. Nini, and S. Wende. 2014. External financing in the life insurance industry:
Evidence from the financial crisis. Journal of Risk and Insurance 81: 529-562.
Bhat, G., and S. Ryan. 2015. The impact of risk modeling on the market perception of banks’
estimated fair value gains and losses for financial instruments. Accounting, Organizations
and Society 46: 81-95.
Bhat, G., S. Ryan, and D. Vyas. 2017. The implications of credit risk modeling for banks’ loan
loss provisions and loan-origination procyclicality. Working paper, New York
University. Available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1978409.
Bond Market Association and American Securitization Forum. 2006. An analysis and description
of pricing and information sources in the securitized and structured finance markets.
October.
Bushman, R., and C. Williams. 2012. Accounting discretion, loan loss provisioning, and
discipline of banks’ risk taking. Journal of Accounting and Economics 54 (1): 1-18.
Bushman, R., and C. Williams. 2015. Delayed expected loss recognition and the risk profile of
banks. Journal of Accounting Research 53: 511-553.
Dickey, J., M. King, and K. Shih. 2008. Subprime-related securities litigation: Where do we go
from here? Insights: The Corporate Securities Law Advisor 22 (4): 2-10.
Einhorn, D. 2008. Accounting ingenuity. Speech at Ira W. Sohn Investment Research
Conference, May 21.
41
Ellul, A., C. Jotikasthira, C. Lundblad, and Y. Wang. 2015. Is historical cost accounting a
panacea? Market stress, incentive distortions, and gains trading. The Journal of Finance
70 (1): 2489-2538.
Hanley, K., A. Jagolinzer, and S. Nikolova. 2017. Strategic reporting of asset fair values.
Working Paper. Available at
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2889160
Hu, J. 2001. Basics of mortgage-backed securities, 2nd Edition, John Wiley & Sons, Hoboken NJ.
Laux, C., and C. Leuz. 2009. The crisis of fair value accounting: Making sense of the recent
debate. Accounting, Organizations and Society 34: 826-34.
Laux, C., and C. Leuz. 2010. Did fair value accounting contribute to the crisis? Journal of
Economic Perspectives 24: 93-118.
Müller, M., E. Riedl, T. Sellhorn. 2015. Recognition versus disclosure of fair values. The
Accounting Review 90 (6): 2411-2447.
Petroni, 1992. Optimistic reporting in the property-casualty insurance industry. Journal of
Accounting and Economics 15: 485-508.
Ryan, S. 2007. Financial instruments and institutions: Accounting and disclosure rules, 2nd
edition. John Wiley and Sons, Hoboken, NJ.
Ryan, S. 2008. Fair value accounting: Understanding the issues raised by the credit crunch.
Council of Institutional Investors, White Paper.
Ryan, S. 2017. Do the effects of accounting requirements on banks’ regulatory capital adequacy
undermine financial stability? Annual Review of Financial Economics, forthcoming, 9.
Stephanou, C. 2010. Rethinking market discipline in banking: Lessons from the financial crisis.
Policy Research Working Paper 5227, The World Bank.
Vyas, D. 2011. The timeliness of accounting write-downs by U.S. financial institutions during
the financial crisis of 2007-2008. Journal of Accounting Research 49 (3): 823-60.
Wyatt, A. 1991. The SEC says: Mark to market! Accounting Horizons 5: 80-84.
42
Table 1: Sample Construction
This table summarizes the sample construction. The sample period covers the years from 2007 to 2011. # Securities refers to the number of unique non-
agency RMBS held by any insurer. # Firms refers to the number of insurers. # Positions refers to the number of insurer-non-agency RMBS-year
observations. All refers to any insurer, PC refers to property-casualty insurers, and Life refers to life insurers.
` # Firms # Positions
Sample Screens # Securities All PC Life All PC Life
Step 0: All insurers on S&P Global Market 22,567 5,552 3,860 1,692
Intelligence database during 2007-2011.
Step 1: Remove insurer-years that do not 22,567 1,723 1,188 535 172,397 46,726 125,671
hold non-agency RMBS.
Step 2: Remove insurer-securities with 12,826 1,376 910 466 95,477 25,629 69,848
purchases after 2006.
Step 3: Remove insurer-years with risk-based 12,675 1,173 756 417 90,090 23,150 66,940
capital ratios outside 200% and 2000%
Step 4: Remove security-years without a least one PC insurer and 1,590 1,000 648 352 24,566 12,723 11,843
one life insurer holding the security during the year.
43
Table 2: Summary of OTT Impairments
This table presents the following statistics for each sample year from 2007 to 2011: the number of sample non-agency RMBS (#All Secs); the number
of insurers holding at least one non-agency RMBS (# Firms); the average number of firms holding a given non-agency RMBS (Avg # Firms per Sec);
the number of OTT-impaired non-agency RMBS (# Imp Secs); the average number of insurers holding a given OTT-impaired non-agency RMBS (Avg
# Firms per Imp Sec); the sum of the cumulative OTT impairment recorded by all insurers on a given security from 2007 to the end of the year divided
by the sum of the actual cost recorded by all insurers in 2006 (Agg CumImp % on Imp Sec); the average across insurers of the cumulative impairment
recorded for a given security divided by the actual cost of the security recorded in 2006 (Avg CumImp % on Imp Sec); the average across securities of
the standard deviation across insurers holding a given security in the year of the cumulative impairment recorded for the security divided by the actual
cost of the security recorded in 2006 (Avg Std Dev Imp Sec); and the interquartile range across securities of the standard deviation across insurers
holding a given security in the year of the cumulative impairment recorded for a given security divided by the actual cost of the security recorded in
2006 (IQR Std Dev Imp Sec). A non-agency RMBS is deemed OTT-impaired during a year if at least one insurer has recorded an OTT impairment for
the security.
Avg Avg # firms Agg Avg Avg IQR
# All # # firms # Imp per Imp CumImp (%) CumImp (%) Std Dev Std Dev
Year Secs Firms per Sec Secs Sec on Imp Secs on Imp Secs Imp Sec Imp Sec
2007 1,563 948 4.52 113 5.13 2.33 4.76 5.85 6.03
2008 1,366 879 4.44 309 5.04 10.66 13.49 16.02 17.80
2009 1,133 759 4.10 409 4.54 10.11 14.56 10.49 14.54
2010 940 665 3.94 390 4.26 13.26 16.60 9.86 12.94
2011 794 604 3.88 363 4.00 14.32 17.51 9.41 12.48
44
Table 3: Timeliness of OTT Impairments
This table presents statistics for the timeliness measure TM by type of insurer (Life or PC) and by year. Panel A presents the number of securities (#
Sec) and number of insurer-security observations (# Obs) for all sample securities. Panel B presents # Sec, # Obs, the mean of TM (Mean), and the
standards deviation of TM (Std Dev) for all OTT-impaired securities. Panel C, D and E present # Sec, # Obs, Mean, and Std Dev for OTT-impaired
securities with NAIC designations from 3 to 5, of 1 or 2, and of 6, respectively. Significant differences in Mean for Life insurers and PC insurers
are designated by asterisks next to Mean for PC insurers. Statistical significance at the 1%, 5% and 10% levels is indicated by ***, ** and *,
respectively. See Appendix A for the definition of TM.
Life PC
2007 2008 2009 2010 2011 2007 2008 2009 2010 2011
Panel A: All Securities
# Sec 1,563 1,366 1,133 940 794
# Obs 3,370 2,891 2,246 1,826 1,510 3,696 3,178 2,401 1,877 1,571
Panel B: All OTT-Impaired Securities
# Sec 113 309 409 390 363
# Obs 265 693 848 792 692 315 865 1,008 869 759
Mean -1.27 0.52 0.34 -0.05 -0.39 1.93*** 4.64*** 4.20*** 3.88*** 3.55***
Std Dev 10.80 18.25 12.95 12.21 11.63 5.98 16.33 13.75 12.87 12.77
Panel C: OTT-Impaired Securities with NAIC Designations from 3 to 5
# Sec 9 46 138 135 124
# Obs 16 103 296 327 276 40 97 312 318 236
Mean -5.91 -2.61 -1.29 -0.61 -0.78 1.69 6.75** 5.36*** 4.84*** 4.05***
Std Dev 29.68 28.64 12.07 11.70 11.24 7.96 24.16 13.51 12.28 11.52
Panel D: OTT-Impaired Securities with NAIC Designations of 1 or 2
# Sec 102 253 237 215 198
# Obs 247 575 481 386 344 272 755 621 458 432
Mean -0.64 1.15 0.80 0.37 -0.13 1.97*** 4.46*** 3.68*** 3.10*** 3.42***
Std Dev 6.43 15.36 11.35 9.34 9.16 5.67 15.02 12.96 11.59 12.86
Panel E: OTT-Impaired Securities with NAIC Designation of 6
45
# Sec 2 10 34 40 41
# Obs 2 15 71 79 72 3 13 75 93 91
Mean -42.30 -2.29 3.94 0.30 -0.17 1.50 -0.42 3.68 4.43 2.91
Std Dev 59.84 26.69 22.50 22.50 20.51 2.60 17.05 19.76 19.21 15.29
46
Table 4: Summary Statistics
Panel A of this table presents the means, medians, standard deviations, means for PC insurers, means for
life insurers, and the difference of the means for PC and life insurers for all the variables in equation (1).
Panel B presents Spearman correlations for all the equation (1) variables except for the NAIC designation
indicators. Definitions of the equation (1) variables are provided in Appendix A. Differences in the means
for PC and life insurers in Panel A and Spearman correlations in Panel B that are statistically significant at
the 1%, 5% and 10% levels are indicated by ***, ** and *, respectively.
Panel A: Summary Statistics
Std Mean: Mean: Diff:
Variable Mean Median Dev PC Life PC - Life
PC 0.52 -- -- -- --
TM 0.60 0.00 7.47 1.16 -0.01 1.17***
NAIC_1_2 0.85 -- -- 0.86 0.85 0.01***
NAIC_3_5 0.12 -- -- 0.11 0.12 -0.02***
NAIC_6 0.03 -- -- 0.03 0.03 0.00
Low_RBCRatio 0.53 -- -- 0.56 0.50 0.06***
Public 0.44 -- -- 0.43 0.45 -0.02***
PartialSale 0.75 -- -- 0.77 0.73 0.04***
Hldrs 7.39 5.00 7.74 7.82 6.94 0.88***
Leverage 8.64 12.03 12.03 5.54 11.96 --
Size 2.35 2.24 2.19 2.00 2.72 --
Liquid -7.80 -8.21 6.00 -9.88 -5.55 --
Risky -0.66 -0.99 7.17 -1.10 -0.19 --
ROAE -0.03 1.56 24.04 3.33 -3.64 --
47
Table 4: Summary Statistics (continued)
Panel B: Correlations
PC TM Low_RBCRatio Public PartialSale Hldrs Leverage Size Liquid Risky ROAE
PC 1
TM 0.08*** 1
Low_RBCRatio 0.06*** 0.00 1
Public -0.02*** -0.01* 0.18*** 1
PartialSale 0.05*** 0.01 0.01 0.05*** 1
Hldrs 0.09*** 0.02*** -0.02*** -0.05*** 0.04*** 1
Leverage -0.36*** -0.05*** 0.23*** 0.20*** -0.01 -0.10*** 1
Size -0.17*** -0.03*** 0.03*** 0.36*** 0.01 -0.22*** 0.51*** 1
Liquid -0.57*** -0.03*** 0.02*** -0.05*** -0.01 -0.03*** 0.08*** -0.20*** 1
Risky -0.20*** -0.02*** -0.05*** -0.05*** -0.02*** -0.18*** 0.12*** 0.32*** -0.06*** 1
ROAE 0.10** 0.01** -0.02*** 0.18*** 0.00 0.03*** -0.02** 0.06*** -0.13*** -0.02*** 1
48
Table 5: Timeliness of PC versus Life Insurers
This table examines the incremental timeliness of OTT impairments of non-agency RMBS for PC insurers
over life insurers. It reports annual cross-sectional estimations of equation (1) on the full sample. Panel A
reports estimations of a nested version of equation (1) that does not distinguish security-year observations
by NAIC designations. Panel B reports estimations of the full equation (1). Panel B reports a Chi-Square
test that assesses the significance of the difference across PC and life insurers in the timeliness of OTT
impairments of non-agency RMBS with NAIC designations from 3 to 5 versus OTT impairments of non-
agency RMBS with NAIC designations of 1 or 2 [i.e., (β3-5-β1-2) > 0]. This panel also reports a Chi-Square
test that assesses the significance of the incremental timeliness of OTT impairments of non-agency RMBS
with NAIC designations from 3 to 5 versus OTT impairments of non-agency RMBS with NAIC
designations of 1 or 2 within PC insurers [i.e., (β3-5+γ3-5-β1-2) > 0]. Equation (1) includes fixed effects for
the state of domicile of each firm insurer. Standard errors are calculated clustering observations by insurer.
t and χ2 statistics that are statistically significant at the 1%, 5% and 10% levels are indicated by ***, **
and *, respectively. Definitions of the equation (1) variables are provided in Appendix A.
Panel A: Effect of Insurer Type without Distinguishing NAIC Designation
(1) (2) (3) (4) (5)
Variable 2007 2008 2009 2010 2011
PC 0.29*** 1.39*** 1.81*** 1.61*** 1.61***
(3.34) (3.02) (3.61) (2.98) (3.16)
Low_RBCRatio 0.15* -0.19 -0.45 -0.07 -0.74*
(1.63) (-0.52) (-0.86) (-0.16) (-1.67)
Leverage -0.00 -0.01 -0.01 -0.02 -0.03
(-0.97) (-0.37) (-0.46) (-0.78) (-1.37)
Size 0.00 0.07 -0.17 -0.02 0.02
(0.20) (0.52) (-0.82) (-0.19) (0.15)
Liquid 0.00 0.00 0.03 -0.00 -0.03
(0.58) (0.05) (1.25) (-0.05) (-0.78)
Risky 0.02* 0.04 0.02 -0.02 0.06
(1.64) (1.19) (0.49) (-0.44) (1.47)
Hldrs -0.00 0.00 0.01 0.01 0.00
(-0.09) (0.10) (0.76) (0.45) (0.20)
PartialSale -0.01 0.50 -0.48 -0.49 0.55
(-0.15) (1.54) (-1.27) (-0.91) (0.56)
Public -0.12 0.42 -0.25 -0.22 0.15
(-1.27) (0.95) (-0.42) (-0.49) (0.31)
ROAE 0.00 -0.03 0.00 0.01 0.02*
(1.34) (-1.50) (0.06) (0.86) (1.90)
Intercept -0.46*** -2.08*** -1.37** -1.62*** -2.57**
(-3.09) (-3.23) (-2.04) (-2.71) (-2.43)
Observations 7,066 6,069 4,647 3,703 3,081
R2 0.014 0.027 0.040 0.043 0.057
FE-State YES YES YES YES YES
Clustering - Firm YES YES YES YES YES
49
Table 5: Timeliness of PC versus Life Insurers (continued)
Panel B: Effect of Insurer Type Distinguishing NAIC Security Designations
(1) (2) (3) (4) (5)
Variable 2007 2008 2009 2010 2011
PC * NAIC_3_5 3.45 6.28** 4.78*** 3.63*** 2.78***
[β3-5] (1.13) (2.06) (5.42) (4.05) (3.19)
PC * NAIC_1_2 0.21*** 1.15*** 1.13*** 0.78 1.12**
[β1-2] (2.80) (2.65) (2.47) (1.59) (2.23)
PC * NAIC_6 29.26 2.04 0.39 3.59 2.38
[β6] (1.27) (0.29) (0.14) (1.09) (0.82)
NAIC_3_5 -2.37 -1.79 -1.07* -0.43 -0.24
[γ3-5] (-0.81) (-0.75) (-1.87) (-0.73) (-0.39)
NAIC_6 -27.99 -2.47 2.31 0.11 0.30
[γ6] (-1.22) (-0.42) (1.16) (0.04) (0.13)
Low_RBCRatio 0.13* -0.21 -0.57 -0.12 -0.79*
(1.74) (-0.58) (-1.09) (-0.29) (-1.79)
Leverage -0.00 -0.01 -0.01 -0.02 -0.03
(-0.95) (-0.29) (-0.36) (-0.74) (-1.35)
Size -0.00 0.06 -0.18 -0.02 0.00
(-0.01) (0.47) (-0.88) (-0.20) (0.03)
Liquid 0.00 0.00 0.03 -0.01 -0.04
(0.55) (0.04) (1.05) (-0.17) (-0.89)
Risky 0.02 0.05 0.02 -0.03 0.06
(1.62) (1.19) (0.35) (-0.57) (1.44)
Hldrs -0.00 0.00 0.01 -0.00 -0.00
(-0.74) (0.05) (0.32) (-0.21) (-0.20)
PartialSale -0.01 0.55* -0.45 -0.21 0.81
(-0.15) (1.76) (-1.22) (-0.46) (0.92)
Public -0.09 0.40 -0.25 -0.24 0.19
(-1.07) (0.92) (-0.42) (-0.53) (0.37)
ROAE 0.00 -0.02 0.00 0.01 0.02*
(1.38) (-1.49) (0.05) (0.78) (1.91)
Intercept -0.37*** -1.83*** -0.63 -1.20** -2.56***
(-2.82) (-2.87) (-1.02) (-2.25) (-2.66)
χ2 Test: (β3-5-β1-2) > 0 3.24 5.13** 3.64*** 2.84*** 1.66***
NAIC 3_5 - NAIC 1_2 for PC-life (1.06) (1.71) (4.76) (3.54) (1.97)
χ2 Test: (β3-5+γ3-5-β1-2) > 0 0.87 3.34** 2.57*** 2.42*** 1.43***
NAIC 3_5 - NAIC 1_2 for PC (0.96) (1.83) (4.84) (4.08) (2.21)
Observations 7,066 6,069 4,647 3,703 3,081
R2 0.074 0.032 0.050 0.052 0.060
FE-State YES YES YES YES YES
Clustering – Firm YES YES YES YES YES
50
Table 6: Controlling for Differences in Investment Horizons
This table examines the relative timeliness of OTT impairments of non-agency RMBS for PC insurers and
life insurers with similar investment horizons. The table reports annual cross-sectional estimations of the
full equation (1) on the subsample of non-agency RMBS for which at least one PC insurer and one life
insurer do not sell their holdings of the security for at least one year. The table reports a Chi-Square test
that assesses the significance of the difference across PC and life insurers in the incremental timeliness of
OTT impairments of non-agency RMBS with NAIC designations from 3 to 5 versus OTT impairments of
non-agency RMBS with NAIC designations of 1 or 2 [i.e., (β3-5-β1-2) > 0]. This panel also reports a Chi-
Square test that assesses the significance of the incremental timeliness of OTT impairments of non-agency
RMBS with NAIC designations from 3 to 5 versus OTT impairments of non-agency RMBS with NAIC
designations of 1 or 2 within PC insurers [i.e., (β3-5+γ3-5-β1-2) > 0]. Coefficients on the control variables are
suppressed. Equation (1) includes fixed effects for the state of domicile of each firm insurer. Standard errors
are calculated clustering observations by insurer. t and χ2 statistics that are statistically significant at the
1%, 5% and 10% levels are indicated by ***, ** and *, respectively. Definitions of the equation (1)
variables are provided in Appendix A.
(1) (2) (3) (4) (5)
Variable 2007 2008 2009 2010 2011
PC * NAIC_3_5 4.76 7.64** 4.76*** 3.78*** 2.11***
[β3-5] (1.50) (2.28) (5.81) (4.01) (2.64)
PC * NAIC_1_2 0.19** 1.29*** 1.41*** 0.93* 1.15***
[β1-2] (2.29) (2.60) (2.85) (1.77) (2.51)
PC * NAIC_6 29.26 1.91 -0.30 2.75 2.87
[β6] (1.27) (0.22) (-0.11) (0.83) (1.03)
NAIC_3_5 -3.89 -1.94 -1.23** -0.47 0.09
[γ3-5] (-1.24) (-0.79) (-2.39) (-0.82) (0.15)
NAIC_6 -28.05 -2.66 2.18 -0.63 -0.29
[γ6] (-1.22) (-0.37) (1.17) (-0.23) (-0.13)
χ2 Test: (β3-5-β1-2) > 0 4.57* 6.36** 3.35*** 2.85*** 0.95
NAIC 3_5 - NAIC 1_2 for PC-life (1.44) (1.93) (4.51) (3.44) (1.20)
χ2 Test: (β3-5+γ3-5-β1-2) > 0 0.68 4.41** 2.12*** 2.37*** 1.05**
NAIC 3_5 - NAIC 1_2 for PC (1.10) (2.04) (3.82) (3.75) (1.79)
Observations 6,379 4,897 3,801 3,086 2,568
R2 0.092 0.036 0.059 0.055 0.069
51
Table 7: NAIC Designation Downgrades
This table presents statistics on the cumulative downgrades of the sample non-agency RMBS from 2006 to
each sample year from 2007 to 2011. A security is deemed to have been cumulatively downgraded if the
assigned NAIC designation at the end of a given year is higher than that recorded at the end of 2006. Panel
A presents the total number of securities with and without cumulative downgrades. # Downgrades (# No
Downgrades) refers to number of securities with (without) a NAIC designation downgrade. Panel B reports
the percentages of the total number of cumulative downgrades up to a given year that are attributable to
downgrades from each NAIC designation to each lower NAIC designation.
NAIC Category 2007 2008 2009 2010 2011
Panel A: # of securities
# Downgrades 20 176 451 334 308
# No Downgrades 1,543 1,189 682 605 486
Panel B: Percentage of cumulative downgrades
1 to 2 15.00 52.27 43.02 25.75 24.35
1 to 3 15.00 11.36 21.06 21.26 20.45
1 to 4 20.00 13.64 13.30 17.07 19.16
1 to 5 20.00 13.07 9.31 17.66 16.23
1 to 6 5.00 3.41 8.43 13.17 14.29
2 to 3 10.00 0.57 0.22 0.60 0.00
2 to 4 0.00 2.27 0.44 1.20 0.65
2 to 5 10.00 0.00 0.44 0.30 0.32
2 to 6 0.00 1.14 2.88 2.10 3.25
3 to 4 0.00 0.57 0.00 0.00 0.00
3 to 5 0.00 0.00 0.22 0.00 0.32
3 to 6 0.00 0.57 0.22 0.30 0.32
5 to 6 5.00 1.14 0.44 0.60 0.65
52
Table 8: NAIC Category Downgrades
This table examines the relative timeliness of OTT impairments of cumulatively downgraded non-agency
RMBS for PC insurers and life insurers. The table reports annual cross-sectional estimations of the full
equation (1) on the subsample of cumulatively downgraded non-agency RMBS. The table reports a Chi-
Square test that assesses the significance of the difference across PC and life insurers in the incremental
timeliness of OTT impairments of non-agency RMBS with NAIC designations from 3 to 5 versus OTT
impairments of non-agency RMBS with NAIC designations of 1 or 2 [i.e., (β3-5-β1-2) > 0]. This panel also
reports a Chi-Square test that assesses the significance of the incremental timeliness of OTT impairments
of non-agency RMBS with NAIC designations from 3 to 5 versus OTT impairments of non-agency RMBS
with NAIC designations of 1 or 2 within PC insurers [i.e., (β3-5+γ3-5-β1-2) > 0]. Coefficients on the control
variables are suppressed. Equation (1) includes fixed effects for the state of domicile of each firm insurer.
Standard errors are calculated clustering observations by insurer. t and χ2 statistics that are statistically
significant at the 1%, 5% and 10% levels are indicated by ***, ** and *, respectively. Definitions of the
equation (1) variables are provided in Appendix A.
(1) (2) (3) (4) (5)
Variable 2007 2008 2009 2010 2011
PC * NAIC_3_5 7.11 7.50** 4.83*** 4.04*** 2.47***
[β3-5] (1.54) (2.07) (5.24) (3.71) (2.62)
PC * NAIC_1_2 -4.97 3.71* 1.29** 0.51 -0.33
[β1-2] (-1.01) (1.74) (1.98) (0.57) (-0.45)
PC * NAIC_6 41.60 6.06 0.70 4.04 2.38
[β6] (1.37) (0.81) (0.25) (1.23) (0.83)
NAIC_3_5 -12.64 -0.17 -0.99* -0.70 -0.79
[γ3-5] (-1.25) (-0.07) (-1.90) (-1.18) (-1.19)
NAIC_6 -45.46 -2.32 2.25 -0.09 0.06
[γ6] (-1.34) (-0.35) (1.07) (-0.03) (0.02)
χ2 Test: (β3-5-β1-2) > 0 12.08** 3.79 3.55*** 3.53*** 2.80***
NAIC 3_5 - NAIC 1_2 for PC-life (1.70) (1.17) (4.50) (3.84) (3.13)
χ2 Test: (β3-5+γ3-5-β1-2) > 0 -0.57 3.63** 2.56*** 2.83*** 2.01***
NAIC 3_5 - NAIC 1_2 for PC (-0.10) (1.68) (4.31) (3.86) (2.86)
Observations 129 768 2,021 1,480 1,272
R2 0.434 0.109 0.088 0.091 0.090
53
Table 9: Alternative Timeliness Measure
This table examines the relative timeliness of OTT impairments of non-agency RMBS for PC insurers and
life insurers using the alternative OTT impairment timeliness measure TM_Alt. The table reports annual
cross-sectional estimations of the full equation (1), with dependent variable TM_Alt, on the full sample.
The table reports a Chi-Square test that assesses the significance of the difference across PC and life insurers
in the incremental timeliness of OTT impairments of non-agency RMBS with NAIC designations from 3
to 5 versus OTT impairments of non-agency RMBS with NAIC designations of 1 or 2 [i.e., (β3-5-β1-2) > 0].
This panel also reports a Chi-Square test that assesses the significance of the incremental timeliness of OTT
impairments of non-agency RMBS with NAIC designations from 3 to 5 versus OTT impairments of non-
agency RMBS with NAIC designations of 1 or 2 within PC insurers [i.e., (β3-5+γ3-5-β1-2) > 0]. Coefficients
on the control variables are suppressed. Equation (1) includes fixed effects for the state of domicile of each
firm insurer. Standard errors are calculated clustering observations by insurer. t and χ2 statistics that are
statistically significant at the 1%, 5% and 10% levels are indicated by ***, ** and *, respectively.
Definitions of the equation (1) variables are provided in Appendix A.
(1) (2) (3) (4)
Variable 2007 2008 2009 2010
PC * NAIC_3_5 -0.55 19.34*** 11.66*** 13.09***
[β3-5] (-0.78) (3.02) (2.55) (2.62)
PC * NAIC_1_2 1.11* 3.07 -6.01 -5.57
[β1-2] (1.87) (1.32) (-1.43) (-1.45)
PC * NAIC_6 -0.07 -2.77 -7.80 2.86
[β6] (-0.18) (-0.22) (-1.31) (0.55)
NAIC_3_5 1.39 -9.77** -8.33** -9.14***
[γ3-5] (0.99) (-1.95) (-2.14) (-2.59)
NAIC_6 0.50 1.38 3.71 -4.06
[γ6] (1.07) (0.11) (0.79) (-1.08)
χ2 Test: (β3-5-β1-2) > 0 -1.66* 16.26*** 17.68*** 18.66***
NAIC 3_5 - NAIC 1_2 for PC-life (-1.60) (2.59) (3.13) (3.79)
χ2 Test: (β3-5+γ3-5-β1-2) > 0 -0.62 6.34* 9.19*** 9.43***
NAIC 3_5 - NAIC 1_2 for PC (-0.59) (1.46) (2.35) (2.76)
Observations 546 536 540 540
R2 0.085 0.231 0.137 0.111
![binary heap, d-ary heap, binomial heap, amortized analysis ... · Amortized Complexity [amortizovaná složitost] In an amortized analysis , the time required to perform a sequence](https://img.pdfslide.us/doc/110x75/5ed29bc1016d386359233e54/binary-heap-d-ary-heap-binomial-heap-amortized-analysis-amortized-complexity.jpg)
