Determinants of Credit Ratings in UK Insurance Industry

Journal of Business Finance & Accounting, 30(3) & (4), April/May 2003, 0306-686X
The Determinants of Credit

Ratings in the United Kingdom
Insurance Industry
MIKE ADAMS, BRUCE BURTON AND PHILIP HARDWICK*
1. INTRODUCTION
The role of credit rating agencies in overseeing corporate finan-

cial strength and promoting the operation of financial markets
has been a topic of intense interest in the finance literature for
more than twenty years (e.g., Weinstein, 1977; Pinches and
Singleton, 1978; Holthausen and Leftwich, 1986; and Cantor
and Packer, 1995 and 1997).
The majority of prior studies have concentrated on external
credit ratings of corporate bonds, with the early focus in the
literature being on the rating process itself.1 More recent studies
have examined the role and value of the ratings in the market-
place with the general conclusions being: (i) that the determinants
of ratings awarded by leading agencies such as Moody’s and
* The authors are respectively from the European Business Management School,
University of Wales, Swansea; the Department of Accountancy and Business Finance,
University of Dundee; and the School of Finance and Law, Bournemouth University.
They would like to thank Mike Buckle, Zou Hong, Clive Lennox, David Wharrier and
Steve Young for their assistance during the course of this study. The paper also
benefited from the comments of an anonymous referee and editor as well as
participants at a seminar hosted by the Accounting and Finance Group, University of
Bristol. However, the usual disclaimer applies. (Paper received August 2001, revised
and accepted February 2002)
Address for correspondence: Bruce Burton, Department of Accountancy and Business
Finance, University of Dundee, Dundee DD1 4HN, UK.
e-mail: b.m.burton@dundee.ac.uk
# Blackwell Publishing Ltd. 2003, 9600 Garsington Road, Oxford OX4 2DQ, UK
and 350 Main Street, Malden, MA 02148, USA. 539
540 ADAMS, BURTON AND HARDWICK
Standard and Poor’s (S&P) can be modelled accurately using

explanatory variables such as profitability, income gearing,
liquidity, and market capitalisation,2 (ii) that external credit
ratings are good predictors of default risk3 and (iii) that share
and bond values adjust significantly to news of changes in debt
ratings.4 While the overwhelming majority of the early studies
are United States (US)-based, the findings reported in Barron
et al. (1997), who employ a sample of 87 firms rated by S&P
between 1984 and 1992, suggest that the United Kingdom
(UK) market reaction to ratings news is broadly similar to
that documented for the US. The authors also provide novel
evidence that although first time ratings of short- and long-
term debt have no significant impact on equity values, the
assessments appear to provide little benefit in terms of
reducing share price volatility or systematic risk and, therefore,
cost of capital.
Investigation into credit rating practices has recently
extended to insurance markets – markets in which the complex
technical nature of insurance transactions leads to policyholders,
investors and others facing particularly acute information
asymmetries at the point-of-sale (Datta and Doherty, 1990).
External credit ratings are particularly important in insurance
(and other financial service) sectors because of the economic
significance of the industry in virtually all developed economies
as well as the fact that policyholders place considerable reliance
on insurers being sufficiently solvent if and when a claim arises.
Pottier (1997 and 1998) points to a recent rise in the attention
paid by researchers to the insurance sector as a whole and
to credit rating practices in particular; he attributes this to
the increased concern exhibited by both regulators and
consumers about the financial condition of many major
insurers. The current political dimension to the reputation of
the insurance industry is demonstrated by the high profile which
the recent problems experienced by a number of apparently
sound, reputable insurers in the UK such as Equitable Life and
Independent Insurance have attracted in the public media.
As Bouzouita and Young (1998) note, published credit ratings
can perform an important role in this context, and alleviate
imperfections in insurance markets by providing interested
parties with:
# Blackwell Publishing Ltd 2003

CREDIT RATINGS OF UK INSURANCE FIRMS 541
an opinion on an insurer’s financial strength, including its opera-

tional performance, and its ability to meet its obligations to policyholders
(p. 23).
In the light of the potential importance of credit ratings of

insurance firms in the current financial environment, a need
appears to exist for better understanding of the empirical links
between insurer ratings and well-established indicators of corpor-
ate health and prospects, as well as the factors which influence
the likelihood of being rated in the first place.
As Cantor and Packer (1995) point out, credit rating agencies
deal closely with issuers of corporate securities and often use both
quantitative and qualitative information when formulating their
rating of a company’s financial condition. The information used
to establish rankings of financial strength is normally obtained
from a combination of both public sources (e.g., the annual
report and accounts) and private information (e.g., managerial
statements). However, factors such as increased competition
amongst credit rating agencies, the failure of those rating agen-
cies to predict insolvency accurately, and pressure from issuers
of securities have caused some commentators (e.g., Klein, 1992;
Ambrose and Seward, 1988; Ambrose and Carroll, 1994; and
Carson and Scott, 1997) to question the credibility of financial
ratings.5 These concerns suggest that credit rating agencies may
not possess unique and unbiased insights into the managerial
systems and strategies employed in the insurance industry and
again point to the need for an accurate knowledge of how the
letter ratings awarded are related to key financial variables, and
whether these same variables influence the likelihood of insurers
being rated and thereby improve the signals provided to the
market about the firms’ true worth.
In this study, we use UK-based insurance company data relat-
ing to two of the most prominent credit rating firms – A.M. Best
and S&P – to examine: (i) the likelihood that an insurer will
have a financial strength rating and (ii) the determinants of the
letter ratings duly assigned, as well as any differences therein.6
Managers frequently seek credit ratings in order to signal to
external parties (e.g. investors) the adequacy of their corporate
financial strength and other considerations such as the efficiency
of systems of internal control and more general managerial

competence (Pottier, 1997 and 1998; and Pottier and Sommer,

1999). As such, our study provides important insights into
managerial motives for obtaining credit ratings.
By comparing the rating activities of A.M. Best and S&P we
are also able to establish whether insurers’ rating likelihoods
and the determinants of the ratings themselves differ according
to the methods used and/or the range of activities traditionally
undertaken by rating agencies. In particular, two major differ-
ences exist between the rating firms examined in this study.
First, while S&P have historically concentrated on rating indivi-
dual debt issues and only began rating insurance firms relatively
recently, A.M. Best have always specialised in the assessment of
insurance firms,7 and as such, the firm is often the preferred
choice for insurance companies (Pottier and Sommer, 1999).
Second, the ratings assigned to UK insurers by S&P during our
sample period of 1993–1997 are based almost exclusively on
publicly-available financial information, whereas A.M. Best bases
all its ratings on a combination of publicly-available financial
data and detailed inside information provided during formal
company visits (Pottier and Sommer, 1999). The financial tests
performed by A.M. Best measure corporate financial performance
in terms of profitability, liquidity and capital structure, while the
qualitative analysis includes an evaluation of an insurer’s spread of
risk, managerial expertise, adequacy of systems of internal control
and so on (Pottier, 1997). The comparison of A.M. Best with S&P
data thus enables us to distinguish between the influence of public
and non-public information on insurance companies’ ratings.8
Additionally, the academic literature on the determinants of the
financial strength of insurance companies, particularly outside of
the US, is limited (Pottier and Sommer, 1999). Our research
therefore contributes to the extant literature by presenting
evidence about the factors that explain letter ratings assigned to
UK-based direct insurers and provides a basis against which prior
and future research can be compared and evaluated.
The present study also improves upon previous US-based
research in that it uses panel data from both the life and general
insurance industries (including composites), thereby increasing
our sample size and enabling cross-business line parameter
estimates to be derived.9 More generally, by focusing on the
insurance industry, the study provides evidence about credit

ratings that are non-mandatory in nature and a contribution is

made to the theoretical debate about the differences between
rated and non-rated firms, given the mixed evidence about the
effect of mandatory bond ratings on corporate values described
above. In particular the study examines whether similarity exists
between the factors which influence (i) the rating likelihood and
(ii) the ratings duly assigned in the insurance industry, an indus-
try which, as noted earlier, is characterised by relatively high
levels of information asymmetry and an associated need for
accurate signalling.
The remainder of the paper is organised as follows: Section 2
sets out the framework for the study and details the specific
hypotheses to be tested while Section 3 discusses the research
design. Section 4 presents our empirical results before, in Section
5, the paper concludes by discussing the implications of our find-
ings and suggesting some possible avenues for future research.
2. FRAMEWORK
(i) Use of Ratios in the Insurance Industry

The standard financial ratios used to assess the financial condition
of insurance companies (and other financial services companies
such as banks) normally focus on capital adequacy, management
operations, earnings and liquidity (the so-called CAMEL criteria).
These measures are incorporated in the Insurance Regulatory
Information System (IRIS) of the US-based National Association
of Insurance Commissioners (NAIC) to assist in solvency monitor-
ing and control (Carson and Scott, 1997). They have also been
used in previous insurance industry studies as a basis for model-
ling the rating process (e.g., Pottier, 1997 and 1998). In this study,
we therefore investigate the relation between corporate-specific
variables and (i) insurance firms’ likelihoods of being externally
rated by either A.M. Best or S&P, and (ii) the determinants of the
letter ratings themselves. The independent variables based on
the CAMEL criteria and included in the analysis are leverage,
profitability, liquidity, growth, company size, organisational form,
reinsurance and business activity. Our framework, including the
definition of variables, is considered in further detail below.

(ii) Leverage
Borde et al. (1994) state that business risk:
can be influenced by the [insurance] company’s degree of leverage, since
leverage tends to magnify corporate returns or losses (p. 181).
This in turn suggests that managers of highly leveraged insurers

will not voluntarily secure a letter rating since a low rating is likely
to exacerbate market uncertainties about their ability to meet
contractual claims. In addition to measuring insurers’ abilities to
meet their fiscal obligations, the leverage proxy that we employ
reflects an insurer’s ability to write new business without undue
financial strain on its capital base.10 We contend that the higher
the financial leverage, the greater the potential adverse effects of
variations in underwriting performance and economic shocks on
an insurer’s ability to meet its commitments to policyholders and
investors. Some authors (e.g., Jensen, 1986) suggest, however,
that high financial leverage (and the associated risk of default)
can help in disciplining managers to act in the interests of fixed
and residual claimants by preventing wasteful investment of free
cash flow, and may therefore represent a positive attribute for
insurers. Nonetheless, most prior studies (e.g., Brotman, 1989;
and Pottier 1997 and 1998) view high financial leverage as
a negative feature of insurance companies’ capital structure. We
therefore propose the following hypotheses:
H1a: Other things being equal, the lower the leverage the
greater likelihood that insurers will be rated.
H1b: Other things being equal, the lower the leverage the
higher the grade of rating assigned to insurers.
(iii) Profitability
Examination of profitability enables financial analysts and indus-
try regulators to assess an insurer’s ability to invest annual
surpluses efficiently in order to generate new business (Fok
et al., 1997; Bouzouita and Young, 1998; and Pottier, 1998).
In addition to providing an indication of an insurer’s ability to
remain as a going concern, measures of profitability also pro-
vide insights into management’s ability to control expenses

effectively and to set competitive rates of premium. Managers of

profitable insurers are expected to obtain a letter rating in order
to enhance the company’s market profile and secure future new
business. Moreover, Brotman (1989), Bouzouita and Young
(1998) and others contend that, in general, the higher the level
of profitability, the lower the insolvency risk and, consequently,
the better the insurance company’s assigned credit rating. We
therefore hypothesise that:
H2a: Other things being equal, the higher the profitability

the greater likelihood that insurers will be rated.
H2b: Other things being equal, the higher the profitability
the higher the grade of rating assigned to insurers.
(iv) Liquidity
The settlement of policyholders’ claims as they fall due is a
critical task for insurance companies (Brotman, 1989; Ambrose
and Carroll, 1994; and Borde et al., 1994). It is plausible that
managers of highly liquid insurers will be keen to signal their
claims-paying ability to the markets through the rating process
in order to promote their reputations for prudent management.
Thus, liquidity could be an important factor in the corporate
decision to obtain a credit rating. If liquidity is too high, however,
managers are provided with the means to invest in projects with
negative net present values, reducing owners’ wealth while
simultaneously increasing managerial remuneration packages
via the consumption of perquisites and the receipt of company
size-related bonuses. Nonetheless, most prior studies (e.g.,
Carson and Scott, 1997; and Bouzouita and Young, 1998)
report the existence of an inverse relation between insurance
companies’ liquidity risk and their credit ratings. This could
reflect the propensity of highly liquid insurance companies to
signal their sound financial condition to prospective investors
and policyholders. As a consequence, we hypothesise that:
H3a: Other things being equal, the higher the liquidity the
greater the likelihood that insurers will be rated.
H3b: Other things being equal, the higher the liquidity the

(v) Growth
Positive growth in annual surplus could be indicative of a
favourable financial condition. Consequently, we predict that
insurers achieving high year-on-year growth in business are
more likely to volunteer for a credit rating than insurers achiev-
ing low growth. On the other hand, significant new business
growth could be due to low underwriting standards and under-
pricing and, as such, result in greater uncertainty about the
adequacy of the insurer as a going concern (Borde et al. 1994;
and Pottier, 1997). However, prior US-based insurance industry
studies (e.g., Pottier and Sommer, 1999) find that strong busi-
ness growth is positively associated with the rating likelihood
and the grade of letter ratings given to companies because
growth indicates prospects for sound future cash flow
performance and improved economic value. The hypotheses
are therefore:
H4a: Other things being equal, the greater the growth in

annual reported surplus the greater the likelihood that
insurers will be rated.
H4b: Other things being equal, the greater the growth in
annual reported surplus the higher the grade of rating
assigned to insurers.
(vi) Company Size

Bouzouita and Young (1998) suggest that company size is likely
to be positively correlated with assigned credit ratings because,
amongst other things, larger entities have access to a relatively
wide pool of managerial expertise and are likely to realise econo-
mies of scale in their operations. Large insurers are also likely
to have prominent market positions and public reputations
for sound governance which prudent management will be keen
to protect. These attributes, when combined with accumulated
reserves, should enable large insurers to survive the impact of
volatile underwriting climates and unfavourable economic
conditions. Indeed, most prior studies (e.g., Pottier, 1997 and
1998; and Bouzouita and Young, 1998) have found that large

insurers are more likely to be rated and have lower insolvency risk
than small insurers. We therefore predict that:
H5a: Other things being equal, the larger their size the
greater the likelihood that insurers will be rated.
H5b: Other things being equal, the larger their size the
(vii) Organisational Form

Adams (1995), Mayers and Smith (1994) and Pottier and
Sommer (1997) contend that managerial decision-making in
the insurance industry partly depends on whether a firm is
a stock company (i.e., shareholder-owned) or a mutual (i.e.,
policyholder-owned). For instance, risk-taking, investment and
product-mix decisions are expected to vary between stock
companies and mutuals as a direct result of their different own-
ership structures, contracting interests and internal governance.
In much of the insurance literature, mutual insurers are
associated with precautionary managerial activities, behaviour
designed to promote policyholders’ fixed claims in the long-
term, whereas stock insurers are likely to engage in more risky
activities in an attempt to maximise shareholders’ short-term wealth
(Adams, 1995). As a consequence, we predict that managers of
mutual insurers will be more inclined to obtain credit ratings to
promote their reputation for prudent management and protect
their market share than will their counterparts in stock companies.
Pottier (1997) also cites evidence suggesting that in the US life
insurance industry mutuals are more highly rated than stock
companies. Therefore, the hypotheses are:
H6a: Other things being equal, mutual insurers are more

likely to be rated than stock insurers.
H6b: Other things being equal, mutual insurers are likely to
receive a higher rating than stock insurers.
(viii) Reinsurance
By transferring risk to a third party, reinsurance can help to
reduce uncertainty regarding the frequency and magnitude of

future losses and enable a primary insurer to sustain an external

economic shock (Adams, 1996). On the other hand, the
insolvency of a reinsurance company could adversely affect the
financial strength of the primary insurer, particularly in cases
where the primary insurer was heavily reinsured. Clearly, the
latter situation would complicate the assessment of a primary
insurer’s future financial condition and its ability to fulfil its
obligations to policyholders, investors and others. In their
recent US-based study, Pottier and Sommer (1999) did indeed
find a negative relation between the amount of reinsurance
undertaken and the letter rating assigned by leading agencies,
suggesting that the insurance and investment markets view
substantive reliance on reinsurance unfavourably. This reasoning
suggests that, all else being equal, highly reinsured companies
are less likely to volunteer for a credit rating than entities with
less reinsurance. Additionally, for the insurance market as a
whole, we expect that lowly reinsured insurance companies will
be given higher letter ratings than highly reinsured insurance
companies (and vice versa). We therefore hypothesise that:
H7a: Other things being equal, the lower the amount of

reinsurance held the greater the likelihood that insur-
ers will be rated.
H7b: Other things being equal, the lower the amount of
reinsurance held the higher the grade of rating
assigned to insurers.
(ix) Business Activity

Insurance companies may specialise in either short-term (e.g.,
vehicle insurance) or long-term business (e.g., life insurance).
Alternatively, they may engage in a combination of both short
and long-term operating lines and hence operate as composite
insurers. Traditionally, long-term (e.g., life) insurance liabilities
have been fairly predictable because of the application of actuar-
ial principles in policy valuation. Moreover, the maintenance of
statutory levels of solvency has helped to allay public concerns
regarding life insurers’ abilities to meet policyholders’ future
claims (Borde et al., 1994). We contend that, other things being
equal, this situation obviates the need for life insurers to volun-

tarily seek a letter rating. In contrast, insurers writing short-term

general (property-liability) business often have to adjust to
unanticipated losses by varying annual premiums, reducing
operating costs and maintaining volumes (e.g., through the
use of new technology), and changing underwriting terms
at short notice. Composites’ motives for seeking a letter rating
may also differ from those of pure life or general insurers (e.g.
strategic investment overseas), but the literature provides no
clear prediction as to whether (or how) the grade of letter rating
is expected to vary between the three types of insurance com-
panies. Accordingly, we propose the single hypothesis that:
H8: Other things being equal, general and composite

insurers are more likely to be rated than pure life
insurance companies.
3. RESEARCH DESIGN
(i) Models
The aims of the research are, first, to investigate the main
determinants of UK insurers’ rating likelihoods and, secondly,
to investigate the main determinants of the actual letter ratings
assigned to UK insurers. To achieve the first aim, we estimate a
‘rating likelihood’ multinomial logit model to determine the
main influences on an insurer’s likelihood of having either an
A.M. Best rating, an S&P rating or no rating.11 In this study, the
dummy variable ( j) is set equal to 0 for non-rated insurers, 1 for
A.M. Best-rated insurers and 2 for S&P-rated insurers. The
model to be estimated may be written in general form as:
0
ej x
Prob ½choice j ¼ P 0 j ¼ 0; 1; 2 ð1Þ
je x
j
where 0j is a vector of the parameters to be estimated for each j

and x is a vector of independent variables. We also estimate two
‘rating likelihood’ probit models: one to investigate the likelihood of
being A.M. Best-rated or non-rated and the second to investigate the
likelihood of being S&P-rated or non-rated. In both probit models,

the dummy dependent variable is set equal to 0 for non-rated

insurance companies and equal to 1 for rated insurance companies.
The second aim is less straightforward to achieve. Insurance
company letter ratings are ordinal measurements and, as such,
their analysis necessitates the estimation of an ordered probit
model, as developed by McKelvey and Zavoina (1975) and
used in Pottier and Sommer (1999). However, since A.M. Best
and S&P do not rate the full population of UK insurers, but
rather only those companies applying voluntarily for a letter
rating, our data set is potentially subject to self-selection bias
(Greene, 1999). This is likely to lead to a skewed distribution
in the sample in favour of better letter ratings and therefore
inconsistent parameter estimates. We control for this possibility
by extending the standard sample selection model of Heckman
(1979) to the ordered probit model in the manner prescribed by
Durbin and Rivers (1990) and Greene (1999). In this modified
model, the ordinal random variable is assessed for self-selection
bias by testing the correlation () between the error terms in the
‘rating likelihood’ probit model and the ‘rating determination’
ordered probit model. The null hypothesis that selection bias
does not exist is then ¼ 0. Next, we re-estimate the coefficients
of both models (including ) using a full-information maximum
likelihood technique, which ensures that the correlation
between the error terms in the two models is taken into account,
so producing consistent estimates.12
The ‘rating likelihood’ probit models and the ‘rating deter-
mination’ ordered probit model (with sample selection) are
written in general form as:
dit ¼ 0 xit þ uit

dit ¼ 0 if dit 0
¼ 1 if dit > 0 ð2Þ
yit ¼ 0 zit þ eit

yit ¼ 0 if yit 0
¼ 1 if 0 < yit 1
¼ 2 if 1 < yit 2
¼ 3 if yit > 2 ð3Þ

uit ; eit
Nð0; 0; 1; 1; Þ
½ yit ; xit is observed if and only if dit ¼ 1:

In these models, the latent variables, d* and y*, are continuous,
unobserved variables: in equation (2), d* may be interpreted as
the propensity to be rated, while in equation (3), y* represents the
propensity to be assigned a particular rating. The binary variable d
equals 0 for an insurer that has no rating and 1 for an insurer with
a rating. The ordinal variable, y, is coded on a four-point scale
from 0 to 3, where 0 represents the poorest condition and
3 superior financial strength.13 In equations (2) and (3), x and z
represent vectors of independent variables and 0 and 0 are the
vectors of coefficients to be estimated. The two error terms, u and
e, are assumed to be normally distributed with means equal to
zero, variances equal to one and a correlation coefficient equal to .
(ii) Data and Variables

Consolidated annual data were obtained for the 40 or so UK-
based insurers that were rated exclusively by A.M. Best and 25
UK-based insurers that were rated only by S&P between the
years 1993 and 1997 inclusive.14,15 This sample represents
approximately 10 per cent of the total population of insurers
operating in the UK over the period of analysis. Additionally,
we collected a sample of 28 UK insurers that were not rated by
A.M. Best, S&P, or any other credit rating agency over the five-
year period studied and for which complete data were available.16
The period 1993–97 represents the earliest and latest years for
which complete information on the A.M. Best-rated and S&P-
rated insurance companies in our sample were available at the
time our study was carried out. The primary source of data used
was the A.M. Best Insurance Companies Directory (1999), supple-
mented by data obtained from A.M. Best and S&P International
Insurer Ratings Lists and insurance company annual reports.
The independent variables included in the study are defined
as follows. Leverage (LEV): As in Bouzouita and Young (1998),
we use financial leverage as a proxy for capital adequacy, which
is the ratio of accumulated reserves to total assets. Profitability
(PROFIT): Again as in Bouzouita and Young (1998), we define

profitability as the ratio of annual investment and underwriting

income (net of expenses), plus unrealised capital gains, to
statutory capital. Liquidity (LIQ): This variable is measured as
the ratio of current assets to current liabilities and therefore
represents insurers’ ability to fulfil their immediate obligations
to policyholders in the event of claims (Pottier, 1998). Business
Growth (GROWTH): Previous studies (e.g., Bouzouita and Young,
1998; and Pottier and Sommer, 1999) have used the absolute
change in annual reported surplus to measure the prospective
surety of policyholders’ fixed claims in periods of adverse
economic conditions. We follow the same procedure in this
study. Company Size (LNSIZE): In most studies of the insurance
industry, company size is measured as the natural logarithm of
total admitted assets because of the highly skewed distribution of
total assets among firms operating in the insurance industry
(Pottier and Sommer, 1997). Accordingly, this surrogate of
insurance company size is also used here. Additionally, this
variable is deflated by the relevant average annual retail price
index (base year ¼ 1993) in order to counteract the effects of
inflation. Organisational Form (OFORM): Like previous studies
(e.g., Adams, 1996), a dummy variable that takes the value 0 for
a stock company and 1 for a mutual is used as a proxy for
organisational form.17 Reinsurance (REINS): As in Adams
(1996), we define our reinsurance variable as the ratio of annual
reinsurance ceded over annual premiums written. Line of
Business Activity (BUS): This variable takes the categorical value
0 for life insurers and 1 for general and composite insurers.
Apart from BUS (which is assumed to influence d* but
not y*),18 the independent variables are predicted to have the
same directional influence on both the rating likelihoods and
the ratings themselves, for the reasons set out in Section 2. The
equations to be estimated may be written in full as:
dit ¼ 0 þ 1 LEVit þ 2 PROFITit þ 3 LIQit þ 4 GROWTHit

þ 5 LNSIZEit þ 6 OFORMit þ 7 REINSit þ 8 BUSit þ uit ; ð4Þ
yit ¼ 0 þ 1 LEVit þ 2 PROFITit þ 3 LIQit þ 4 GROWTHit

þ 5 LNSIZEit þ 6 OFORMit þ 7 REINSit þ eit : ð5Þ

4. RESULTS
(i) Descriptive Data

Panel A of Table 1 presents the means and standard deviations
of the independent variables used in the study. Separate figures
are given for the A.M. Best-rated, S&P-rated and non-rated
insurance companies in the sample. The most striking finding
to emerge from an inspection of these figures is the large varia-
tion in surplus growth rates across the sample, particularly for
the A.M. Best-rated insurance companies where a coefficient of
variation of 5.49 resulted. However, when the data for each A.M.
Best-rated company were examined, it emerged that this figure
was largely driven by an observation of 992 per cent in 1997 for
the newly merged conglomerate of Royal Sun Alliance. Remov-
ing this outlier from the sample causes the coefficient of variation
to fall to 2.3. A further interesting finding to emerge from the
figures is that the non-rated companies tend on average to be
larger and have lower profitability and growth rates and more
reinsurance than the rated companies in the sample.
Panel B of Table 1 details the number of insurance firms
obtaining each of our coded values of the A.M. Best and S&P
rating categories in all five years of the sample period. The
results show that in all years, over 90 per cent of the A.M. Best-
rated insurers were given an A or higher rating and therefore
categorised as ‘excellent’ or ‘superior’. This proportion is higher
than that found for US property-liability insurers by Pottier and
Sommer (1999), who report that just over 72 per cent of all A.M.
Best-rated insurers in their study were ‘excellent’ or ‘superior’ in
1996. The S&P-rated insurers in our sample generally have
substantially lower letter ratings than those assessed by A.M.
Best. Only 64 per cent of the S&P-rated insurers were assigned
an A or higher rating (roughly equivalent to a Bþ A.M. Best
rating) in 1993–94, rising to 88 per cent in 1996–97. These
proportions are lower than those reported for US property-
liability insurers by Pottier and Sommer (1999) who found that
over 93 per cent of the S&P-rated insurers in their sample were
assigned ratings of A or higher. The figures in Panel B of Table
1 also reflect the fact that the S&P ratings for UK insurers
changed more often than those rated by A.M. Best over the

Table 1
Descriptive Statistics for the Sample of UK Insurance Companies
Panel A: Key Summary Statistics for the Sample

A.M. Best S&P Non-rated Companies
Variables Mean SD Mean SD Mean SD
LEV 0.73 0.12 0.88 0.06 0.90 0.09

PROFIT 0.33 0.11 0.08 0.05 0.07 0.08
LIQ 2.32 0.84 1.43 0.29 1.77 0.90
GROWTH 14.15 77.62 22.12 69.49 0.88 2.82
LNSIZE 6.40 1.66 6.51 1.76 7.60 1.55
OFORM 0.13 0.34 0.00 0.00 0.29 0.46
REINS 0.27 0.23 0.23 0.14 0.31 0.39
BUS 0.83 0.38 0.92 0.27 0.61 0.49
Panel B: A.M. Best and S&P Ratings, 1993–97
No. of Firms
Coded Rating Grade Categorya 1993 1994 1995 1996 1997 Total
A.M. Best
3 Aþ/Aþþ Superior 6 6 6 6 5 29
2 A Excellent 25 25 25 26 24 125
1 A Excellent 6 6 6 6 6 30
0 Bþþ or less Very good 2 2 2 2 3 11
or below 39 39 39 40 38 195
S&P
3 Aþ or above Very good
or better 1 3 3 3 4 14
2 A Good 8 6 6 6 5 31
1 A Good 7 7 8 13 13 48
0 Bþþ or less Adequate 9 9 8 3 3 32
or below 25 25 25 25 25 125
Panel C: Correlation Coefficient Matrix and Variance-Inflation Factors
(VIFs)
Correlations
VIFs PROFIT LIQ GROWTH LNSIZE OFORM REINS BUS
LEV 1.58 0.52** 0.16 0.01 0.34** 0.08 0.05 0.13

PROFIT 1.59 1.00 0.25** 0.03 0.26** 0.20** 0.11 0.23**
LIQ 1.10 1.00 0.01 0.08 0.10 0.09 0.04
GROWTH 1.28 1.00 0.37** 0.05 0.08 0.06
LNSIZE 1.38 1.00 0.38** 0.27** 0.20**
OFORM 1.35 1.00 0.30** 0.28**
REINS 1.24 1.00 0.25**
BUS 1.25 1.00

Table 1 (Continued)
Notes:
This table presents summary statistics for the sample of financial and insurance firm
data used in the study. The sample consists of an unbalanced panel of 195 company/year
observations for UK-based insurance companies rated by A.M. Best, 125 company/year
observations for UK-based insurance companies rated by S&P, and a further 140
company/year observations for UK non-rated insurance companies, all for the period
1993–97. Panel A provides descriptive data for the rated and non-rated insurers, while
Panel B details the numbers of firms in each of A.M. Best’s and S&P’s rating categories
for the years 1993–97. Panel C shows the correlation matrix and variance-inflation
factors (VIF) for the independent variables used in the study, calculated from the total
sample of 460 observations. LEV is the ratio of accumulated reserves to total assets.
PROFIT is the ratio of annual income plus unrealised capital gains to statutory capital.
LIQ is the ratio of current assets to current liabilities. GROWTH is the absolute change
in reported annual surplus. LNSIZE is the natural logarithm of total admitted assets.
OFORM is a dummy variable that takes the value of 0 for a stock (shareholder-owned)
company and 1 for a mutual (policyholder-owned) company. Over the sample period,
no mutuals approached S&P for a credit rating. REINS is the ratio of annual reinsur-
ance ceded over annual premiums written. BUS is a dummy variable that takes the
value of 0 for life insurers and 1 for general and composite insurers. SD refers to
standard deviation. Correlations involving non-metric variables are measured using
the Spearman rank coefficient; the remainder are computed using the Pearson
product-moment coefficient.
**Denotes significance at the 0.05 level (two tail).
sample period. This evidence may reflect a change in the relative

weighting criteria used by S&P in their assignment of letter
ratings (e.g. Pottier, 1997, p. 112).
Finally, Panel C of the table provides a matrix of Pearson and
(where non-metric variables are involved) Spearman correlation
coefficients for all the independent variables included in the
pooled model. The correlation coefficients between pairs of
independent variables are generally low (all are less than
0.53), which suggests that multicollinearity is unlikely to be a
problem. Nevertheless, since collinearity can exist among more
than two independent variables, we computed variance-
inflation factors (VIFs) by regressing each independent variable
in turn on all the others and then calculating 1(1 – R2). As
the calculated VIFs are all less than 1.60, we can be reasonably
confident that multicollinearity is not a problem in this study.19
(ii) The Rating Likelihood Multinomial Logit Model

The ‘rating likelihood’ multinomial (i.e., trichotomous) logit esti-
mation results are shown in Table 2. Interpreting the parameter

estimates in this model is not straightforward, so we concentrate

here on the statistical significance of the estimates, shown in
Panel A, and the marginal effects, shown in Panel B. The mar-
ginal effects represent the partial derivatives of the probability
of being rated by each agency or non-rated (see equation (1))
with respect to each independent variable, evaluated at the
means of the independent variables.
Table 2
The Rating Likelihood Multinomial Logit Model: Estimation Results
Panel A: Parameter Estimates

A.M. Best S&P
Variable Parameter Estimate Standard Error Estimate Standard Error
CONSTANT 0 12.08 3.51** 10.17 2.48**

LEV 1 17.06 3.61** 4.72 2.19**
PROFIT 2 36.50 4.81** 8.43 2.92**
LIQ 3 0.98 0.44** 1.15 0.36**
GROWTH 4 0.05 0.02** 0.05 0.02**
LNSIZE 5 0.99 0.22** 0.85 0.16**
REINS 7 1.44 1.20 3.45 0.73**
BUS 8 2.69 0.88** 2.06 0.55**
Panel B: Marginal Effectsa
Non-rated S&P A.M. Best
( j ¼ 0) ( j ¼ 2) ( j ¼ 1)
LEV 0.990 1.930 2.920

PROFIT 2.090 4.440 6.530
LIQ 0.030 0.370 0.400
GROWTH 0.003 0.001 0.002
LNSIZE 0.070 0.010 0.080
REINS 0.140 0.390 0.250
BUS 0.180 0.070 0.250
2 656.2 (Reject hypothesis that all coefficients are zero)b
Wald test statisticc 58.8
McFadden’s R2 0.66
Panel C: Predicted and Actual Values of the Dependent Variable
Predicted Totals
Actual 0 1 2
0 108 7 25 140
1 9 182 4 195
2 16 1 108 125
Totals 133 190 137 460

Table 2 (Continued)
Notes:
This table shows the estimation results for the rating likelihood multinomial logit model,
using data from our complete sample, which consists of an average of 92 UK insurance
companies over five years (making 460 company/year observations in total). Panel A of
the table lists the parameter estimates and standard errors for the S&P and Best rated
companies, while Panel B shows the marginal effects for the non-rated, S&P-rated and
Best-rated companies respectively, together with a set of diagnostic statistics. Panel C
shows a cross-tabulation of predicted and actual values of the rating likelihood dummy
variable. LEV is the ratio of accumulated reserves to total assets. PROFIT is the ratio of
annual income plus unrealised capital gains to statutory capital. LIQ is the ratio of
current assets to current liabilities. GROWTH is the absolute change in reported annual
surplus. LNSIZE is the natural logarithm of total admitted assets. REINS is the ratio of
annual reinsurance ceded over annual premiums written. BUS is a dummy variable that
takes the value of 0 for life insurers and 1 for general and composite insurers. In the
S&P sample, all insurers are stock companies, so the coefficients of OFORM could not
be estimated.
* Significantly different from zero at the 0.10 level.
** Significantly different from zero at the 0.05 level.
a
Computed at the means of the independent variables.
b
The critical value of 2 for this test is 14.07.
c
The critical value of 2 for the Wald Test is 14.07.
Leverage, profitability, liquidity, growth, company size and busi-

ness activity all have statistically significant influences (at the five
percent level, two tail) on the probability of being A.M. Best-rated.
Similarly, leverage, profitability, liquidity, growth, company size,
reinsurance and business activity all have significant influences on
the probability of being S&P-rated. According to the marginal
effects relating to the non-rated insurers (shown in the first column
of Panel B), increases in leverage, company size or reinsurance
would increase the probability of being non-rated, while increases
in profitability, liquidity, growth or business type would all reduce
the probability of being non-rated. The model therefore suggests
that the likelihood of being rated (by either agency) is positively
related to profitability, liquidity, growth and business type and
negatively related to leverage, company size and reinsurance.
An examination of all three columns of marginal effects in
Panel B of Table 2 suggests further that: first, an increase in
leverage or company size reduces the probability of being A.M.
Best-rated, but increases the probability of being S&P-rated or
non-rated. Second, that a rise in profitability, liquidity or business
type increases the probability of being A.M. Best-rated, but
reduces the probability of being S&P-rated or non-rated. Third,
that an increase in reinsurance reduces the probability of being

S&P-rated while raising the probability of being A.M. Best-rated or

non-rated and fourth, that an increase in growth leads to a small
rise in the probability of being rated by either agency and a small
decrease in the probability of being non-rated. These findings
suggest that managers of UK insurers may consciously evaluate
the benefits of an A.M. Best rating (e.g., increased market con-
fidence) against the potential opportunity costs of subjecting their
operations to external scrutiny (e.g., disruption costs). That is,
managers of insurance companies have incentives to use A.M.
Best only if the internal financial management systems, annual
financial performance and solvency conditions are good.
The other statistics shown in Table 2 include a 2 statistic
for testing the null hypothesis that the regression coefficients
(excluding the constant term) are all zero. The calculated
value of 656.2 enables us to reject this hypothesis. The McFad-
den R2 value of 0.66 suggests a reasonably good fit. The Wald
test statistic enables us to test for a significant difference between
the A.M. Best and S&P coefficient estimates. The calculated
value of 58.8 provides evidence that the trichotomous model
that differentiates between A.M. Best-rated, S&P-rated and non-
rated insurers is superior to a simpler binomial model that
differentiates only between rated and non-rated insurers.
Panel C of Table 2 shows a cross-tabulation of the actual and
predicted values of j. The estimated model correctly predicts an
A.M. Best rating in 182 out of 195 within-sample cases, an S&P
rating in 108 out of 125 within-sample cases and no rating in
108 out of 140 within-sample cases.
(iii) The Rating Likelihood Probit Models

The ‘rating likelihood’ probit estimation results20 are summarised
in Panel A of Table 3. Two binomial models are estimated: the
first examines the likelihood of being A.M. Best-rated or non-
rated, while the second examines the likelihood of being S&P-
rated or non-rated. This procedure is similar to that followed
in Pottier (1997 and 1998) and may help determine whether
non-rated insurers have firm-specific characteristics that
explain why they are not rated.
In the first model, leverage, profitability, liquidity and business
activity all have expected signs and are statistically significant

Table 3
The Rating Likelihood Probit Models: Estimation Results

A.M. Best S&P
Expected
Variable Parameter Sign Estimates Estimates
CONSTANT 0 2.56 (1.68)* 5.73 (1.48)**

LEV 1 6.92 (1.76)** 0.96 (1.34)
PROFIT 2 þ 14.47 (2.14)** 2.98 (1.61)*
LIQ 3 þ 0.52 (0.21)** 0.66 (0.21)**
GROWTH 4 þ – 0.12 (0.05)**
LNSIZE 5 þ 0.13 (0.11) 0.67 (0.10)**
OFORM 6 þ 0.37 (0.50) –
REINS 7 0.80 (0.55) 1.91 (0.42)**
BUS 8 þ 1.45 (0.48)** 0.87 (0.27)**
Chi-squared test ( 2): 367.5 151.0

(Reject hypothesis that all coefficients are zero)a
Kullback-Leibler R2: 0.80 0.41

Panel B: Predicted and Actual Values of the Dependent Variable
A.M. Best S&P
Predicted Totals Predicted Totals

Actual 0 1 Actual 0 1
0 133 7 140 0 113 27 140
1 11 184 195 1 25 100 125
Totals 144 191 335 Totals 138 127 265
Notes:
This table shows the estimation results for the rating likelihood probit models. To obtain the
A.M. Best estimates, data were used from a sample of an average of 39 A.M. Best-rated and
28 non-rated UK-based insurance companies over five years (making 335 company/year
observations in total). To obtain the S&P estimates, data were used from a sample of 25 S&P-
rated and the 28 non-rated UK-based insurance companies over five years (making 265
observations in total). Panel A lists the parameter estimates, with standard errors in parenth-
eses, while Panel B provides diagnostic statistics and cross-tabulations of predicted and actual
values of the rating likelihood dummy variables. LEV is the ratio of accumulated reserves to
total assets. PROFIT is the ratio of annual income plus unrealised capital gains to statutory
capital. LIQ is the ratio of current assets to current liabilities. GROWTH is the absolute
change in reported annual surplus. LNSIZE is the natural logarithm of total admitted assets.
OFORM is a dummy variable that takes the value of 0 for a stock (shareholder-owned)
company and 1 for a mutual (policyholder-owned) company. REINS is the ratio of annual
reinsurance ceded over annual premiums written. BUS is a dummy variable that takes the
value of 0 for life insurers and 1 for general and composite insurers. In the A.M. Best model,
GROWTH proved to be insignificant and so has been omitted. In the S&P sample, all
insurers are stock companies, so the coefficient of OFORM could not be estimated.
* Significantly different from zero at the 0.10 level (one-tailed tests).
** Significantly different from zero at the 0.05 level (one-tailed tests).
a
The critical value of 2 for this test is 14.07.

influences (at the five per cent level, one tail) on the propensity to
be rated by A.M. Best, while company size, reinsurance and
organisational form are not statistically significant. The estimated
coefficient of the measure of financial leverage (negative, as
expected) suggests that insurance companies with lower leverage
have a greater propensity to be A.M. Best-rated than non-rated.
The estimated coefficients of profitability and liquidity have
expected positive signs, suggesting that insurance companies
with higher levels of profitability and liquidity (i.e., ‘sound’
financial performance) have a greater propensity to be A.M.
Best-rated than non-rated. The estimated coefficient of business
type is also positive, as expected, which suggests that general
and composite insurers have a greater propensity to be A.M.
Best-rated than non-rated. These results are consistent with the
multinomial logit results described in the previous section.
In the second model, surplus growth, reinsurance and business
activity all have the expected signs and are significant at the five
per cent level (one tail) as influences on the propensity to be S&P-
rated. Profitability (which also has its expected sign) is only
significant at the ten per cent level, while leverage is statistically
insignificant. Company size and liquidity have unexpectedly
negative signs, but are significant at the five per cent level. These
results suggest that smaller insurance companies with lower
levels of liquidity and reinsurance will have a greater propensity
to be S&P-rated than non-rated. The estimated coefficients
of profitability, growth and business type are all positive, as
expected, suggesting that, as with A.M. Best-rated insurers, fast-
growing general and composite insurers with sound profitability
have a greater propensity to be S&P-rated than non-rated. One
notable difference in the findings relates to liquidity. As in the
multinomial logit model, the evidence suggests that improvements
in liquidity lead to insurers being more likely to have a rating from
A.M. Best, but less likely to have a rating from S&P. Improved
liquidity could thus encourage managers to expose their financial
systems to the additional scrutiny of an A.M. Best visitation.
The other statistics listed in Table 3 include the 2 statistic for
testing the null hypothesis that the regression coefficients
(excluding the constant term) are all zero. The 2 values of
367.5 and 151 respectively allow us to reject this hypoth-
esis. The Kullback-Leibler R2 values of 0.80 and 0.41 suggest

reasonably good fits.21 Finally, Panel B shows the cross-

tabulations of the actual and predicted values of the rating
dummy variable. The estimated models correctly predict an
A.M. Best rating in 317 out of 335 within-sample cases and an
S&P rating in 213 out of 265 within-sample cases.
(iii) The Rating Determination Ordered Probit Model

The results obtained from estimating the pooled ‘rating deter-
mination’ ordered probit model with sample selection are shown
separately for the A.M. Best-rated and S&P-rated samples in
Panel A of Table 4. In the A.M. Best regression, just three out
of the six independent variables included in the model have
statistically significant effects on the assigned rating, namely prof-
itability, liquidity and organisational form, all of which have the
expected positive signs. The results thus enable us to conclude
that mutual insurance companies with higher-than-average levels
of profitability and liquidity tend to have higher A.M. Best ratings.
Of the other independent variables, company size and rein-
surance both have the expected signs, but are not significantly
different from zero at the five per cent level. Leverage has a
negative sign, as expected, but is insignificant. Therefore, leverage
may be given less weight in the overall rating process by A.M. Best
analysts than might be expected. The estimate of the error corre-
lation coefficient, , is 0.86 with a standard error of 0.24. This
suggests that selection bias is present in our sample of A.M. Best-
rated companies and justifies the application of the ordered probit
model with selection as a way of producing consistent estimates.
In the S&P regression, three of the five independent variables
included in the model have the expected (statistically significant)
influences on the assigned rating, these being leverage, profitability
and liquidity. Company size and reinsurance have unexpected
negative and positive signs respectively, but are both insignificant.
The estimate of the error correlation coefficient, , is only 0.34
with a standard error of 0.35, suggesting that selection bias is not
present in the sample of S&P-rated insurance companies.
In interpreting the ordered probit results, it is important to
emphasise that the coefficients are not the same as the marginal
effects, and the latter are therefore shown separately in Panel B
of Table 4.22 In both the A.M. Best and S&P models, we find that

Table 4
The Rating Determination Ordered Probit Models with Sample
Selection: Estimation Results

A.M. Best S&P
Expected Standard Standard

Variable Parameter Sign Estimate Error Estimate Error
CONSTANT 0 0.40 1.04 9.44 3.75**

LEV 1 1.34 1.14 12.55 3.78**
PROFIT 2 þ 2.80 0.95** 4.42 2.60**
LIQ 3 þ 0.19 0.10** 2.35 0.42**
LNSIZE 5 þ 0.11 0.11 0.14 0.11
OFORM 6 þ 0.70 0.32**
REINS 7 0.35 0.56 0.47 1.06
Division 1 0.78 0.17** 1.68 0.28**
2 2.72 0.22** 3.11 0.34**
Error correlation 0.86 0.24** 0.34 0.35
Panel B: Marginal Effects
A.M. Best Rating S&P Rating
0 1 2 3 0 1 2 3
LEV 0.17 0.24 0.15 0.26 2.18 2.45 3.67 0.96

PROFIT 0.35 0.49 0.30 0.54 0.77 0.86 1.29 0.34
LIQ 0.02 0.03 0.02 0.04 0.41 0.46 0.69 0.18
LNSIZE 0.01 0.02 0.01 0.02 0.02 0.03 0.04 0.01
OFORMa 0.09 0.12 0.08 0.13
REINS 0.04 0.06 0.04 0.07 0.08 0.09 0.14 0.04
Notes:
This table shows the estimation results for the rating determination ordered probit model
with sample selection. To obtain the A.M. Best estimates, data were used from a sample of an
average of 39 A.M. Best-rated and 28 non-rated UK-based insurance companies over five
years (making 335 company/year observations in total). To obtain the S&P estimates, data
were used from a sample of 25 S&P-rated and the 28 non-rated UK-based insurance
companies over five years (making 265 observations in total). Panel A provides the parameter
estimates and standard errors, while Panel B summarises the marginal effects for each
independent variable. LEV is the ratio of accumulated reserves to total assets. PROFIT is
the ratio of annual income plus unrealised capital gains to statutory capital. LIQ is the ratio of
current assets to current liabilities. GROWTH is the absolute change in reported annual
surplus. LNSIZE is the natural logarithm of total admitted assets. OFORM is a dummy
variable that takes the value of 0 for a stock (shareholder-owned) company and 1 for a mutual
(policyholder-owned) company. REINS is the ratio of annual reinsurance ceded over annual
premiums written. BUS is a dummy variable that takes the value of 0 for life insurers, 1 for
general insurers and 2 for composites. GROWTH proved to be insignificant in both models
and so has been omitted from the estimation. The coefficient of OFORM cannot be estimated
in the S&P model because all of the S&P-rated companies in the sample are stock companies.
** Denotes significance at the 0.05 level (one-tailed tests).
a
The marginal effect for the dummy variable OFORM is calculated as the change in
probability that results when the dummy variable changes by one unit.

an increase in profitability or liquidity will, as expected, reduce

the probability of being assigned a 0 or 1 coded (‘low’) rating and
increase the probability of being assigned a (‘high’) rating of 2 or
3. However, Panel B also shows that, in both models, a higher
value of leverage increases the probability of being assigned a 0
or 1 rating and reduces the probability of a 2 or 3. The two
models differ with respect to the effects of a change in company
size and reinsurance. For the S&P-rated insurance companies, an
increase in company size or a decrease in reinsurance will
increase the probability of being assigned a 0 or 1 rating and
reduce the probability of a 2 or 3; while exactly the opposite
applies to the A.M. Best-rated insurance companies. These obser-
vations indicate that A.M. Best-rated insurers and S&P-rated
insurers could be distinguished not only by their size, but also
in terms of their reliance on reinsurance. One interpretation of
this evidence is that the relative weighting of company size and
reinsurance in the rating process could be different between
A.M. Best and S&P analysts. This contention is also supported
by Pottier and Sommer (1999) when they state that their results:
are consistent with the hypothesis that rating agencies differ systematically in
the relative importance given to the different factors they consider (p. 639).
Finally, a change in organisational form from stock to mutual

insurer will reduce the probability of being assigned an A.M. Best-
rating of 0 or 1 and increase the probability of a 2 or 3, suggesting
that UK mutual insurers may be more prudently managed than
stock companies. This could indicate that, over the period of
analysis, mutual insurers were more capital constrained relative
to stock insurers and that managerial prudence may have been a
partial response to avoid the demutualisation option.23
5. CONCLUSIONS
This is the first study to investigate empirically the rating likelihood

by UK insurance companies, and the first to focus specifically on the
grade of ratings assigned to life, property-liability and composite
insurance companies by A.M. Best and S&P using ordered probit
models that control for self-selection bias. The main implications of
our findings are summarised below in three sub-sections. The first
discusses the results of the investigation into the determinants of
UK insurers’ rating likelihoods. The second discusses the results

of the investigation into the determinants of the ratings actually
assigned by A.M. Best and S&P. The third presents some general
conclusions and suggestions for further research.
(i) The Rating Likelihood

Our first main finding is that the likelihood of being rated is posi-
tively related to an insurer’s profitability and negatively related to
leverage. We also find that general and composite insurers have a
greater propensity to be rated than pure life insurers and that
smaller insurers have a greater propensity to be rated. These results
lend some support to the view expressed by Pottier and Sommer
(1999) that insurers frequently seek to obtain ratings in order to
resolve ex ante uncertainty about their future levels of financial risk
and solvency. However, there are some differences in the determi-
nants of the likelihood of being rated by A.M. Best or by S&P. For
example, the multinomial and binomial models both support the
conclusion that insurance companies with higher levels of liquidity
are more likely to have an A.M. Best rating and less likely to have
an S&P rating. Similarly, the multinomial results suggest that
a decrease in financial leverage will increase the likelihood of
being A.M. Best-rated, but reduce the probability of being S&P-
rated. This suggests that managers may value the perceived benefit
of an A.M. Best investigation of capital management procedures
(e.g., enhanced market signalling) to be greater than the possible
additional cost (e.g., costs of disruption). The extent to which man-
agers consider the perceived quality of the rating process provided
by A.M. Best and S&P and the degree to which the ex ante choice of
rating agency helps to fulfil the self-interest objectives of manage-
ment are interesting issues that can be pursued in future research.
(ii) The Rating Determinants

Our second major finding concerns the ratings assigned by
A.M. Best and S&P. We find that the ratings assigned by A.M.
Best are related to profitability, liquidity and organisational
form. In particular, we have strong evidence to support our
hypotheses that higher levels of profitability and liquidity lead
to higher A.M. Best ratings and that mutual insurers tend to

have higher A.M. Best ratings than stock insurers. This latter
finding is consistent with the view expressed by Adams (1995)
that mutual insurers tend to engage in more precautionary
managerial behaviour and less risk-taking than do stock
insurers, who face market pressures to maximise shareholder
wealth. Turning to the ratings assigned by S&P, we find that the
three statistically significant determinants are profitability,
liquidity and financial leverage. As with the A.M. Best ratings
and in line with our prior expectations, higher levels of profit-
ability and liquidity lead to higher S&P ratings. Also in line with
our expectations, we find a significantly negative relation
between LEV and the S&P rating, suggesting that insurers
with lower financial leverage will be more likely to be assigned
a higher S&P rating. This evidence suggests that S&P place
more weight on leverage in determining the overall rating
assigned to insurers than do A.M. Best.24
(iii) General Conclusions

We contend that the results of our research have potentially
important implications for firms operating in insurance mar-
kets as well as for policy-makers and industry regulators. For
example, the evidence that mutual insurers are generally assigned
higher letter ratings than stock companies suggests that certain
publicly-traded insurance firms, in particular new entrants,
might not possess sound financial strength and may require
closer regulatory scrutiny than more established insurers. Also,
the finding that liquidity has a significantly positive effect on the
ratings assigned by both credit rating agencies should provide a
measure of confidence in the robustness of the ratings to industry
regulators, policyholders and investors in the UK. The results
of this study suggest that such ratings could eventually play a
role in substituting for costly industry regulation (as is
currently the case, for example, in the New Zealand insurance
market). However, from the point of view of UK insurance
firms considering applying for an external credit assessment,
the most important finding may be that the factors which influ-
ence the propensity to be rated differ in a number of important
respects from those which appear to determine the ratings them-
selves. In particular, the likelihood of having an external credit

assessment is negatively related to firm size despite the fact that

the agencies’ ratings do not appear to be influenced by this type
of information in practice. In fact, only one variable, profitability,
influenced the propensity to be rated by both credit rating
agencies and the assigned ratings in a consistent manner.
This evidence suggests that a gap may exist between insurance
company managers’ perceptions and the practical reality of the
external rating process. For example, managers of small insurers
might seek a rating in order to enhance their prospects for
corporate growth and development, while credit rating agencies
may take the view that the strain associated with new business
growth will reduce insurers’ reserves and so inhibit their ability to
meet policyholders’ claims. Theoretically, therefore, the likelihood
of being rated may not be conjointly related to the grade of letter
rating that is actually assigned. Further, managerial decisions
to be rated could be motivated by possible signalling effects
such as the desire to convey financial strength to the insurance
and financial markets in order to secure new business growth and
a lower market cost of capital. In contrast, the credit rating
agencies may, through the rating process, be helping to resolve
inherent agency conflicts between policyholders and other
contracting constituents in insurance firms. For example, a ‘good’
rating could help to assure policyholders that managers are ade-
quately protecting their long-term fixed claims through prudent
financial practices – thus reducing their monitoring expenditures.
The results of this study point to the need for further theoretical
work on what motivates the rating process in credit rating agencies
and their client companies. In particular, conventional signalling
theory purports that the role of external credit ratings and other
financial intermediaries is to partially eliminate information
asymmetry regarding the fundamental worth of future income
streams (Diamond, 1984; and Millon and Thakor, 1985). The
evidence presented here suggests that in practice insurers may
volunteer for an external rating because they discern significant
validation benefits in the possession of a rating, over and above
the achievement of a particular letter. The results therefore
appear to pose the question of whether or not the theoretical
and practical benefits of attaining any class of rating, over and
above those of achieving a specific level of rating, are exclusive to
credit ratings of a voluntary nature (such as those in the insurance

industry) or if instead the bond rating process is perceived in the

same way by debt issuers.
We believe that this study addresses an important omission in
the literature about players in the UK insurance market, but
acknowledge that the interpretation of our results should be
tempered by limitations in the research design, such as the use
of a relatively small sample size compared with US studies. We
have, however, attempted to derive robust results, for example,
through the use of panel data and by employing the ordered
probit model in a form that takes full account of possible self-
selection bias. Finally, we hope that this study will stimulate
more research into the impact of company-specific factors on
the financial strength of insurance firms and indeed, the like-
lihood of financial distress in the economically and politically
important financial services sector as a whole. In particular,
future studies could usefully examine why a marked difference
exists in the way a variable such as liquidity affects the likelihood
of being assessed by two of the world’s leading rating agencies,
as well as investigating why the factors which influence the
likelihood of being rated differ in a number of key aspects
from those which determine the ratings themselves. This line
of enquiry should also provide useful developments to the
insights provided in this study about the rating process in
a non-mandatory context, and help determine whether the
evidence is exclusive to the insurance industry.
APPENDIX
Ordered Probit Models and Marginal Effects
In our models, where we assume a normal distribution in

which y can take on four values (0, 1, 2 or 3), we can write the
probabilities (P) associated with each y-value as:
Pð y ¼ 0Þ ¼ ð 0 xÞ
Pð y ¼ 1Þ ¼ ð1 0 xÞ ð0 xÞ
Pð y ¼ 2Þ ¼ ð2 0 xÞ ð1 0 xÞ
Pð y ¼ 3Þ ¼ 1 ð2 0 xÞ

where is the cumulative normal distribution function. The

marginal effects are then given by:
@Pð y ¼ 0Þ
¼ ’ð 0 xÞ
@x
@Pð y ¼ 1Þ
¼ ½’ð0 xÞ ’ð1 0 xÞ
@x
@Pð y ¼ 2Þ
¼ ½’ð1 0 xÞ ’ð2 0 xÞ
@x
@Pð y ¼ 3Þ
¼ ’ð2 0 xÞ
@x
where ’ is the standard normal density function. In these

models, with four y-categories, it is likely (but not certain) that
the marginal effects will have opposite signs to those of the coeffi-
cients for y ¼ 0 and y ¼ 1, but the same signs for y ¼ 2 and y ¼ 3.
NOTES
1 A useful review of some of the early literature is provided in Perry et al.
(1988). Chattopadhyay et al. (1997) return to the issue of the determinants
of bond ratings using Canadian data and demonstrate that ratings are not
affected by changes in the accounting treatment of deferred tax.
2 Pottier and Sommer (1999) provide a helpful historical review of this
literature and describe the relevance of such studies to examinations of
insurance company ratings.
3 For example, no bond given an initial rating of AAA by S&P has defaulted in
the twelve months after issue for more than thirty years.
4 Although early studies by Weinstein (1977), Pinches and Singleton (1978)
and others suggested that equity and debt prices change prior to external
bond rating decisions, more recent work by Holthausen and Leftwich
(1986) and Nayar and Rozeff (1994) suggests that bond ratings do provide
price-sensitive information to market participants. Liu et al. (1999) develop
the earlier studies by demonstrating that, for a sample of 258 US bonds
rated by Moody’s between 1977 and 1983, the effect on share prices of
downgradings is greater than in the case of upgrades. Leftwich (1998)
concludes that ‘bond yields are associated more strongly with ratings than
with publicly available data alone’ (p. 240).
5 In the early 1990s several US insurers, including Mutual Benefit and Uni-
versal Life, became insolvent. The credit rating agencies failed to predict
these events (Carson and Scott, 1997).
6 Theoretically, there is a complementary linkage between the voluntary
decision to seek a rating and the grade of rating received. For example,
insurance company managers will be motivated to opt for a rating to

achieve some goal such as signalling to the markets that their companies
are financially strong. However, the inherent self-selection bias in the
rating process means that managers will only volunteer to obtain a credit
rating ex-ante if they expect the outcome to be favourable and consistent
with their self-interest objectives.
7 Although A.M. Best have rated the financial condition of US insurers since
the 1920s (Cantor and Packer, 1995 and 1997) the use of credit rating
agencies outside North America is a more recent phenomenon; for example,
the firm only began rating UK insurers in the 1980s.
8 The requirement of a company visit means that during our sample period
the cost of a rating by A.M. Best is likely to have been higher than that for
S&P. Fee negotiations are also confidential and will tend to vary among insurers
depending on the complexity of the task. As a result, an analysis of company-
specific rating charges between different agencies could not be carried out in this
study.
9 We acknowledge that some measures of financial strength (e.g., risk-based capital
models) may vary between the life and non-life insurance industries. However,
the variables used in this study to measure financial condition (e.g., size and
liquidity) are reasonably consistent between life and non-life insurers and in any
case industry differences will be highlighted by our business activity variable.
Additionally, grouping life and non-life insurers together helps to increase the
sample size and enhance the efficiency of our parameter estimates.
10 Insurers, particularly those operating in life insurance markets, often
incur high expenses when acquiring new business (e.g., in the form of
medical expenses and sales commission). This can place a considerable
strain on the insurer’s ability to undertake new insurance unless there is an
adequate capital base.
11 A multinomial logit model was used by Powell (1997) to investigate take-
over likelihood in the UK.
12 If the null hypothesis, ¼ 0, cannot be rejected, then the results obtained
from the ordered probit model with sample selection will not differ
significantly from the results that would be obtained from an ordered
probit model without sample selection.
13 A.M. Best defines superior strength as Aþþ/Aþ (coded here as 3); excellent
as both A (coded 2) and A (coded 1); very good as Bþþ (also coded 1) and
good, adequate or uncertain claims-paying ability as Bþ and below (coded 0).
S&P define superior strength as AAA, excellent as AAþ/AA/AA and very
good as Aþ (all coded here as 3); good as A (coded 2) and A (coded 1) and
adequate or uncertain claims-paying ability as BBBþ and below (coded 0).
The codings reflect our judgement given the distributional properties of the
observed data set.
14 Subsidiaries of life and non-life insurance companies that separately
obtained external ratings were excluded to avoid duplication. Our action
in this respect was also motivated because users of ratings (e.g., prospective
investors) are more likely to be interested in the financial condition of
the corporate group rather than that of satellite entities. Reinsurance
companies and Lloyds were also excluded from our study because they
are substantively different from direct insurers. For example, reinsurance
companies do not write direct business and are therefore not subject to the
same underwriting risks as conventional insurers. Lloyds, meanwhile,
is not a corporate form of organisation, but is instead a collection of

underwriting interests (or Names). Furthermore, the Lloyds market

operates a three year accounting cycle to reflect the special (long-tail)
nature of much of the risks that it insures/reinsures.
15 The number of A.M. Best-rated companies varied across years as follows:
38 in 1997; 40 in 1996; 39 in 1995; 39 in 1994; and 39 in 1993. Our total
A.M. Best sample thus represents an unbalanced panel of 195 company/
year observations. Our sample of 25 S&P rated companies was constant
across all years, 1993–97. Few UK-based insurers use the ratings of Moody’s
Investors Services, so tests involving this agency could not be undertaken.
The 10 or so UK insurers that were jointly rated by both A.M. Best and S&P
over the period of analysis were excluded from our sample to facilitate
testing between the A.M. Best and S&P rated firms.
16 Data limitations prevented us from collecting a larger sample of non-rated
insurance companies.
17 Most stock insurers in our sample are publicly listed insurers and as such,
distinguishing between different forms of stock ownership would not have
rendered meaningful results.
18 As we argued above, the line of business is a likely influence on the
likelihood of being rated, but it is unlikely to influence the rating itself
separately from the other company-specific variables included in the rating
determination model. Indeed, experiments in which BUS was included in
the rating determination model found that the estimated coefficient was
consistently insignificant at the 0.10 level and beyond (two tail).
19 See Gujarati (1995, pp. 338–39), for a discussion of ‘rule-of-thumb’
methods of detecting multicollinearity. Gujarati suggests that a VIF in
excess of 10 would indicate a high degree of collinearity.
20 All estimation results were calculated using LIMDEP7.
21 The Kullback-Leibler R2 is a measure of goodness-of-fit relative to a model
with just a constant term.
22 The probabilities associated with each y-value and the equations of the
marginal effects are shown in the Appendix.
23 Senior management might favour demutualisation if, for example, it
enhances their compensation through equity options. On the other hand,
demutualisation can expose senior management to the discipline of the
market for corporate control and thus threaten their job security.
24 One possible reason for this result is that in our period of analysis S&P did
not generally perform ‘rating audit’ visits, and as such they may have had to
place heavy reliance on conventional financial measures of strength (like low
leverage). In contrast, A.M. Best representatives visited insurance compa-
nies and talked to managers on an ongoing basis, thereby allowing a first
hand view of managerial control consciousness to be formed. This in turn
suggests that A.M. Best may have been in a position to place relatively more
emphasis on their own judgement (of the adequacy of managerial systems)
than on leverage measures derived from published accounting sources.
REFERENCES
Adams, M.B. (1995), ‘Balance Sheet Structure and the Managerial-Discretion
Hypothesis: An Exploratory Empirical Study of New Zealand Life Insur-
ance Companies’, Accounting and Finance, Vol. 35, pp. 21–45.

Adams, M.B. (1996), ‘The Reinsurance Decision in Life Insurance Firms: An

Empirical Test of the Risk-Bearing Hypothesis’, Accounting and Finance,
Vol. 36, pp. 15–30.
Ambrose, J.M. and A.M. Carroll (1994), ‘Using Best’s Ratings on Life Insurer
Insolvency Prediction’, The Journal of Risk and Insurance, Vol. 61, pp. 317–27.
——— and J.A. Seward (1988), ‘Best’s Ratings, Financial Ratios and Prior
Probabilities in Insolvency Prediction’, The Journal of Risk and Insurance,
Vol. 55, pp. 229–44.
Barron, M.J., A.D. Clarke and S.H. Thomas (1997), ‘The Effect of Bond
Rating Changes and New Ratings on UK Stock Returns’, Journal of Busi-
ness Finance & Accounting, Vol. 24, pp. 497–509.
Best, A.M. (1999), Insurance Companies Directory (International Edition, Old-
wick, NJ: A.M. Best).
Borde, S.F., K. Chambliss and J. Madura (1994), ‘Explaining Variation in Risk
Across Insurance Companies’, Journal of Financial Services Research, Vol. 8,
pp. 177–91.
Bouzouita, R. and A.J. Young (1998), ‘A Probit Analysis of Best Ratings’, The
Journal of Insurance Issues, Vol. 21, pp. 23–34.
Brotman, B.A. (1989), ‘Reliability of Best’s Insurer Ratings Using Financial
Information Published in the Annual Report’, Journal of Insurance Issues
and Practices, Vol. 12, pp. 58–70.
Cantor, R. and F. Packer (1995), ‘The Credit Rating Industry’, The Journal of
Fixed Income, Vol. 5, pp. 10–34.
——— ——— (1997), ‘Differences of Opinion and Selection Bias in the Credit
Rating Industry’, The Journal of Banking and Finance, Vol. 21, pp. 1395–417.
Carson, J.M. and W.L. Scott (1997), ‘Life Insurers and the ‘‘Run on the Insurer’’
Exposure’, Journal of the American Society of CLU & ChFC, pp. 44–48.
Chattopadhyay, S., F.J. Arcelus and G. Srinivasan (1997), ‘Deferred Taxes and
Bond Ratings: A Canadian Case’, Journal of Business Finance & Accounting,
Vol. 24, pp. 541–57.
Datta, P. and N.A. Doherty (1990), ‘The Effects of Organisational Form on
Capital Structure: The Case of Stock and Mutual Property-Liability
Insurance Firms’, in H. Lonberge (ed.), Risk, Information and Insurance
(Norwell, MA: Kluwer Academic Publishers).
Diamond, D. (1984), ‘Delegated Monitoring and Financial Intermediation’,
Review of Economic Studies, Vol. 51, pp. 393–415.
Durbin, J.A. and D. Rivers (1990), ‘Selection Bias in Linear Regression, Logit
and Probit Models’, in J. Fox and J.S. Long (eds.), Modern Methods of Data
Analysis (Newbury Park, CA: Sage Publications).
Fok, L.Y., W.M. Fok, P.P. Wei and S.M.L. Lee (1997), ‘The Effects of Firm
Size on Risk and Profitability of the Property and Casualty Insurance
Industry’, The Journal of Insurance Issues, Vol. 20, pp. 25–36.
Greene, W.H. (1999), Econometric Analysis (2nd ed. Englewood Cliffs, NJ:
Prentice-Hall).
Gujarati, D.N. (1995), Basic Econometrics (3rd ed. New York: McGraw-Hill).
Heckman, J. (1979), ‘Sample Selection Bias as a Specification Error’, Econo-
metrica, Vol. 47, pp. 153–61.
Holthausen, R.W. and R.W. Leftwich (1986), ‘The Effect of Bond Rating
Changes on Common Stock Prices’, Journal of Accounting and Economics,
Vol. 17, pp. 57–89.

Jensen, M.C. (1986), ‘Agency Costs of Free Cash Flow, Corporate Finance and Take-
overs’, American Economic Review: Papers and Proceedings, Vol. 76, pp. 323–29.
Klein, R.W. (1992), Insurance Company Rating Agencies: A Description of their
Methods and Procedures (Washington DC: National Association of Insur-
ance Commissioners).
Leftwich, R. (1998), ‘Evaluating the Bond Rating Agencies’, in Mastering
Finance (FT Pitman, London).
Liu, P., F.J. Seyyed and S.D. Smith, (1999), ‘The Independent Impact
of Credit Rating Changes – The Case of Moody’s Ratings Refinements
on Yield Premiums’, Journal of Business Finance & Accounting, Vol. 26,
pp. 337–63.
Mayers, D. and C.W. Smith (1994), ‘Managerial Discretion, Regulation and
Stock Insurer Ownership Structure’, Journal of Risk and Insurance, Vol. 61,
pp. 638–55.
McKelvey, R.D. and W. Zavoina (1975), ‘A Statistical Model for the Analysis of
Ordinal Level Dependent Variables’, Journal of Mathematical Sociology,
Vol. 4, pp. 103–20.
Millon, M.H. and A.V. Thakor (1985), ‘Moral Hazard and Information Shar-
ing: A Model of Financial Information Gathering Agencies’, Journal of
Finance, Vol. 40, pp. 1403–22.
Nayar, N and M.S. Rozeff (1994), ‘Ratings, Commercial Paper, and Equity
Returns’, Journal of Finance, Vol. 49, pp. 1431–49.
Perry, L.G., P. Liu and D.A. Evans (1988), ‘Modified Bond Ratings: Further
Evidence on the Effect of Split Ratings on Corporate Bond Yields’,
Journal of Business Finance & Accounting, Vol. 15, pp. 231–41.
Pinches, G.E. and J.C. Singleton (1978), ‘The Adjustment of Stock Prices to
Bond Rating Changes’, The Journal of Finance, Vol. 33, pp. 29–44.
Pottier, S.W. (1997), ‘Life Insurer Risk Characteristics and the Rating Process’,
The Journal of Insurance Issues, Vol. 20, pp. 111–30.
——— (1998), ‘Life Insurer Financial Distress, Best’s Ratings and Financial
Ratios’, The Journal of Risk and Insurance, Vol. 65, pp. 275–88.
——— and D.W. Sommer (1997), ‘Agency Theory and Life Insurer Ownership
Structure’, The Journal of Risk and Insurance, Vol. 64, pp. 529–45.
——— ——— (1999), ‘Property-Liability Insurer Financial Strength Ratings:
Differences Across Rating Agencies’, The Journal of Risk and Insurance,
Vol. 66, pp. 621–42.
Powell, R.G. (1997), ‘Modelling Takeover Likelihood’, Journal of Business
Finance & Accounting, Vol. 24, pp. 1009–35.
Weinstein, M.I. (1977), ‘The Effect of a Rating Change Announcement on
Bond Price’, Journal of Financial Economics, Vol. 5, pp. 329–50.

Determinants of Credit Ratings in UK Insurance Industry

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Determinants of Credit Ratings in UK Insurance Industry

Uploaded by

Copyright:

Available Formats

Journal of Business Finance & Accounting, 30(3) & (4), April/May 2003, 0306-686X

The Determinants of Credit

MIKE ADAMS, BRUCE BURTON AND PHILIP HARDWICK*

The role of credit rating agencies in overseeing corporate finan-

Standard and Poor’s (S&P) can be modelled accurately using

# Blackwell Publishing Ltd 2003

an opinion on an insurer’s financial strength, including its opera-

In the light of the potential importance of credit ratings of

# Blackwell Publishing Ltd 2003

competence (Pottier, 1997 and 1998; and Pottier and Sommer,

# Blackwell Publishing Ltd 2003

ratings that are non-mandatory in nature and a contribution is

(i) Use of Ratios in the Insurance Industry

# Blackwell Publishing Ltd 2003

This in turn suggests that managers of highly leveraged insurers

# Blackwell Publishing Ltd 2003

effectively and to set competitive rates of premium. Managers of

H2a: Other things being equal, the higher the profitability

# Blackwell Publishing Ltd 2003

H4a: Other things being equal, the greater the growth in

(vi) Company Size

# Blackwell Publishing Ltd 2003

(vii) Organisational Form

H6a: Other things being equal, mutual insurers are more

# Blackwell Publishing Ltd 2003

future losses and enable a primary insurer to sustain an external

H7a: Other things being equal, the lower the amount of

(ix) Business Activity

# Blackwell Publishing Ltd 2003

tarily seek a letter rating. In contrast, insurers writing short-term

H8: Other things being equal, general and composite

where 0j is a vector of the parameters to be estimated for each j

# Blackwell Publishing Ltd 2003

the dummy dependent variable is set equal to 0 for non-rated

dit ¼ 0 xit þ uit

yit ¼  0 zit þ eit

# Blackwell Publishing Ltd 2003

½ yit ; xit is observed if and only if dit ¼ 1:

(ii) Data and Variables

# Blackwell Publishing Ltd 2003

profitability as the ratio of annual investment and underwriting

dit ¼ 0 þ 1 LEVit þ 2 PROFITit þ 3 LIQit þ 4 GROWTHit

yit ¼ 0 þ 1 LEVit þ 2 PROFITit þ 3 LIQit þ 4 GROWTHit

# Blackwell Publishing Ltd 2003

(i) Descriptive Data

# Blackwell Publishing Ltd 2003

Panel A: Key Summary Statistics for the Sample

Variables Mean SD Mean SD Mean SD

LEV 0.73 0.12 0.88 0.06 0.90 0.09

VIFs PROFIT LIQ GROWTH LNSIZE OFORM REINS BUS

LEV 1.58 0.52** 0.16 0.01 0.34** 0.08 0.05 0.13

# Blackwell Publishing Ltd 2003

sample period. This evidence may reflect a change in the relative

(ii) The Rating Likelihood Multinomial Logit Model

# Blackwell Publishing Ltd 2003

estimates in this model is not straightforward, so we concentrate

Panel A: Parameter Estimates

Variable Parameter Estimate Standard Error Estimate Standard Error

CONSTANT 0 12.08 3.51** 10.17 2.48**

LEV 0.990 1.930 2.920

# Blackwell Publishing Ltd 2003

Leverage, profitability, liquidity, growth, company size and busi-

# Blackwell Publishing Ltd 2003

dit ¼ 0 xit þ uit

yit ¼ 0 zit þ eit

dit ¼ 0 þ 1 LEVit þ 2 PROFITit þ 3 LIQit þ 4 GROWTHit

yit ¼ 0 þ 1 LEVit þ 2 PROFITit þ 3 LIQit þ 4 GROWTHit

LEV 1.58 0.52 0.16 0.01 0.34 0.08 0.05 0.13

CONSTANT 0 12.08 3.51 10.17 2.48

CONSTANT 0 0.40 1.04 9.44 3.75**

where is the cumulative normal distribution function. The