Dynamic Financial Analysis in Insurance

Yung-Ming Shiu
Assistant Professor
Department of Finance, Tunghai University, Taiwan
Tel: +886-4-23590121 ext. 3540
Fax: +886-4-23506835
Email: yungming@thu.edu.tw

Abstract

The past decade has witnessed the gradual emergence of dynamic financial analysis (DFA) in
the insurance industry. Actuaries employ DFA techniques to model the uncertainty of insurance
operations. This industry has various areas with potential applications for DFA such as solvency
testing, asset allocation, and capital allocation. Since DFA has enjoyed wide popular support
among insurance practitioners and academics, it is not surprising that insurance regulators across
the world have spent particular attention on DFA and its variants such as dynamic solvency testing
(DST) and dynamic capital adequacy testing (DCAT). This article presents an overview of DFA
issues, approaches, applications and practices. The specific purposes of the survey are two-fold:
first, to review DFA applications in insurance; and second, to document the extent to which DFA
techniques have been employed.

Keywords: Dynamic financial analysis; Dynamic solvency testing; Dynamic capital adequacy
testing; Scenario testing; Stochastic simulation; Insurance

1
1. Introduction

Dynamic financial analysis (DFA) involves projecting the financial position of an insurer into
the future under various groups of consistent assumptions with a view to evaluating its solvency
status and management strategies. Over the last decade, a number of the world’s major insurers
have developed sophisticated DFA models. Since there has been a recent trend among insurance
supervisory authorities towards adopting risk-based approaches to regulating insurance business,
the issue of DFA’s potential applicability for supervisory purposes has gradually gained
prominence. It is anticipated that insurance firms currently not using DFA or its variants will
probably adopt something along the lines of DFA in the near future.

It is worth noting at the outset that DFA is given different names in different contexts. Dynamic
solvency testing (DST) and dynamic capital adequacy testing (DCAT) are variants of DFA
specifically applied in solvency and capital adequacy testing respectively, and financial condition
reports (FCR) are the outputs of DFA/DST/DCAT models. In this article, DFA is sometimes
employed as the generic term for the various types of related applications such as DST and DCAT.
It is also worthwhile to mention that in 2003 the Dynamic Financial Analysis Committee of the
Casualty Actuarial Society (CAS) in the USA, the dedicated advocate of DFA, changed its name
to the Dynamic Risk Modeling (DRM) Committee to reflect the fact that DFA is just one type of
DRM. Nevertheless, because market practitioners are relatively familiar with the term of DFA
rather than DRM, the term of DFA will be used in this article.

This paper presents an overview of DFA approaches, issues and relevant studies in insurance,
and provides a description of the state of practice in DFA modeling. The purposes of this research
effort are two-fold: first, to survey a number of DFA applications in insurance; and second, to
document the extent to which DFA techniques have been utilized in the industry. Many
bibliographic references are given throughout the whole paper. I concentrate on a few important
issues concerning DFA and offer my views on what we have learned from the research on DFA.

The remainder of this article is structured as follows. The next section introduces and describes
various financial modeling techniques and some basic ideas about DFA. Section 3 addresses a
number of important DFA issues and examines DFA techniques. Section 4 deals with places where
DFA models have been applied in practice. Section 5 presents the current guidance to actuaries
conducting DFA/DST/DCAT/FCR in the UK, the USA and Canada. A DFA example is then given
in Section 6. The final section summarizes and concludes the paper.

2. Financial modeling techniques

2.1 The Evolution of Financial Modeling Techniques

2
 Over the years, financial institutions have utilized and adapted financial modeling techniques
which have been well developed for quite a while to evaluate the effects of a vast array of risks on
their financial condition. These techniques evolve from simple and static forecasting to
sophisticated and dynamic modeling. Before investigating the techniques in more detail, it is
worthwhile to discuss two main types of approaches: deterministic and stochastic. The
deterministic approach is actually a “what if” analysis. The values of the variables in the model
are predetermined using this method. However, those in stochastic modeling are randomly
selected from an assumed probability distribution. As illustrated in Fig. 1., at least three stages of
this evolution can be identified.

Deterministic Stochastic
approach approach

Static forecasting

Sensitivity testing

Scenario testing Stress testing Scenario testing Stress testing

Fig. 1. The evolution of financial modeling techniques

Stage I: Static forecasting

Static forecasting assuming that the state of the insurer and the environment in which it
operates will continue unchanged is actually a simple

3
 model which projects the future financial condition of the company. It predicts the firm’s future
likely situation based on one set of assumptions regarding key variables in question. For instance,
the insurer can project its surplus for the next five years using a group of consistent assumptions
concerning assets, liabilities, economic conditions, and other important variables.

Stage II: Sensitivity testing

Sensitivity testing is also known as sensitivity analysis. Unlike static forecasting, sensitivity
testing expands the projection by forecasting a number of possible circumstances. This technique
is usually conducted by changing one key variable at a point in time. By carrying out a series of
calculations it is possible to arrive at an overall picture of the nature of the key risks facing the
company and their impact on financial strength. There are two main problems with sensitivity
testing. The first is that it may be difficult to determine the extent by which the actuary should
change the variable in question when conducting the analysis. The second problem is that the
interrelationships between variables could be disrupted due to the change of only one variable at a
time.

Stage III: Scenario and stress testing

The main difference between sensitivity testing and deterministic scenario testing is that the
former is conducted by changing one variable at a time whereas the latter is carried out by
changing a group of variables. As stated previously, a deterministic scenario actually means that
the values of the variables are predetermined. The interrelationships between variables in one
scenario may be assumed. Deterministic scenario testing projects the trends of a company’s
financial condition under various future scenarios. Stress testing is just an extreme case of
deterministic scenario testing and usually involves the worst and relatively unlikely scenarios. If
the financial condition of the firm under these extremely rare scenarios is still acceptable, then the
risks are in general assumed to be tolerable.

Stochastic simulation is often incorporated into the scenario building process of scenario
testing. In this case, stochastic simulation is used for generating numerous scenarios. The
difference between stochastic modeling and deterministic scenario testing mentioned above is that
the values for the variables in the scenarios established in stochastic simulation are randomly
selected from the probability distributions assumed, whereas those in deterministic scenario
testing are predetermined deterministically. The scenarios established using stochastically
simulated values for

4
 the variables are referred to as stochastic scenarios whereas those established using
predetermined values for the variables are referred to as deterministic scenarios. It is worth noting
that sensitivity tests and stochastic simulation actually are all scenario testing. It’s all in how the
scenarios are selected. Further discussion about stochastic simulation and scenario testing can be
found in Subsection 3.2.

In the course of evolution of financial modeling techniques, at least two trends may be
identified. The first is that the nature of modeling techniques changes from deterministic to
stochastic and from static to dynamic. This trend has been shown and discussed above. The
second trend is that current approaches place more emphasis than ever on the interrelationship
between the variables in question. Both trends lead to the emergence of DFA.

2.2 Dynamic financial analysis: basic ideas

The definitions of DFA vary. From the investors' or shareholders' point of view, DFA is a
process whereby an actuary analyzes an insurer’s financial situation that refers to the prospective
ability of the firm’s capital and surplus to sufficiently support its future operations, according to
the DFA handbook published by the CAS (Casualty Actuarial Society, 1995). From the insurance
regulator's perspective, DFA is an approach to monitoring the financial soundness of an insurer
and its capability of meeting its obligations to policyholders (Canadian Institute of Actuaries,
1998). The Casualty Actuarial Society (2000) defines DFA as a systematic approach to financial
modeling in which financial results are forecasted under a wide variety of possible scenarios,
showing how financial indicators might be influenced by changing internal and external
conditions. DFA can also be defined in terms of its uses as follows. DFA is a tool that can be used
to quantify the financial effects of likely future economic conditions, and to evaluate the impact of
implementing different management strategies on an insurer’s performance.

Strictly speaking, DFA is an old approach to financial modeling. It actually originates from
asset-liability management (ALM). Due to their similarity, it is difficult to draw a distinction
between them. Christofides (2000) implicitly suggests that DFA models have better economic
scenario generators than traditional asset-liability models. Moreover, Kaufmann et al. (2001) point
out that DFA is applied almost exclusively to property-casualty insurance especially in North
America, whereas a similar concept in life insurance is still called ALM. Furthermore,
Cumberworth et al. (2000) simply regard DFA as ALM. It appears that Gorvett (1998) does as
well. In his research, he points out that DFA became important to the insurance industry mainly
because of the increased levels and volatility of interest rates in the past decades. Since the life
insurance industry was more exposed to interest rate risk than the non-life industry, the former
developed the idea of DFA earlier.

5
 To sum up, it is safe to say that DFA is a variant of ALM, and it seems that greater emphasis is
placed on both economic scenario generators and the interrelationships between assets and
liabilities in DFA models than in relatively traditional ALM models. This can be further confirmed
by the comments of D’Arcy et al. (1998) on DFA:

“It (DFA) provides a far more effective tool for forecasting future financial and operating
conditions of an insurance company than prior methods for two primary reasons. First, the
interactions between the underwriting and investment sides of the insurance business are formally
integrated. Second, this approach utilizes advances in computer technology and modeling
techniques to provide almost instantaneous feedback to decision makers, allowing for the
evaluation of numerous operating alternatives.”

2.3 Dynamic Financial Analysis and insurance companies: some history

Like other financial institutions, insurance firms have been utilizing and adapting these above-
mentioned financial modeling techniques to meet their needs. Since 1992 valuation actuaries of
life insurance firm operating in Canada have been conducting DST in accordance with the
Standard of Practice on Dynamic Solvency Testing for Life Insurance Companies issued by the
Canadian Institute of Actuaries (1991). According to the Canadian Institute of Actuaries (1993),
this Standard of Practice also had been applied to the appointed actuaries of fraternal benefits
societies since 1994. In 1999 the Standard of Practice on Dynamic Solvency Testing was replaced
by the Standard of Practice on Dynamic Capital Adequacy Testing. Based on the latest Standard
of Practice, all appointed actuaries of insurance companies operating in Canada are required by
the Superintendent of Financial Institutions to conduct Dynamic Capital Adequacy Testing
(DCAT) and prepare financial condition reports.

In the UK, a Working Party under the auspices of the Joint Actuarial Working Party was set up
in 1993 to consider whether DST should be formally introduced into the process of monitoring the
solvency of life insurers and whether a financial condition report should be prepared and made
available to the insurance regulatory authority. In 1996 the Faculty and Institute of Actuaries
issued Guidance Note 2 (GN2). According to GN2 appointed actuaries responsible for long-term
insurance business are encouraged to prepare FCR using DST.

Generally speaking, DST and DCAT in Canada use scenario testing of a deterministic nature.
There are ten deterministic scenarios suggested in the Canadian DST for initial testing, including
worsening mortality, morbidity and withdrawal rates, increasing and decreasing interest rates,
level and high new sales, sudden worsening in mortality and morbidity, and increased default and
expense rates. The Standard of Practice on DCAT lists 10 and 11 risk categories for life and non-
life insurers and suggests that at least three plausible adverse scenarios posing the greatest risk to
the company require scenario testing and reporting each year. As for the UK, GN2 lists four
6
assumptions that “there would need to be specific reasons for not testing”, eight assumptions that
“may be of considerable importance in some companies but not others”, and ten assumptions to
which the appointed actuary needs to be alert. GN2 points out that it may be helpful, as a matter
of routine, to test the effect of each assumption using sensitivity testing and that certain
assumptions, in particular those which are a consequence of the economic environment, are best
treated as a group using scenario testing.

Beard (1964) appears to be one of the first to state that an insurer can be regarded as collection
of liabilities, against which a group of assets is held. In his work, Beard argues that both assets
and liabilities have to be considered when assessing insurer solvency. Krouse (1970) considers
simultaneous selection of investments, underwriting lines, and capital financing to form efficient
mean-variance corporate portfolios. In particular, he focuses on the balance sheet accounts of the
company and considers the interrelationships among these accounts.

In general, all the variables in question in the above work were deterministically decided.
From the 1980s, the use of stochastic modeling has emerged. The members of the Finnish
Solvency Working Party, Pentikäinen and Rantala (1982), employed a stochastic model to assess
solvency margins. Limb (1984) employed a stochastic model to investigate life insurer solvency.
Unlike Limb (1984) focusing on life business, Ryan (1984) applied simulation techniques to
solvency testing for a non-life insurance firm. Pentikäinen et al. (1989) developed a relatively
complete model to demonstrate how to stochastically model the risks that may affect an insurer’s
financial position. A number of relevant works using stochastic simulation has been published.
Some of them include Ryan (1984), Daykin and Bernstein (1985), Daykin et al. (1987), Daykin et
al. (1990), Hardy (1993), Macdonald (1995), Hardy (1996), Berketi (1998), and Consigli (1998).

In fact, the evolution of solvency testing techniques can be generally regarded that of asset-
liability modeling, because the modeling of assets and liabilities is required when conducting
solvency testing. This evolution can be summarised in Fig. 2.

7
Fig. 2. The evolution of asset-liability modeling in insurance

Pentikäinen and Rantala (1982)

Beard (1964) Limb (1984)
Krouse (1970) Ryan (1984)
An insurance company can be regarded Pentikäinen et al. (1989)
as collection of liabilities, against which Focuses on the balance sheet Daykin et al. (1990)
a group of assets is held. accounts of the company Hardy (1993)
Both assets and liabilities have to be Considers the interrelationships Macdonald (1995)
simultaneously considered. among these accounts. Hardy (1996)

Stochastic simulation is used for

conducting solvency testing.

8
3. Dynamic financial analysis: issues and techniques

3.1 The process of conducting dynamic financial analysis

The process of conducting a DFA largely depends on its objectives and purposes. However, the
main steps in conducting such an exercise can be outlined as follows1:

Step 1: Investigate the risks faced by the insurer and the factors affecting its performance.

The first step in developing a DFA model for an insurer is to examine the risks faced by the
company. It should be noted that insurance firms are risk intermediaries and as a result run not
only general business risks such as financial and operational risks that are common to other
ordinary businesses but insurance-based risks such as reinsurers’ default and underwriting risks
which are specific to insurance. Several approaches to identifying the risks have been
recommended. First, the actuary of the insurer may consider the risks suggested by actuarial
professional bodies or researchers in insurance areas. These risks are summarized in Table 1. This
table could be used as a checklist and the actuary could investigate whether the company is
exposed to the risk factors listed in this table. It should be stressed that this table gives a non-
exhaustive list of risks. Since the circumstances of each insurer are unique, these risks should be
used as a stimulant to thinking about risk, rather than a definitive checklist. In addition, in order to
understand the possible risks in great detail, it is deemed advisable to examine the financial
statements of the company such as technical account (revenue account), non-technical account
(profit and loss account), and balance sheet. By examining these financial statements, the actuary
has a better understanding of the risk profile of the firm.

1
The reader can refer to Society of Actuaries (1996) and Burkett, Mclntyre and Sonlin (2001) for other ways of
listing the steps in conducting a DFA.

9
Table 1
Risks faced by insurance companies
Professional bodies/ Researchers
Feldblum (1992)
Society of Actuaries (1996)
Risks faced by insurance companies
1. Underwriting risk
• Catastrophes  • Underwriting cycles
• Regulatory action • Parameter risk
• Process risk
2. Reserving risk
• Unforeseen liabilities • External changes
• Internal changes • Inappropriate methods
3. Asset risk
• Default risk • Loss of principal
• Asset-liability matching
4. Other risks
• Reinsurance risk • Credit risk
• Management risk
1. Operational risk:
• Mortality risk • Morbidity risk
• Persistency risk
• Expense risk
• Investment risk
• Liquidity risk
10
 • Risk of mismanagement

2.
Environment risk:
• General economy
•Efficiency of economic markets
•Legal environment
•Competitive forces
•Society’s perception of insurance
•Governmental actions
•Accounting actions
•Changes in demographic structure
• Technological changes
•Public health
Faculty and Institute of Actuaries 1. Concentrations of assets in particular risk areas
(1996) 2. Derivatives
3. Assets containing unusual provisions which may be susceptible to particular risks
4. Sources of new business which have unusual characteristics
5. Impending major claims or litigation that might affect the company
6. Operational exposure to accidents, terrorism, or malicious damage
7. Unusual contracts or relationships which may have financial implications
8. Risks created by deficient product literature or policy documentation
9. Loss of a distribution channel
10. The effect in different scenarios of options and guarantees in the insurance liabilities.
D’Arcy et al. (1997) 1. Balance sheet risk
• Asset risk

11
 • Liability risk

2. Operating risk
• Underwriting risk
• Investment risk
Walling et al. (1999) 1. Pricing
2. Loss reserve development
3. Catastrophe
4. Investment
Casualty Actuarial Society (2000) 1. Asset risk
2. Obligation risk
3. Interest risk
4. Mismanagement risks
Australian Prudential Regulation 1. Insurance risk
Authority (APRA, 2000) 2. Investment risk
3. Concentration risk
Ryan et al. (2001) 1. Underwriting risk
Market
Premiums
Claims
Expenses
Social
Reinsurance
Legal/Legislative
Other economic

12
Currency
Political

2. Asset risk
Assets
Premium reserves
Financing risk
Other economic
Currency
Political

3. Other risk
Operational
Policyholders’ reasonable expectations
Dependency
Group structure

13
 In practice the actuary normally uses professional judgment to determine which risk factors
should be included based on his or her understanding of the company’s risk exposures. In general,
it is sufficient to do so. On some occasions, however, the actuary might not be able to identify all
the important factors or to give appropriate weights to the factors identified. Therefore, it is
necessary to develop a more scientific means to assist the actuary in finding out these important
factors.

The second approach is to empirically conduct a statistical/econometric analysis to determine

which economic and firm-specific factors should be incorporated within a DFA model. The
factors to be included are supposed to have financial impact on insurer performance. Sufficient
detail should be captured in the model, otherwise the model would be too simple to be
meaningful. Nevertheless, it is neither feasible nor necessary to include all the factors affecting
insurer performance in the model to represent the complicated reality. Therefore, the actuary
usually only needs to consider the factors that pose material threats to company performance.

The first recognition that the identification of important economic and firm-specific factors
provides a basis for actuaries to build insurer-specific DFA models was due to Browne, Carson,
and Hoyt (2001). In their paper, they estimate one-factor panel data models for three risk-
unadjusted and risk-adjusted measures of financial performance, including return on assets, return
on equity, and the percentage change in capital and surplus from year to year. They conclude that
over the period from 1985-1995 life insurer risk-unadjusted performance is positively related to
bond returns, disposable personal income per capita, and negatively related to unexpected
inflation. Also, over the same period risk-adjusted financial performance is positively associated
with firm size, and negatively associated with leverage and assets in separate accounts.

Particular attention should be paid to the significant factors affecting not only financial
performance but also other performance measures such as operational performance, solvency and
insolvency rate. Adams and Buckle (2003) investigate the determinants of operational
performance in the Bermuda insurance market and find that insurer operational performance is
positively related to underwriting risk and leverage, and negatively related to asset liquidity.
Using investment yield, percentage change in shareholders’ funds and return on shareholders’
funds to capture different aspects of insurance operations, Shiu (2004a) find that liquidity,
unexpected inflation, interest rate level and underwriting profits are statistically significant
determinants of the performance of UK non-life insurers. Shiu (2005) further conclude that
solvency determinants changed from one epoch to another based on panel data for the three
subperiods, 1986-1990, 1990-1994, and 1994-1999. He also found that solvency was positively
related to bonds-to-total assets, equities-to-total assets, and level of new business, while
negatively related to unexpected inflation, market competition, assets held to cover linked
liabilities-to-total assets, life and general annuity reserves-to-total reserves, pension reserves-to-
14
total reserves, permanent health reserves-to-total reserves, other reserves-to-total reserves, firm
size, and insurance leverage. Table 2 lists the empirical research on insurer performance
determinants. Among the articles in this study that identified market and economic predictors of
insolvency in the US property-liability and life-health insurance markets are Browne and Hoyt
(1995), and Browne, Carson, and Hoyt (1999) respectively. Their key findings are also
summarized in Table 2.

15
Table 2: Empirical studies regarding the determinants of insurance company performance
Researchers Nature of Sample Methodology Dependent Explanatory variable
variable Significant at the 5%
level*
Browne and Quarterly data of US Logistic Insolvency rate 1. Number of
Hoyt (1995) property-liability regression companies (+)
insurance companies model 2. Industry combined
for the period 1970
ratio (+)
through 1990
3. First quarter of the
year(+)

Browne, Quarterly data of US Poisson Insolvency rate 1. ∆(Disposable

Carson and life-health insurance regression personal income
Hoyt (1999) companies for the model per capita) (+/–)
period 1972 through 2. Number of insurers
1994
(+)
3. Second quarter
dummy (+)
4. Quarter number
(+/–)

Browne, Annual data of US One-way Percentage 1. (Separate

Carson and life insurance fixed-effects change in Accounts) / (Total
Hoyt (2001) companies for the model capital and assets) (–)
period 1985 through surplus 2. Log (Total assets)
1995
(+)
3. (Written
premiums) /
Surplus (–)
4. Reserves/ (Capital
and surplus) (–)
5. IRIS 9 (Surplus
relief) (–)
6. IRIS 10 (Change in
premium) (−)

One-way Return on 1. Disposal personal

random-effects assets income (+/–)
model 2. (Liquid assets)/
(Total assets) (+)
3. (Real Estate)/
(Total assets) (–)

16
 4. (Separate
Accounts)/ (Total
assets) (–)
5. (Ordinary life
reserves)/ (Total
reserves) (+/–)
6. (Annuity
reserves)/ (Total
reserves) (+/–)
7. (Written
premiums)/
Surplus (–)
8. Reserves/ (Capital
and surplus)
9. (–)
10. IRIS 6 (Non-
admitted to
admitted assets)
(–)
11. IRIS 10 (Change
in premium) (–)

One-way Return on 1. Disposal personal

random-effects equity income (+/–)
model 2. Unanticipated
inflation (–)
3. Bond portfolio
returns (+/–)
4. (Separate
Accounts)/ (Total
assets) (–)
5. Log (Total assets)
(+)
6. (Ordinary life
reserves)/ (Total
reserves) (+/–)
7. (Written
premiums)/ Surplus (–)
8. Reserves/ (Capital
and surplus) (–)

17
 9. IRIS 9 (Surplus
Relief) (–)
10. IRIS 10 (Change
in premium) (–)

Adams and Accounting data Two-way Percentage 1. Underwriting risk

Buckle of 47 major non- random-effects difference (–)
(2003) captive registered model between the 2. Leverage (+)
insurance and ratio of
3. Liquidity (–)
reinsurance annual
companies for the operating 4. Company type
period 1993 expenses (Direct insurance
through 1997 (including
commission) company = 0,
plus net reinsurance company
premiums =1) (+)
written and
the ratio of
net
investment
income to
net
premiums
earned

Shiu (2004b) Annual data of UK Ordinary least Investment 1. Liquidity (+)

non-life insurers squares, one- yield, 2. Unexpected
for the period way fixed- percentage
inflation (–)
1985 through effects and change in
1999 random-effects shareholders’ 3. Interest rate level
models funds, and (+)
return on
shareholders’ 4. Underwriting
funds profits (+)

Shiu (2005) Annual data of UK Ordinary least Free asset ratio 1. Bonds-to-total
life insurers for the squares and assets (+)
period 1985 one-way 2. Equities-to-total
through 1999 random-effects
assets (+)
model
3. Level of new
business (+)
4. Unexpected
inflation (–)
5. Market competition
(–)

18
 6. Assets held to cover
linked liabilities-to-
total assets (–)
7. Life and general
annuity reserves-to-
total reserves (–)
8. Pension reserves-to-
total reserves (–)
9. Permanent health
reserves-to-total
reserves (–)
10. Firm size (–)
11. Insurance leverage
(–)

Step 2: Choose reward and risk measures

The second step is to choose reward and risk measures in accordance with the purpose of the
analysis. Reward measure, also known as objective function (Almagro and Sonlin, 1995; Burkett,
Mclntyre, and Sonlin, 2001) or return measure (Bohra and Weist, 2001), serve as an instrument in
evaluating insurer performance under a wide range of scenarios or strategies. Expected surplus,
policyholder surplus, and shareholder’s equity are commonly seen reward measures. The standard
deviations of the abovementioned reward measures, expected policyholder deficit, tail Value-at-
Risk, and the probability of ruin over a three year period are risk measures. It is worth noting that

19
since insurance operations are complicated and interrelated, there could be a number of
theoretically and/or practically appropriate measures that can serve the purpose. Therefore, it may
be desirable for the actuary to simultaneously employ several measures in the analysis, so that the
risks inherent in the business venture can be more readily understood.

Step 3: Determine projection period

The projection period mainly depends on the characteristics of the risks to which the company
is exposed and management’s business planning horizon. This period should be long enough to
capture the full effects of the risks. As a result, the projection period of a life insurer generally is
longer than that of a non-life insurer. There are two primary reasons for this. First, the liabilities
and assets of a life firm are relatively long-term compared with those of a non-life company. The
effects of the risks of a life company generally take longer to become apparent than those of a
non-life firm. Second, the liabilities of a non-life company are relatively uncertain compared to
those of a life company because both occurrence time and claims amount of non-life policies are
unknown at the outset of a contract, whereas claims amount of life products generally is fixed.
There is no point in projecting cash flows of a non-life company for a very long time period
because the spread of variability increases with time and the projection accordingly would
become relatively unreliable as the projection period is lengthened.

Step 4: Build a DFA model

To build a stochastic DFA model, for example, the actuary normally chooses one or more
driving factors from economic factors. Then the driving factors are stochastically simulated to
generate likely future economic conditions. Common factors including interest rate and inflation
rate will be discussed in Subsection 3.3. It should be noted that the stochastic processes should be
calibrated before they can be employed to simulate. The aim of calibration is to determine the
appropriate values for parameters of stochastic processes. Once the future economic conditions
are simulated, the financial market returns are then consistently determined. Other important
issues on building a DFA model are discussed throughout this paper.

Step 5: Project cash flows

Once built, the model projects future cash flows. Thousands of iterations of financial results are
generated and output distributions of the results are produced. The higher the number of iterations,
the more reliable the distribution of financial results is likely to be. However, it should be noted
that the reliability could not be increased further as the simulation runs reach a certain limit
number.

Step 6: Interpret the results and provide feedback

20
 The financial results obtained from a DFA exercise should be carefully discussed and
interpreted. If the results under some plausible adverse scenarios are not acceptable, the actuary
has to identify the causes and suggest alternative possible corrective measures to be taken.

Step 7: Prepare a written report and present it to the Board

The preparation of a written report to senior management is the final step of the DFA. The
content of the report depends on the purpose of the analysis. For instance, the report is often
referred to as the financial condition report if the purpose of the analysis is to test the solvency of
an insurer under plausible adverse scenarios. The report should normally, at a minimum include
the purpose of the analysis, methods and assumptions, scenarios, findings, restrictions to the
analysis, recommendations, and so forth. It is worth noting that the presentation of the output of
the DFA results to the Board of Directors is important because an effective presentation may
facilitate the acceptance of DFA models. The interested reader can refer to Larsen et al. (2004).

3.2 Dynamic Financial Analysis techniques

Feldblum (1992) indicates that DFA techniques fall into two broad categories: scenario testing
and stochastic simulation. Shiu (2004b) further divides the techniques into three types including
sensitivity tests, scenario testing and stochastic simulation. It is worthwhile to note that insurers
may be doing sensitivity testing within scenarios, without identifying that specifically. In fact,
sensitivity tests and stochastic simulation are all scenario testing. It is all in how the scenarios are
selected. Scenario testing projects financial results under groups of assumptions of variables that
are assumed to change in a consistent way. Each set of consistent assumptions represents a
scenario. For instance, it is common to assume that high (low) inflation rates accompany high
(low) interest rates since inflation is highly correlated with interest rates. However, it should be
noted that the assumptions of consistency are not always valid. There could be some periods when
consistency has been invalidated. One of the possible reasons for inconsistency is the time lag
existing between variables. For example, it takes time for an insurer to adjust its capital shock in
response to a rise in the demand for its products. Stochastic simulation models uncertainty by
randomly selecting values from probability distributions for each variable. These values for each
variable are then used to calculate a large number of resulting scenarios. The primary difference
between scenario testing and stochastic simulation is that the former starts with building scenarios
in which variables are assumed in a consistent way, whereas the latter usually starts with the
assumption of independence between the variables which need to be simulated. In general, the
scenarios built for scenario testing are therefore more realistic than most of the resulting scenarios
generated stochastically for stochastic simulation. As to the main similarity, all the values for each
variable in the scenarios for scenario testing and stochastic simulation are input into a cash flow
21
model to calculate outcomes.

It is worthwhile to note that since the variables in questions are unlikely to be uncorrelated with
each other, the interdependence between them should be taken into account when building a DFA
model. For example, the changes in the value of a given asset may be correlated with those of
another asset. Similarly, the loss experiences of difference insurance products may tend to vary in
concert. In scenario testing, the interrelationship is considered by assuming these variables in a
consistent fashion, while in stochastic simulation the Cholesky Factorization, the Normal Copulas,
or the Cario-Nelson method can be utilized to incorporate correlation into the simulation process.
Additionally, a number of time-series approaches such as transfer functions can also be employed
to impose a covariance structure on a set of variables. These approaches provide methods to
generate correlated variables. If analysts fail to include these correlations in stochastic models, the
risk or uncertainty of the overall position of a company would be underestimated. Pentikäinen
(1988) indicates that the variables should not be assumed to be mutually independent and such an
assumption would lead to an underestimation of the risks. Feldblum (1992) also points out that
separate consideration of interrelated risks is insufficient and this interdependence of risks carries
the most danger for insurance solvency. Finally, the DFA analysts must bear in mind that there are
two major problems concerning correlation. The first is that correlations between variables are not
time-invariant over the forecast period. The second problem is that known dependency
relationships may not be maintained and past causal relationships are sometimes not indicative of
future relationships.

The techniques mentioned above have their own weaknesses and merits. For instance, one of
the drawbacks of stochastic simulation is that its results such as expected policyholder deficit
(EPD) are sometimes difficult to calculate or understand for the Board of Directors, management
or non-technical audiences. Stochastic simulation has several advantages. First, a wide range of
scenarios can be simulated with the help of the capability of modern computers. By increasing the
number of simulation runs, more scenarios can be obtained and simulation results are more stable.
Second, stochastic simulation can account for the stochastic nature of insurance operations
(D’Arcy et al., 1997). Through a number of stochastic simulations, an idea of likely future
prospects of a company in a variety of different circumstances can be obtained. Other advantages
of stochastic simulation such as graphical presentations of the uncertainty in the projections are
identified in Pentikäinen (1988).

Compared with those for stochastic simulation, the scenarios built for scenario testing are more
meaningful, tangible and consistent, and the results are accordingly relatively easy to understand.
However, these scenarios for scenario testing are limited to the actuary’s preconceived notions
about likely future adverse developments of economic and financial conditions, and may not be
exhaustive. Nevertheless, it should be noted that the scenarios for scenario testing may be
sufficient although they may not be exhaustive. If the actuary has a complete picture of the risk
22
profile of the firm and a good understanding of the market environment where the company
operates, it is possible to build sufficient scenarios.

Of these three techniques, what approach is better? The short answer is that it depends on the
types of questions asked. Each of them is more appropriate in certain situations. For example,
scenario testing would be the relatively appropriate approach to answering a question like: “how
would the net cash flows of an insurer change under the following conditions?”. When analytic
solutions become too complex to obtain or there are no closed-form solutions, a stochastic
approach would be more suitable than scenario testing. For instance, simulations are particularly
useful to determine “how much capital would be required in order that there is a 95 per cent
probability that the surplus of an insurer will fall below the beginning statutory surplus over the
next five years?”.

3.3 Driving factors and cascade structure in dynamic financial analysis models

In addition to employing the abovementioned approaches to addressing the interrelation issue,

the use of driving factors also known as generators, with a cascade structure also is commonly
seen. As depicted in Fig. 3, factors at the top influence those below and the impact is usually one-
way, i.e. irreversible. That is, factors at the lower tier of the structure cannot influence those at the
higher tier. It should be stressed that the structure does not imply any causality between factors.
Instead, it is the statistically significant correlations among variables that actuaries are trying to
capture. Driving factors usually are the factors at the highest tier of the structure. After selecting
driving factors and simulating their values, actuaries can subsequently and consistently determine
the values of some other variables, including those on both sides of the balance sheet.

Tier I:
Driving factor(s)

Tier II:
Factor II,1 on the asset side Factor II,2 on the liability side

Tier III:

Factor III,1 Factor III,2 Factor III,3 Factor III,4

Fig. 3. An illustrated example of a three-tier cascade structure

23
 Economic factors such as interest rate and inflation rate often serve as driving factors. The
former is used as the driving factor in DynaMo (Version 3.0), a public access DFA model for non-
life insurers (D’Arcy et al., 1997; 1998), whereas the latter is the generator in the Wilkie
investment model (Wilkie, 1986; 1995). Although it has been argued that it makes no huge
difference which economic factor is taken as the driving factor (Daykin, Pentikäinen and Pesonen,
1994), actuaries’ choices of generators reflect their emphasis on which side of the balance sheet of
an insurer. For instance, the movements of interest rates have financial impact on both sides of the
balance sheet especially on the asset side, and inflation is a major factor in determining the claims
costs. As a result, it seems reasonable to choose interest rate as the driving factor if more emphasis
is placed on the asset side, while inflation rate may be selected if great emphasis is put on the
liability side.

It is also noteworthy that the number of driving factors most depends on the purpose of the
DFA model. If it is built for forecasting, actuaries may improve the accuracy of forecasting by
including more driving factors in the model. For example, the CAP: Link, developed primarily for
ALM by Towers Perrin, uses both interest rate and inflation rate as driving factors. The highest
tier of the cascade structure consists of short and long interest rates, and price inflation (Mulvey
and Thorlacius, 1998). If the model is constructed for testing purposes such as resilience testing, it
appears that one driving factor is sufficient, controlling the other relevant variables by means of
assumed interrelationships (Daykin, Pentikäinen and Pesonen, 1994).

The rest of this subsection will briefly illustrate how the interest rate and inflation rate can be
employed as driving factors, using DynaMo and the Wilkie investment model as the examples. It
is worthwhile to note that DynaMo is a complete asset and liability, while the Wilkie model only
is an asset model.

It should be noted that different DFA models using interest rates as the driving factors may
employ various interest rate models to generate future likely interest rates. For instance, a variant
of the two-factor Brennan-Schwartz approach is used in the CAP: Link to generate future long
and short interest rates (Mulvey and Thorlacius, 1998). However, for the reasons provided by
D’Arcy et al. (1997, 1998), Walling et al. (1998), and Ahlgrim (2001) such as simplicity and non-
life insurers’ relatively less exposure to interest rate risk than life insurers, the Cox, Ingersoll and
Ross model (CIR model) is used in DynaMo. This model is a one-factor equilibrium model and
takes the following form:

dr t = a ( m − rt ) dt + σ rt dz t (1)

where rt is instantaneous short rate at time t; a is the parameter controlling average length of the
time of the mean reversion; m is the long-run mean level to which the short-term rate reverts; σ is
24
the volatility (standard deviation) of the short-term rate; dzt is a standard Wiener process
(Brownian motion).

For practical applications, the discrete-time form of this model shown below is employed:

∆rt = a ( m − rt ) ∆t + σ rt ∆z t (2)

After projecting the values of interest rates using formula (2), actuaries subsequently determine,
the inflation rate, the prices of bonds and other investments, premiums income and loss related
items. The reader can refer to D’Arcy et al. (1997, 1998) and Walling et al. (1998). The appendix
to D’Arcy and Gorvett (2004) also provides a concise account of DynaMo 3.

DynaMo allowing for two different lines of business consists of a number of modules. These
modules include investment, underwriting, reinsurance, and catastrophe. In the investment
module, there are various investment asset categories including bonds, stocks, cash, real estate
and mortgage loans. In the underwriting module, insurance business is divided into three types:
new business, first renewal business, and second and subsequent renewal business. This division
is to reflect the aging phenomenon, under which past loss experience improves with the length of
time the policies have been written by the insurer. As to reinsurance, two types of reinsurance
programs, stop loss reinsurance and catastrophe reinsurance are considered. In the catastrophe
module, the frequency and severity of catastrophic losses are assumed to follow Poisson
distribution and Lognormal distribution. A contagion matrix is provided to reflect the contagion
effect of catastrophic losses.

The Wilkie investment model is the first comprehensive UK stochastic asset model and is
extensively used by actuaries in the UK. Wilkie first proposed his model in 1986 (Wilkie, 1986).
He then updated this model by renewing the data and extending it to encompass more variables
such as wages index, yields on index-linked stock, short-term interest rates, property yield and
income, and currency exchange rates, and so on (Wilkie, 1995). Since these extensions left the
structure of the original model virtually unaltered (Huber, 1997) and it has a relatively succinct
cascade structure, the discussion here is mainly focused on the original model instead of the
extended one.

The original model is constructed based on a cascade structure that interrelates four variables,
including inflation rate, share dividend yield, share dividends, and yield on 2.5% Consols. This
model is displayed in Fig. 4.

25
 Inflation rate

Share yield

Share dividends Consols yield

The generator, the inflation rate, of the Wilkie model is actually a statistically stationary AR(1)
series and takes the following form:

I ( t ) = QMU + QA( I ( t − 1) − QMU ) + QSD * QZ ( t ) (3)

where I(t)= ln (Q(t)/Q(t-1)) is the inflation rate over the period of t-1 through t; QMU is the
parameter of the fixed mean; QA is the adjustment parameter, which decides the deviation of this
year’s inflation rate from the mean, by including the deviation of last year’s inflation rate from the
mean; QSD is the standard deviation and QZ(t) is a series of unit normal variable, i.e. QZ(t) ~ iid
N (0,1).

The best estimates of QMU, QA, and QSD recommended by Wilkie from an actuary’s point of
view are 0.05, 0.6,and 0.05 respectively. For space-saving purposes, the remaining processes in
the Wilkie model will not be discussed here. Interested readers are referred to Wilkie’s paper
(Wilkie, 1986). In addition, the Wilkie model has become the standard UK asset model that is
popular in the fields of actuarial science and finance. However, it certainly is not perfect. For a
review of the Wilkie model, see, for example, Kitts (1990), Geoghegan (1992), Daykin et al.
(1994) and Huber (1996). The reader interested in comparing Wilkie-type stochastic asset models
can refer to Lee and Wilkie (2000) and Rambaruth (2003) in which a number of actuarial models
for simulating future investment-related variables are discussed and their features are described.

4. Applications of dynamic financial analysis

26
 The uses of DFA are extremely extensive and the examples cannot be exhaustive. The current
applications include but not limited to: (1) solvency-testing the status of an insurer and assess its
financial status under a wide range of adverse economic and operating scenarios, (2) evaluation of
management strategies such as asset allocation and reinsurance programs, and (3) determination
of the amount of capital allocated to business units. In addition, DFA is also instrumental in
helping develop the business plan by identifying the potential external and internal threats to
company operations, and in pricing insurance products under a wide range of likely future
economic and financial conditions.

This section reports on the following insurance areas where DFA has been implemented:
solvency testing and management strategy evaluation. The former is concerned with assessing the
ability of an insurer to survive under different circumstances using DFA techniques. The latter
relates to the application of DFA to an insurer to examine the effectiveness and efficiency of the
company’s management strategies such as asset allocation, and capital allocation. Only studies
that focus on DFA are reviewed in this section. Also, some of the articles fit into more than one
application category, and were arbitrarily assigned a category.

4.1 Solvency testing

Ryan (1984) was one of the first to apply stochastic simulation to insurer solvency. He argues
that since the nature of insurance business is stochastic one who wishes to examine insurer
solvency has to carry out the investigation in a stochastic manner, especially in the context of non-
life insurance. In this article, he illustrates how simulation can be used to determine the capital
and retention levels of a non-life insurer in relation to different business strategies. The first step is
to show the probability that the company will hit the minimum statutory underwriting margin
during the period in question. Then the probability that the solvency margin will fall below the
starting solvency margin over a five-year period is calculated. By conducting simulations many
times, relatively reliable results can be obtained. Management would be able to identify primary
risk factors such as equity and liquidity exposures that affect the solvency margin and remove
them to bring the insolvency probability to an acceptable level.

Another solvency testing-related study was Daykin et al. (1984) which put great emphasis on
the importance of the market values of assets and liabilities rather than their cash flows. Daykin et
al. (1987) further examine the solvency of a non-life insurer and investigate the impact of
different scenarios on its financial health using stochastic simulation and an emerging costs
model, also known as a cash flow model, instead of employing the traditional balance sheet
concept. In order to show how to employ the emerging costs approach a model that projects cash
flows from non-life insurance operations is first built by Daykin et al. (1987). Following their
notation, the model is briefly presented as follows:

27
∆U j = ∑Ak , j −1 (1 + y k , j )(1 + g k , j ) − ∑Ak , j −1 + EP j − EXP j − DT j − ∑C i , j (4)
k k i≤ j

where ∆U j =U j −U j −1 , U j is the solvency margin at the end of year j , Ak , j the total value
of component k of the asset portfolio at the end of year j , y k , j the yield on asset component k
during year j , g k , j the percentage change in the value of component k of the asset portfolio
during year j , EP j the earned premiums in year j , EXP j the expenses including the cost of
administration, commission, and reinsurance in year j , DT j the amount paid out in dividends
and tax in year j , C i , j the claim paid in year j with regard to claims arising in year i .

The above model involves determining the changes in the solvency margin by projecting the
cash inflows and outflows. The cash inflows in this model include the changes in asset values, if
positive, and earned premiums, whereas the cash outflows expenditures, dividend payments, taxes
and, most importantly, claims.

With a view to investigating the effect of management strategies on solvency, Hardy (1996)
simulates the relative solvency of life insurance firms using a stochastic model office representing
a life mutual writing 25-year with-profits endowment insurance and non-participating term
insurance policies to lives age 30. In her article, an insurer is defined to be relatively insolvent if it
significantly falls out of line with the insurance market, and technically insolvent if the market
value of the assets is less than that of the liabilities, valued using the statutory interest basis
without allowance for the minimum solvency margin or the mismatching reserve. The
abovementioned office is referred to as the base office, and in this stochastic model office three
elements were assumed to be dynamic, including reversionary bonus declaration, valuation
interest rate, and equity proportion in asset portfolios. In addition to this base office, four variant
offices were considered, including high and low equity, high reversionary bonus, low estate
offices. She concludes that both the technical and relative solvency risks should be examined
when testing the solvency implications of different management strategies.

Muir and Sarjant (1997) use two examples to illustrate how to carry out DST using
deterministic scenario testing. The first example is concerned with a life insurer writing unit-
linked business. For simplicity, the only type of business written is a single premium pension
policy. The first step in examining the future solvency of the firm was to generate a base case
scenario. The base case scenario generally reflected the insurer’s business plan for the coming
years ahead or the company’s best estimate of the future. In this instance, three assumptions were
made: first, both the levels of new business and expenses increased at a constant rate; second, all
surplus arising was transferred to shareholder funds and shareholder funds over ₤2.5 million were
paid out as dividends. The second step was to select alternative scenarios. These scenarios are
generally considered by the actuary to have adverse effects on financial health of insurers. There
were five alternative scenarios employed by Muir and Sarjant (1997), including new business
28
volumes, withdrawal rates, investment returns, the level of the fund management charge, and the
level of expenses. Various approaches may be used to identify the scenarios such as checklists of
business exposures, analyses of the firm’s financial statements, and discussions with managers
throughout the firm and risk management consultants, etc.

The second example illustrated by Muir and Sarjant (1997) concerns a with-profits office. As is
known, with-profits business is more sensitive to investment performance than unit-linked
business whose assets and liabilities, by definition, are closely matched. Thus, more attention
should be paid to the office’s investment operations. Similarly, in addition to a base scenario, a
number of adverse scenarios were employed, including an immediate fall in the market values of
equities along with an increase in the redemption yield on fixed interest investments, and a
gradual increase in the market values of equities followed by a sudden fall.

Muir and Sarjant (1997) also conduct a stochastic modeling exercise and 500 simulations were
carried out. Noteworthy is that the model used can be operated on a one-step-at-a-time basis,
which mean that the proportions of components of the model can be rebalanced at the end of each
projection step to reflect the modeler’s decision under a particular circumstance.

For solvency testing purposes, the most commonly seen measure probably is the probability of
ruin. It is frequently estimated based on the simulation results. More specifically, it is expressed as
the number of insolvencies divided by the total number of simulations. The other commonly seen
measure is the probability that the surplus (or shareholders’ fund) will fall below the beginning
statutory surplus over a certain period of time. Similarly, this measure is expressed as the number
of simulations whose surplus is below the beginning surplus divided by the total number of
simulations.

4.2 Asset allocation

A number of papers have investigated the impact of different asset allocation strategies on the
solvency of a life insurer, such as Ross (1991), Ross and McWhirter (1991), Macdonald (1995),
and Ong (1995). The Wilkie investment model was utilized in these papers to model the asset
returns. By employing different investment strategies, the company would be able to judge how
sensitive the ultimate financial strength is to the investment strategy. A general discussion
regarding the above-mentioned papers can be found in Berketi (1998). Since the example
illustrated later in this paper is concerned with a non-life insurer’s asset allocation, the following
subsection will be focused on the major literature regarding the application of DFA techniques to
asset allocation for non-life insurers.

As illustrated in Fig. 5., the literature may be categorized as belonging to one of the following
groups in terms of the DFA models used.
29
30
 Fig. 5. Major research regarding the application of dynamic financial analysis in asset allocation for non-life business

Asset-Liability Management Efficient Frontier Russel-Yasuda Kasai Model FINANS Model ARMS Model The Smith Model
(ALMEF) Model (TSM)

Correnti and Sweeney (1994) Carińo et al. (1994) Kaufman and Ryan (2000) Christofides and Smith(2001)
Almagro and Sonlin (1995) Carińo and Ziemba (1998) Bohra and Weist (2001)

Total Integrated Risk Management Carińo, Myers and Ziemba (1998)

(TIRM) Model

Correnti, Nealon and Sonlin (1996)

Falcon Integrated Risk Management (FIRMTM)) Model

Correnti, Sonlin and Isaac (1998)

Sweeney et al. (1998)
Laster and Thorlacius (2000)
Burkett, Mclntyre and Sonlin (2001)

31
 The Asset-Liability Management Efficient Frontier (ALMEF) model was first employed by a
property and casualty insurance company, called USF&G (Correnti and Sweeney, 1994). The
main steps in allocating assets using the ALMEF model are as follows. The first step is to identify
representative asset classes such as fixed-income securities and equities. Asset indices are used as
proxies for asset classes such as fixed-income securities and equities. Then the economic
movements are modeled in order to join together the projected asset class returns and liability
cash flows from insurance operations (Almagro and Sonlin, 1995). The projected economic
movements at least include interest and inflation rates. Next, the objective function (reward
measure) and risk measure are determined and used for evaluating different asset allocation
strategies. Finally, the asset allocation strategies that provide the most efficient risk/reward
tradeoffs are said to be on the Asset-Liability Efficient Frontier. One of the main disadvantages of
the ALMEF model is that incorrect asset evaluation may take place because inappropriate proxies
are selected to serve as representations of assets.

Correnti, Nealon and Sonlin (1996) presents the Total Integrated Risk Management (TIRM)
model which is an extension of the ALMEF model. The main difference between the two is that a
stochastic, top-down and relatively complicated simulation model is explicitly included in the
former model. This stochastic simulation model is used for generating future interest rates using a
variant of the two-factor Brennan-Schwartz model. That is, interest rate is the driving factor of the
model. Once the future term structure of interest rates is determined, the top-down structure of the
model produces price inflation, stock dividend growth rates and yields. Finally, primary asset
class returns can be obtained accordingly. This simulation model is the predecessor of the CAP:
Link described in the previous section. The difference between the simulation model in Correnti,
Nealon and Sonlin (1996) and the CAP: Link is that the former has only one driving factor
(interest rate), whereas the latter has two driving factors (interest rate and inflation rate).

The TIRM model was further developed into the Falcon Integrated Risk Management (FIRM)
model by Correnti, Sonlin and Isaac (1998). One of the characteristics of the FIRM model is that
the total risk faced by an insurance company can be decomposed into its key components. This
model allows management to “loop” through the process by selecting either the investment loop
such as asset allocation or through the operations loop such as business mix. Sweeney et al.
(1998) modify this model to explicitly consider currency factors when assessing insurance asset
allocation strategies in a multi-currency environment. The FIRM model used in Laster and
Thorlacius (2000) is similar to that in Correnti, Sonlin and Isaac (1998). Burkett, Mclntyre and
Sonlin (2001) further improve this model by allowing a high level of flexibility in describing how
the underlying stochastic variables behave in an attempt to minimize model risk. Nevertheless, the
increase in flexibility has the result of moving a significant burden from the model to the model
user.

The Russell-Yasuda Kasai model is an asset-liability management model developed at a

32
Japanese non-life insurance company, Yasuda Fire and Marine Insurance Co., Ltd. This model
incorporates the company’s asset and liability mix, along with institutional constraints as well as
uncertain cash flows, disbursements, and taxes. It is designed to assist the company in deciding
how available funds should be allocated among potential investments to provide returns to cover
liabilities and provide for long-term growth of the company’s wealth (Carińo, Myers and Ziemba,
1998).

The Russell-Yasuda Kasai model consists of a matrix generator, a scenario generator, a liability
generator, a coefficient generator, and a solver. The main output of this model includes the optimal
asset allocation, the expected financial statements for each period (Carińo et al., 1994). The
procedure of asset allocation involves a number of meetings to develop the input data for the
model, including scenarios for the asset returns, business sales, and redemptions. Then the model
is run using these inputs. The detail about formulation of the model is described in Carińo and
Ziemba (1998).

It is noted that due to its multiperiod nature this model enables decision makers to change
portfolios through time to react to changing circumstances. Carińo, Myers and Ziemba (1998)
argue that Markowitz models are inadequate for decision problems that are inherently dynamic
and are progressively worse with increasing time horizons. In addition, as the time horizon
increases, the forecast error increases as well. In this paper, Carińo, Myers and Ziemba
demonstrate that the Russell-Yasuda Kasai model performs better than static mean-variance
Markowitz type models in terms of general understanding of the company’s business, and the
effect of various policies and constraints.

The FINANS model is a DFA model developed at Milliman & Robertson Inc. The main module
of the model is an economic scenario generator. This generator is a multi-equation econometric
model which develops quarterly projections of six economic and financial variables, including
gross domestic product growth, inflation rate, short and long term interest rates, and stock returns
and dividends. These projected variables are then used to produce accounts in financial statements
(Witcraft, 1998). Kaufman and Ryan (2000) investigate the impact of increasing equity holdings
on the financial performance of a non-life insurance company. In their paper, the FINANS model
is initially run for a base scenario using current financial statistics and asset allocation of the
insurance industry. Next the amount of equity holdings is varied to determine how selected
performance (return) and risk measures change.

Bohra and Weist (2001) use the ARMS model to evaluate different asset allocation strategies.
The ARMS model is developed at American Reinsurance Company using DFA techniques. It
consists of five main modules including the global economic module, asset module, liability
module, reinsurance module, and accounting module. The global economic module generates
future economic variables and capital market conditions. The generated future economic variables
33
and capital market conditions are then fed to the asset module, the liability module, and the
reinsurance module. The asset module prices current asset portfolios; the liability module projects
losses and expenses; the reinsurance module models the impact of all reinsurance terms. The
accounting module brings together all balances, cash flows, and accruals into an accounting
framework.

The Smith Model (TSM) is a multi-line, multi-period, multi-currency, multi-asset stochastic

plan generator. An economic scenario generator is used for projecting future term structure of
interest rates and inflation rate. Returns for eight asset classes, including cash, stocks, and bonds
of various durations are then determined accordingly. One of the characteristics of this model is
that it includes a strategy file which contains the user inputs that describe the company’s business
plan. Four types of inputs are needed to run the model, including the asset details, insurance
business details, catastrophic loss details, and simulation control inputs. The asset details include
company capital structure, asset strategies, and dividend policy; the insurance business details
include premiums, losses, outstanding claims amounts, and loss frequency and severity
distributions; the catastrophic loss details include distributions for frequency and severity of
catastrophic losses, and payment patterns; the simulation control inputs include number of
simulation, seed number, and scenarios to run (Christofides and Smith, 2001).

The common features of the above-mentioned models can be summarized as follows:

 A stochastic model is used for simulating economic variables and capital market conditions.
 These projected economic variables and capital market conditions are then used for
determining asset returns and projecting future liabilities.
 Interest rate and inflation rate are two most common driving factors.
 The main output of modeling is projected cash flows and financial statements.

4.3 Capital allocation

Capital allocation is the process of distribution through which an insurer determines how much
capital is needed for a particular operating division or a line of business within a company. One of
the main purposes of allocating capital is to properly quantify capital charges for each division or
product line. The issue on capital allocation is frequently related to capital adequacy.
Management, in general, decides on a total required capital amount before allocating to the
divisions. Capital adequacy has been a popular issue in the actuarial field for many years. Various
papers concerning this issue have been published. For instance, both Mango and Mulvey (2000)
and Philbrick and Painter (2001) first explored the adequacy issue, then move to the allocation
issue. In these papers, a number of risk measures and their merits and weaknesses were also
discussed. Examples included probability of ruin, EPD, tail conditional expectation, expected
default loss rate on surplus, and standard deviation of surplus.
34
 The quantity of the capital required by a division generally increases with the risk of the
business written by that division. Several approaches to allocating capital among divisions have
been proposed (Mango, 1998; Philbrick, 1999; Mango and Mulvey, 2000; Philbrick and Painter
2001). Suppose an insurer only writes three distinct lines, X, Y and Z. The first approach is that
we calculate the total required capital of the company and the capital needed if the insurer only
wrote lines X, Y or Z. Suppose the capital amounts required are 10,000, 4,000, 5,000 and 6,000.
We multiply 10,000 by 4/15, 5/15, 6/15 and obtain the capital amounts for lines X, Y and Z. These
are 2,667, 3,333 and 4,000 respectively. The second allocation approach is referred to as the
marginal capital method. We need to calculate the marginal required capital contribution of lines
X, Y, and Z. For instance, the required capital amounts for a company writing lines XYZ, YZ, XZ,
XY are 10,000, 9,000, 8,000, and 7,000. In this case, the marginal required capital contribution of
lines X, Y, and Z are 1,000, 2,000, and 3,000. We then rescale these three numbers to make their
sum equal 10,000 by multiplying 10/6, and obtain the capital amounts for lines X, Y and Z are
1,667, 3,333 and 5,000 respectively. The final approach is to apply the Shapley value of the game
theory field to the allocation question and is referred to as the Shapley value allocation method.
The required capitals for X, Y, and Z are based on their Shapley values. For demonstration
purpose, we calculate the Shapley value for line X. The Shapley value is the average of the
marginal variances over all the possible permutations. It is worth noting that the Shapley value is
order dependent. In the case of an insurer writing lines X, Y, and Z, the possible permutations are
XYZ, YXZ, YZX, ZXY, ZYX, and XZY. If the risk measure chosen was variance of a particular
variable, the Shapley value for line X is the sum of Var(X), Cov(X, Y), and Cov(X, Z). Interested
readers are referred to Philbrick (1999) for the Excel spreadsheet examples.

Philbrick and Painter (2001) applied DFA to a hypothetical insurer, DFAIC, to deal with the
capital allocation problem. The model employed was the FIRMTM Asset Model. In this article,
they divided the company’s business into five lines including Workers Compensation, Auto
Liability, Property, General Liability, and CMP liability coverage. First, the model was
parameterized to reflect the behavior of the variables in question. Philbrick and Painter (2001)
employed the Shapley method in a slightly different way. They used the marginal volatility
associated with a particular line instead of comparing the insurer with and without of the
individual line being added to all combinations of the remaining lines. They completed the section
with a discussion of capital allocation results obtained from using the Shapley value and marginal
last-in allocation methods.

5. Current guidance to actuaries conducting DFA/DST/FCR in the UK, the USA and
Canada

There has been a recent trend among actuarial professional bodies in developed countries
35
towards providing guidelines or standard of practice to help appointed actuaries evaluate the
financial condition of their individual companies using DFA or its variants. It is anticipated that
the insurance regulatory authorities in other countries will follow the trend and require insurers to
conduct DFA in the future. The purpose of this section is to review the current guidance on
DFA/DST/FCR in the UK, the USA and Canada.

5.1. UK

In March 1996, the Life Board of the Faculty and Institute of Actuaries introduced Guidance
Note 2 (GN2) on FCR as Recommended Practice for appointed actuaries responsible for long-
term insurance business in accordance with the Insurance Companies Act 1982 and the Friendly
Societies Act 1992. After the Financial Services and Markets Act 2000 took effect, GN2 is
subsequently revised to a small extent. In Section three of GN2, how to employ deterministic and
stochastic techniques to obtain the firm’s important information required by the FCR is discussed.
It is worth noting that GN2 is not statutory practice for life insurers and that there is no
counterpart in non-life insurance.

GN2 divides DST techniques into three categories: sensitivity testing, scenario testing, and
stochastic modeling. The actuary may employ these techniques at their discretion to project an
insurer’s solvency position into the future under different assumptions with a view to assessing its
financial strength and identify the main risk factors affecting the company.

GN2 does not specify any scenarios that must be examined. As a whole, they are at the
appointed actuary’s discretion. However, GN2 classifies the assumptions into two categories:
assumptions that all insurers are encouraged to test and assumptions that may be of great
importance to some firms. These assumptions can be found in Table 3.

GN2 as guidance on FCR for life business suggests that projection period of five years in DST
is suitable in most cases. However, a longer forecast period can be used if the actuary considers it
necessary. The DST results will be presented in the FCR, including the aim of the report, the
development and business of the firm and the environment where the company is expected to
operate, the techniques and scenarios employed, comments on the implications of DST results,
etc. For the UK case, there is currently no relevant guidance on DFA in non-life insurance.

5.2. USA

In 1996, the Society of Actuaries (SOA) published “Dynamic Financial Condition Analysis
(DFCA) Handbook” designed as a resource to assist actuaries in assessing the financial condition
of life and health insurers. However, it should be noted that this handbook has not yet been
developed as a standard of practice (Society of Actuaries, 1996).
36
 This handbook suggests that the actuary find out the sensitivity of surplus to likely changes of
internal and external environment by testing the most likely scenario and other plausible
scenarios. The actuary determines the scenarios that may be tested after identifying the risks faced
by the insurer. No specific scenarios are recommended by the handbook. A five-year forecast
period is considered appropriate for general purposes. The handbook points out that DFCA is an
ongoing, not necessarily annual, exercise. What items a DFCA report should contain are also
illustrated.

In 1995, the CAS first developed “Dynamic Financial Analysis Handbook” for property and
casualty insurance companies. In 2000, the DFA committee of the CAS published “Dynamic
Financial Analysis Research Handbook” by combining the original handbook with other newly
produced papers on DFA. The purpose of this updated handbook is to provide actuaries with
guidance and a list of considerations when conducting DFA. The CAS specifically points out that
the selection of a scenario depends on the environment in which the insurance company operates.
It is noted that the handbook does not prescribe reporting requirements as regards DFA. The
format of the relevant report is at the discretion of the actuary conducting DFA. In addition, the
handbook does not prescribe a specific projection period, either (Szkoda et al., 1995).

5.3. Canada

From January 1 1999, all appointed actuaries of insurance firms operating in Canada have been
required by the Superintendent of Financial Institutions to prepare financial condition reports in
accordance with the Standard of Practice on Dynamic Capital Adequacy Testing issued by the
Canadian Institute of Actuaries (1998). This Dynamic Capital Adequacy Testing (DCAT) standard
covers not only life insurers and fraternal benefits societies, but also property-casualty insurers.
Before the DCAT standard was issued, the DST standard only covered life companies and
fraternal benefits societies.

DCAT involves testing different scenarios and investigating their adverse effect on the firm’s
financial well-being and capital adequacy. According to the DCAT standard, a scenario refers to as
a set of consistent assumptions relating to not only underwriting but also investment operations.
The actuary should select and test, at least, the three most adverse, but possible, scenarios except
the base scenario. The unfavorable scenarios suggested by the Canadian Institute of Actuaries for
life and non-life firms can be found in Table 3.

The DCAT standard also suggests that the projection period of five years is appropriate for a
life insurance firm and that of two years for a non-life company. An annual investigation into the
operations of the insurer during the past few years is required. However, an interim review can be
conducted if necessary. A sample report as well as the manner of application of the DCAT
37
standard is outlined in the Canadian Institute of Actuaries (1999).

Overall, most of the risk categories listed by the professional bodies are not concerned with
investment conditions. Nonetheless, at present investment-related risk is one of the most
significant risks faced by insurance firms. Over the recent years insurers have been adversely
affected by the falling price of shares and interest rates. For instance, the UK life insurance
industry has been badly hurt because it invests in equities a very large proportion of the assets it
manages and accordingly many life insurers cut bonus payments to policyholders and raise exit
penalties. This highlights the importance of testing the financial health of the company under a
variety of investment-related scenarios. As indicated in the DCAT standard, the actuary should
consider threats to capital adequacy under plausible adverse scenarios that include but are not
limited to the scenarios listed above. Therefore, the increased uncertainty over future investment
returns requires that several plausible adverse scenarios concerning investment conditions be
tested such as sharp falls in equity prices.

The actuary should constantly review relevant guidance notes or standards of practice issued by
regulators or actuarial professional bodies in order to comply with the relevant regulations. For
comparison purposes, Table 3 summarizes the current guidance to actuaries carrying out
DFA/DST/FCR in the above-mentioned countries.

38
Table 1. Current DFA/DST/FCR guidance in the UK, USA, and Canada
UK (Life) USA (Life, Non-life) Canada (Life, Non-life)
Assumption (scenario) •GN2 prescribes no assumption that ․No specific scenarios are Life:
must be tested and appointed prescribed. •Mortality
actuaries are left with discretion. •Morbidity
•Assumptions that all insurers are •Persistency
encouraged to test: future •Cash flow mismatch
investment conditions, levels of new •Deterioration of asset values
business, expenses and persistency. •New business
•Assumptions that may be of •Expense
importance to some firms: allocation •Reinsurance
of profits, mortality and morbidity, •Government and political action
taxation, exercising of options by •Off balance sheet
policyholders, exercising of options
by the insurer, effects of asset- Non-life:
defaults, unit pricing bases, and •Frequency and severity
reinsurers’ default risk. •Pricing
•Misestimation of policy liabilities
•Inflation
•Interest rate
•Premium volume
•Expense
•Reinsurance
•Deterioration of asset values
•Government and political action
•Off balance sheet
Forecast period 5 years Life: 5 years Life: 5 years
Non-life: none Non-life: 2 years
Projection frequency Annual None Annual
FCR • Potential adverse developments •Background, purpose and • Executive summary
• Possible remedial actions objectives • Introduction to DCAT
• Especially significant risks •Scope • Capital adequacy measurement
•Limitations and reliance • Base scenario

39
•scenarios tested • Adverse scenario
•Summary report by line of • Analysis of risks by
business • Line of business
•Detailed analysis by line of • Conclusions and
business recommendations
•Tentative findings and • Appendices
observations at total company
level
•Next steps

40
5.4. Empirical survey of the practices of dynamic financial analysis/ dynamic solvency testing

It is worthwhile to note that the actuary should investigate the current DFA practices with a
view to understanding what level of detail and specific components are included in DFA models
by practitioners. Although DFA/DST related techniques have been used in the insurance industry
for some time, there exist relatively few surveys of current DFA/DST practices. The results of
these surveys which have been undertaken are summarized in Table 4.

In order to draft guidance for appointed actuaries on FCR, the Dynamic Solvency Testing
Working Party of the Faculty and Institute of Actuaries posted a questionnaire to appointed
actuaries in 153 UK life offices in 1994 investigating the practices of DST in the life insurance
industry. In the initial analysis of 29 with-profit offices, the Dynamic Solvency Testing Working
Party (1994) found that 34 per cent of the survey respondents reported only carrying out a
sensitivity analysis, and 21 per cent scenario testing. Thirty eight per cent of the respondents
conducted both a sensitivity analysis and scenario testing, whereas seven per cent did not do any
of them. Only 14 per cent of the respondents carried out stochastic projections. Inflation,
investment returns and investment yields are the variables which were usually treated
stochastically. More than two thirds of the respondents indicated that their offices used model
points to represent the liability structure, while just over one third reported using most of the in
force policies to forecast future liabilities. The most common forecast period was five years. The
DST results were normally presented to the board of directors by the appointed actuary.

As mentioned previously, GN2 as Recommend Practice on DST and FCR took effect in March
1996. At the end of 1996, Muir and Sarjant sent a questionnaire to appointed actuaries in UK life
offices and friendly societies covering the practical issues associated with DST and FCR. The
total number of questionnaires for this survey was not reported in Muir and Sarjant (1997). Forty-
nine replies to this survey were received, including those from 31 offices writing with-profit
business. Muir and Sarjant (1997) reported that about 50 per cent of respondents carried out
scenario testing as well as sensitivity. They also found that offices writing with-profits business
are usually able to conduct more complicated asset modeling than offices which do not write any
with-profit business. For instance, 19 per cent of offices writing with-profits business had the
capacity to model individual assets, whereas only eight per cent of office which do not write any
with-profit business had the same capacity. Regarding liability modeling, roughly a quarter of the
respondents used individual policies to project liabilities rather than model points. In addition,
according to the survey, the most common projection period was 20 years, but generally only the
first five-year results were presented to the Board.

Because the survey by the Dynamic Solvency Testing Working Party (1994) was administered
before GN2 took effect, whereas the survey by Muir and Sarjant (1997) was administered less
than one year after, these two surveys should have been comparable to some extent. However, it is
41
doubtful whether the findings of the surveys may be validly compared. Most of the questions
asked in the questionnaire by the Dynamic Solvency Testing Working Party are different from
those by Muir and Sarjant. Very few of the questions in these two surveys are similar. Even if they
are, the questions in the survey of Muir and Sarjant are generally more specific and detailed than
those in the survey of the Dynamic Solvency Testing Working Party. Moreover, the findings of the
Dynamic Solvency Testing Working Party are solely derived from the analysis of with-profit
offices while Muir and Sarjant included both life offices and friendly societies in their survey
population, in spite of whether they wrote with-profit business or not. In general, offices writing
with-profit business are likely to have more complicated techniques of solvency testing, asset and
liability modeling than offices which do not write any with-profit business. Further, there was no
test for non-response bias conducted for these two surveys. Therefore, it is unknown that to which
extent the results of these two surveys were affected by this problem.

Oakden, Friedland and Périgny (2001) invited 36 Canadian property-casualty insurance and
reinsurance companies to participate in a study of Appointed Actuaries’ approach to DCAT
analysis and reporting. Twenty-two companies responded the invitation and were interviewed.
Oakden, Friedland and Périgny (2001) reported that Appointed Actuaries were significantly
involved in determining input for the base scenario. Scenarios considered significant and included
in more than one-half of the DCAT reports of the companies surveyed include frequency and
severity of loss, understatement of unpaid claim liability, single catastrophic loss, increase in
inflation, increase in interest rates, and deterioration in asset values. On average, more than six
scenarios were used. In addition, the length of projection period of DCAT was roughly in line with
that of business plan. The DCAT projection period of more than 90 per cent of the companies was
less than two years, while the projection period of business plan of more than 90 per cent of the
companies was less than three years.

In May 2002, Shiu (2004b) administered two DFA/DST/FCR surveys for non-life and life
insurance companies authorized to carry on insurance business in the UK. The questionnaire
confirmed whether DFA/DST were being used, how they were used, which risk factors were
incorporated, and how assets and liabilities were modelled within the individual organisations.
Moreover, the questionnaire would also seek to confirm whether FCR was being produced,
whether FCR was available to third parties, whether it is necessary to introduce a Guidance Note
on FCR (for non-life insurers), and to what extent the Guidance Note 2 is acceptable (for life
insurers). The numbers of survey populations were 131 and 92 for the non-life and life sectors
respectively. Based on the survey results, Shiu reported that there is a significant difference in the
DFA /DST/FCR practices between non-life and life insurers, and between with-profits and non-
profit offices. A comparison of findings from surveys that focus on DFA practices is shown in
Table 4.

42
 Table 4 Empirical surveys of practices of dynamic financial analysis
Professional body/ Methodology Major findings
Researcher(s)
Dynamic Solvency Postal survey: The following findings are based on the results of the initial analysis of 29 with-profit offices:
Testing Working appointed • 72% of respondents carried out sensitivity analysis; 59% scenario testing; 38% both; 7% neither.
Party, the Faculty
actuaries in 153 • 14% of respondents carried out stochastic projections; inflation, investment returns, investment
and Institute of
Actuaries (1994) UK life offices yields are usually treated stochastically.
• 69% used model points to model liabilities; 35% used the whole policy file.
• The most common projection period was five years.
• Results of DST exercises were normally presented to the Board by the appointed actuary.
Muir and Sarjant Postal survey: • Approximately 50% of respondents carried out scenario testing as well as sensitivity testing.
(1997) Appointed • 19% of offices writing with-profit business had the capability to model individual assets; 8% of
Actuaries in UK offices which do not writing with-profit business had the same capacity.
life offices and • Only 26% used individual policy to project liabilities rather than model points.
friendly • The most common projection period was 20 years, but generally only the first 5-year results were
societies presented to the Board.
Oakden, Friedland Interviews: • Scenarios considered significant and included in more than one-half of the DCAT reports of the
and Périgny (2001) appointed companies surveyed include frequency and severity, understatement of unpaid claim liability, single
catastrophic loss, increase in inflation, increase in interest rate, and deterioration in asset values.
actuaries in 22
• On average, more than 6 scenarios were used in the companies surveyed.
Canadian
• DCAT projection period: 50% (1 year); 41% (2 years); 9% (3 years).
property-
 Business plan projection period: 37% (1 year); 27% (2 years); 27% (3 years); 9% (5 years).
casualty
insurance and
reinsurance
companies
Shiu (2004b) Two postal Non-life survey:
43
surveys: 131 non-  The use of DFA techniques in the industry was limited.
life and 92 life  Scenario testing was the most frequently used technique.
insurers in the UK.
 The most common DFA application was evaluation of reinsurance programmes.
 Less than ten scenarios were run regularly.
 Inflation was the most frequently modeled economic variable.
 The capability of asset modeling of general insurers was restricted. 7. The most common method of
liability modeling was to use all in force policies in aggregate.
 The most common projection periods in DFA and BP were three years.
 The main reason for not using DFA techniques and producing FCR was lack of need.
 Views on whether a Guidance Note on FCR specifically for general insurers should be introduced
differed.

Life survey:
 Scenario testing was the most commonly used DST techniques.
 Most life insurers regularly run less than ten scenarios in scenario testing.
 Most life insurers reported using a five-year forecast period in DST.
 The two most commonly seen difficulties are: difficulties in communicating complex issues to non-
specialists, and how to present extremely adverse scenarios without causing undue concern.
 Nearly all life firms use FCR.
 Guidance Note 2 is generally considered acceptable.
 Compared with the results reported by previous studies, the use of DST techniques is now more
common in life firms.
 There is a significant difference in DST/FCR practices between with-profits and non-profit insurers.

44
6. An illustrated DFA example

The objective of this section is to illustrate an example to show how a DFA model works in
general. It was decided to apply DynaMo 3 to solvency testing and asset allocation in non-life
business. DynaMo 3 is a public access Excel based model developed by the actuarial consulting
firm Miller, Herbers, Lehmann & Associates, Inc2. As alluded to in Section 3, this model consists
of five primary modules, including the interest rate generator module, the investment module, the
underwriting module, the reinsurance module, and the catastrophe generator module. The
rationale of including these modules in the model is as follows. First, both investment and
underwriting operations are a non-life insurer’s operations. Second, non-life companies generally
focus more on the underwriting side. The risk exposure to catastrophe is one of their main
concerns. They usually rely heavily on reinsurers due to the risky nature of their business. The
other features of this model can be found in D’Arcy et al. (1997, 1998), and Walling et al. (1999).

6.1. The change made to DynaMo 3 and key modeling assumptions

It is worth noting that the default parameters which are included in DynaMo 3 are mainly
chosen to be illustrative. The actuaries of individual firms should select their own parameters to
reflect the actual situations in their companies. For instance, historical data is used to calibrate the
parameters required in the interest rate generator of DynaMo. The reader may refer to Chan et al.
(1992) and Nowman (1997) for approaches to calibrating models. According to D’Arcy et al.
(1998), the data is from an actual US non-life insurer. The initial set of inputs of DynaMo remains
unchanged except where otherwise stated.

For simplicity, the invested asset mix comprises two categories of investments: equities and
bonds. Specifically, the company only invests in ordinary shares and long-term government bonds
with a maturity of more than 20 years. It is assumed that the hypothetical company invests
$96,650 thousand in these two assets (bond investment: $64,433 thousand; equity investment:
$32,217 thousand). The starting point in time of the DFA exercise is at the end of the year 2005
and the model is used to project five years into the future. The company’s balance sheet as of 31
December 2005 is shown in Table 5.

2
DynaMo 3 is available to be downloaded from the Pinnacle website: http://www.pinnacleactuarialresources.com
45
Table 5
The balance sheet of the hypothetical company as of 31 December 2005 ($000)
Assets Liabilities

Invested assets Technical reserves:

Long-term government bonds 64,433 Outstanding claims 34,402

Unearned premium 25,500
Ordinary shares 32,217
Other liabilities 2,598
Other assets

Agents balances 5,200

Reinsurance recoverables 650 Surplus 40,000

Total 102,500 Total 102,500

Expected surplus, also known as shareholders’ funds or solvency margin, is chosen as the
performance (return) measure in this analysis. Within an insurance context surplus not only
represents the level of shareholders’ wealth but also provides a cushion for absorbing unexpected
losses. In non-life business, the level of surplus is of particular importance because the
permissible level of premiums written is generally expressed as a multiple of surplus such as the
NAIC Property/Casualty IRIS Ratio 1 (Gross premiums written to policyholders’ surplus) and
Ratio 2 (Net premiums written to policyholders’ surplus).

Two risk measures are used in this analysis. The first is the standard deviation of surplus. This
indicator measures the dispersion of the outcomes, but does not evaluate the severity (Lowe and
Stanard, 1996). The second risk measure is the probability that the surplus will fall below the
beginning statutory surplus over the next five years. The primary difference between the two is
that the former focuses on the dispersion of the outcomes, whereas the latter the probability and
severity of adverse outcomes.

Some of the other key assumptions used in the model include the following:

• Starting point: year-end 2005.

• Forecast period: five years (2006-2010).

46
• Only two lines of business: Homeowners multiple peril (HMP) and workers’ compensation

(WC).

• Trade-off ratio = (Percentage change in return measure) / (Percentage change in risk measure).

• Parameters for the interest rate model: a = 0.25, m = 6 per cent, σ = 5 per cent, r2005 = 4.91 per

cent.

• Parameters for the inflation rate model: a = 0.015, b = 0.75, s = 0.7 per cent.

• No impact of inflation on the number and severity of claims for HMP, and on the number of

claims for WC. That is, inflation only has impact on the severity of claims for WC.

• Market risk premium in: 8.5 per cent.

• New business, first renewal business, and second and subsequent renewal business are

modeled separately in order to reflect the aging phenomenon. The frequencies of claims are

assumed to be different, new business being the highest, and second and subsequent renewal

business being the lowest.

• The company equally underwrites business in two places, where the frequency and severity of

claims follow Poisson and Lognormal distributions respectively. Place A: frequency ~ Poisson

(0.6667), severity ~ Lognormal (12.76965, 1.516347); Place B: frequency ~ Poisson (1.0556),

severity ~ Lognormal (11.28113, 0.584328).

• Non-catastrophe losses are projected using the loss development method, which is based on

the assumption that claims move from unreported to reported-and-unpaid to paid in a pattern

which is sufficiently consistent that past experience can be used to predict future development

(Casualty Actuarial Society, 1990).

• Written premiums in 2005: HMP ($58 million) and WC ($21.8 million).

• Expenses: HMP (32 per cent) and WC (27 per cent) of written premium.

• Two reinsurance programs: Stop loss reinsurance (cost: 1 per cent of premium; maximum loss

& allocated loss adjustment expense ratio: 76 per cent; maximum ceded amount: $5 million)

and catastrophe reinsurance (retention: $ million; size: $25 million; cost: 2 per cent of

premium).

47
• Underwriting cycles: phase 1 = mature hard, phase 2 = immature soft, phase 3 = mature soft;

phase 4: immature hard. The transition probabilities between underwriting cycle phases are

shown in Table 6.

Table 6

Transition probabilities between underwriting cycle phases

Transition probability
Phase 1 2 3 4
1 50% 40% 5% 5%
2 5% 50% 40% 5%
3 5% 5% 50% 40%
4 40% 5% 5% 50%

Both fixed and dynamic asset allocation strategies are considered in this paper. It should be noted
that these strategies are only concerned with the invested assets, i.e. equities and bonds and do not
apply to other assets including agents balances and reinsurance recoverables. The fixed strategies
used in the paper are similar to those of Macdonald (1995), and Bohra and Weist (2001). Table 7
defines the five fixed strategies, ranging from 100 per cent to 0 per cent in equities and the
remaining in bonds.

Table 7

Fixed asset allocation strategy

Strategy Percentage of invested assets invested in equities
Fixed asset allocation strategy 1 (FAAS 1) 100%
Fixed asset allocation strategy 2 (FAAS 2) 75%
Fixed asset allocation strategy 3 (FAAS 3) 50%
Fixed asset allocation strategy 4 (FAAS 4) 25%
Fixed asset allocation strategy 5 (FAAS 5) 0%
Note: Invested assets include bond and equity investments.

The fixed asset allocation strategies were denoted FAAS 1 (100 per cent in equities) to FAAS 5
(0 per cent in equities), whereas the corresponding dynamic asset allocation strategies DAAS 1
(100 per cent in equities) to DAAS 5 (0 per cent in equities). The dynamic asset allocation
strategies used are variants of those of Macdonald (1995) and Hardy (1996). Take DAAS 1 as an
example. As stated displayed in Table 5, the A/L ratio at the commencement of modeling is 1.64.
From the first year to be modeled, the company, by default, invests 100 per cent of its funds in
equities. However, if the A/L ratio falls below 1.54, the company progressively switches out of
equities and into bonds, with 100 per cent bond investment if the A/L ratio falls below 1.44. As
long as the A/L ratio is greater than 1.54, the company will return to 100 per cent equity
investment. It should be noted that even if the A/L ratio falls below 1.54, DAAS 5 is the same as
48
FAAS 5 because all the invested assets are bonds.

There are two main reasons behind this dynamic asset switching algorithm. First, although
equity returns are usually higher than bond returns, but the risk associated with the former is also
higher than that associated with the latter. A high proportion of equity investments may increase
the risk of insolvency. The company has to strike a balance between solvency and high returns. It
seems advisable not to take extra risk and to be relatively conservative when the A/L ratio is low.
Second, most importantly, the valuation of equities does not allow for possible capital gain, but
allows for dividend yield, which is usually lower than bond yield. Therefore, in order to
demonstrate solvency a company usually switches out of equities and into bonds.

6.2. Model output

This Subsection summarizes the results, which are intentionally presented in terms that are
familiar to actuaries and managers of non-life insurers. Simulation is performed 1,000 times. In
order to reproduce the results and make them comparable, the starting random seed value has to
be set at a fixed and randomly selected number. A number of 44 is chosen from a random number
table to serve this purpose.

Table 8 shows the numbers of insolvencies (A/L<1) out of 1,000 simulations for each of the ten
strategies defined above. The numbers of insolvencies are all relatively small. This is probably
because the company whose A/L at the commencement of modeling is 1.64 is financially strong.
In addition, it seems that in the particular case in question changes in the proportion of equity
investments make relatively little difference in the number of insolvencies. There is a very small
probability that the company will become insolvent in the period concerned, possibly due in part
to the short forecast period and the firm’s financial soundness.

Table 8
Number of insolvency out of simulation 1,000 times
Strategy Number of insolvency Strategy Number of insolvency
FAAS 1 5 DAAS 1 5
FAAS 2 3 DAAS 2 3
FAAS 3 4 DAAS 3 4
FAAS 4 4 DAAS 4 4
FAAS 5 5 DAAS 5 5

As indicated by Ryan (1984), management is concerned not only about solvency, but also about
the overall stability of the results. As such, it would be interesting to investigate the distributions
of surplus in more detail. Two observations are obtained from the results. First, the distributions of
the ending surplus of the fixed asset allocation strategies are very similar to those of their dynamic

49
counterparts. Due to the similarity between the fixed and the dynamic results, the following
discussion will be only focused on the fixed strategies for simplicity reasons. Second, for the five
fixed strategies FAAS 1 through FAAS 5, the probabilities that the surplus will fall below the
beginning statutory surplus ($40 million) over a five-year period are 14.3 per cent, 13.4 per cent,
12.0 per cent, 10.4 per cent, and 21.3 per cent respectively. The lowest equity strategy (FAAS 5) is
demonstrated to be much riskier than the highest equity strategy (FAAS 1). As shown in Figures 6
to 10, the simulated distributions of ending surplus and cumulative percentages of these strategies
are fairly different. The above five figures again confirm that equity investments result in a
distribution of ending surplus with a higher mean and standard deviation. It is noted that the
simulated ending surplus of FAAS 1 is relatively normally distributed, whereas that of FAAS 5 is
highly skewed. In addition, in the particular case in question, it can be seen from Figure 10 that
most of the simulated ending surpluses are relatively small, although the distribution has a
relatively low variance.

Table 9 shows the trade-off ratios of these strategies. The trade-off ratio is defined as percentage
change in performance (return) measure divided by percentage change in risk measure. All the
trade-off ratios are positive, which reflects efficient portfolios. That is, return and risk change in
the same direction. Figure 9.1 is a mean-standard deviation diagram for the five fixed asset
allocation strategies. FAAS 1 with 100 per cent in equities has the highest risk and return, whereas
FAAS 5 with 0 per cent lowest. This confirms that equity achieves much higher mean returns but
with a much higher standard deviation of returns.

Table 9
Trade-off ratios of fixed asset allocation strategies (2005-2009)
Mean Surplus Standard deviation of surplus Trade-off ratio
FAAS 1 76,918,000 35,348,000 FAAS 1,2 = 39.26%
FAAS 2 69,020,000 27,371,000 FAAS 2,3 = 33.36%
FAAS 3 60,814,000 19,489,000 FAAS 3,4 = 27.80%
FAAS 4 52,329,000 12,309,000 FAAS 4,5 = 48.18%
FAAS 5 43,578,000 8,688,000 N/A

50
 90000
Return- expected surplus after five years

80000

70000

60000

50000

40000

30000

20000

10000

0
0 5000 10000 15000 20000 25000 30000 35000 40000

Risk- standard deviaiton of surplus after five years

Fig. 10. Risk-return trade off for five asset allocations ($000)

6.3. Robustness testing

Burkett, Mclntyre, and Sonlin (2001) point out that sensitivity testing is required in a DFA
exercise to ensure that the results obtained are robust and are not the product of a particular set of
assumptions. By performing sensitivity analyses, the modeler can examine how the original
results react to the changes in the initial assumptions. Specifically, sensitivity testing helps
actuaries identify and evaluate key levers for management strategy decision making (Sweeney et
al., 1998). Thus, with a view to testing the robustness of the results of the analysis, sensitivity tests
of key input factors are carried out such as interest rates, new exposure growth rates, renewal
ratios, and loss and expense ratios. As stated previously, sensitivity testing involves changing a
variable at a time. However, there are too many key input factors which can be varied. For
illustrative purposes, it was decided to vary the assumptions regarding the interest rate. The reason
why interest rate was chosen is because it is the driving factor in the model. Both the long-term
mean interest rate and the reversion parameter are varied to test whether their changes have
significant impact on ending surplus. Since the results of varying long-term mean interest rate and
reversion parameter are similar, only the former is reported and discussed here.

The long-term mean interest rate in the base case is 6 per cent. Two alternative long-term mean
interest rates (5 per cent and 7 per cent) are selected to assess their impact on ending surplus.
Table 10 shows the mean, standard deviation, minimum, maximum and first and ninety-ninth
percentile values from 1,000 simulations for each of the five fixed strategies. The sensitivities to
the changes in the value for long-term mean interest rate are different. On average, every
percentage point change in the long-term mean interest rate results in the increase in ending
surplus by $356,000, $201,000, $100,000, $48,000, and $44,000 for strategies FAAS 1 through
FAAS 5. The above changes in surplus only account for less than 1 per cent of beginning surplus.
This is shown graphically in Fig. 11. These results suggest that the conclusions obtained in this
51
chapter are robust in terms of the sensitivity test of long-term mean interest rate.

Table 10
Sensitivity test of long-term mean interest rate (unit: $000)
Long-term mean interest rate m = 5%
Mean S.D. Minimum Maximum 1% 99%
FAAS 1 76,571 35,026 –39,396 198,551 12,495 166,420
FAAS 2 68,784 27,088 –48,898 162,480 17,871 136,969
FAAS 3 60,667 19,234 –58,921 125,388 22,351 106,980
FAAS 4 52,259 12,056 –69,292 89,046 22,451 77,798
FAAS 5 43,573 8,434 –80,018 61,109 16,974 54,706

Long-term mean interest rate m = 6% (Base case)

Mean S.D. Minimum Maximum 1% 99%
FAAS 1 76,918 35,348 –39,752 200,656 11,286 167,580
FAAS 2 69,020 27,371 –49,422 164,188 16,978 138,171
FAAS 3 60,814 19,489 –59,582 126,701 22,159 107,954
FAAS 4 52,329 12,309 –70,073 90,184 21,955 77,707
FAAS 5 43,578 8,688 –80,900 60,101 15,627 54,916

Long-term mean interest rate m = 7%

Mean S.D. Minimum Maximum 1% 99%
FAAS 1 77,283 35,660 –40,183 202,819 10,711 169,127
FAAS 2 69,186 27,556 –50,175 165,347 15,882 139,179
FAAS 3 60,866 19,627 –60,532 127,199 20,745 108,411
FAAS 4 52,354 12,496 –71,104 90,252 20,684 78,461
FAAS 5 43,660 8,931 –81,894 60,724 14,412 54,987

350 120.00%

300
100.00%
250 12 0 .0 0 %

250
10 0 .0 0 % 80.00%
200
200
Frequency

8 0 .0 0 % F re quenc y
15 0 60.00%
F re q ue n c y C umulat iv e %
150
Frequency

6 0 .0 0 % C u m u l a t iv e %
10 0 40.00%
4 0 .0 0 % 100

50
2 0 .0 0 % 20.00%
50

0 .0 0 %
0 2 0 0 0 0 4 0 0 0 0 6 0 0 0 0 8 0 0 0 0 10 0 0 0 012 0 0 0 014 0 0 0 016 0 0 0 0 M o r e 0 .00%
0 20000 40000 60000 80000 100000 120000 140000 160000 M or e
E n d in g S u r p l u s ( u n it : 0 0 0 )
E n din g sur p lus (un it : 0 0 0 )

Fig. 6. Distribution of ending surplus (FAAS 1)Fig. 7. Distribution of ending surplus (FAAS 2)

52
 450 120.00%

400
100.00%
350

300 80.00%

Frequency 250
60.00%
Frequency

200 Cumulative %

150 40.00%

100
20.00%
50

0 .00%
0 20000 40000 60000 80000 100000 120000 140000 160000 More

Ending Surplus (unit: 000)

Fig. 8. Distribution of ending surplus (FAAS 3)

900 12 0 .0 0 %
8 00 12 0 .0 0%
800
7 00 10 0 .0 0 %
10 0 .0 0%
70 0
6 00
8 0.0 0 % 600 8 0 .0 0 %
5 00
50 0
Frequency

F req uency
Frequency

F r e qu e n c y
4 00 6 0.0 0 % 6 0 .0 0 %
C u m u la t iv e % 400 C umulative %

3 00
4 0.0 0 % 300 4 0 .0 0 %
2 00
200
2 0.0 0 % 2 0 .0 0 %
10 0 10 0

0 .0 0% 0 .0 0 %
0 20 0 0 0 4 00 00 60 00 0 8 00 0 0 10 0 0 0 0 120 00 0 14 0 0 0 0 16 0 00 0 M o r e 0 200 00 40 000 6000 0 8000 0 10000 0 1200 00 14 0000 160 000 M o r e

E n din g sur p lus ( un it ) E n din g sur p lus (un it : 0 0 0 )

Fig. 9. Distribution of ending surplus (FAAS 4) Fig. 10. Distribution of ending surplus (FAAS 5)

90,000

80,000
Ending surplus (unit: 000)

70,000

60,000
FAAS 1

50,000 FAAS 2
FAAS 3
40,000
FAAS 4

30,000 FAAS 5

20,000

10,000

0
5% 6% 7%

Long-term mean interest rate

Fig. 11. Ending surplus versus long-term mean interest rate

53
In this section, DynaMo 3 is utilized to illustrate how a stochastic DFA model works in non-life
business for the purposes of asset allocation and solvency testing. It is worth noting that the
illustrative evaluation of asset allocation strategies are carried out in the context of the whole
insurance operations which include both underwriting and investment activities, unlike traditional
asset allocation which only focuses on the investment side. Some conclusions from the results
obtained can be summarized as follows. First, the dynamic asset allocation strategies considered
appear to offer little advantage over the fixed asset allocation strategies in terms of the numbers of
insolvencies. Second, equity investments achieve much higher mean returns than do bond
investments, but also with a much higher standard deviation of returns. Third, the trade-off ratios
of the fixed strategies are all positive, reflecting efficient portfolios. That is, return and risk
change in the same direction. Fourth, it appears that the results are robust in terms of the
sensitivity tests conducted.

7. Conclusions

Over the years DFA has gradually emerged as one of the most important approaches to financial
modeling. Using DFA the actuary may yield much more information about the risk profile of an
insurer and the implications of management strategies chosen than statistic techniques. However,
it does not mean that DFA is perfect. Building an insurer-specific DFA model which represents the
real situation is by no means easy. It often involves model risk and parameter risk. The former
refers to the risk of employing an incorrect model and the latter the risk of using inappropriate
probability distributions and specifying the wrong values for the variables included in the model.
If stochastic modeling is employed, then the model is additionally exposed to stochastic risk
reflecting the stochastic element of the model. If a DFA model were to be used in determining
insurer solvency, regulators would need some means of ensuring that an insurer’s internal model,
if there was no standard models, reflects accurately the level of the risks that may jeopardize the
financial position of the insurer. Back-testing may provide a way of continually checking model
performance.

The purpose of this paper has been to provide an overview of DFA issues, including its
approaches, applications and practices in insurance. Thanks to the availability of modern
computers, the modeling of the insurance process using stochastic simulation gained effectiveness
and efficiency in recent times. Stochastic modeling is becoming of increased importance to
actuaries. As we have seen, DFA has been applied in many insurance areas including solvency
testing, asset allocation, and capital allocation. The overviews verify that DFA has been widely
implemented in insurance. Given the fact that there are many more problems that could be
resolved using DFA, we are likely to witness the emergence of a number of new applications
within insurance in the foreseeable future.

54
References

Adams, M., Buckle, M., 2003. The determinants of operational performance in the Bermuda
insurance market, Applied Financial Economics, 13, 133-143.
Ahlgrim, K. C., 2001. The effects of multifactor term structure models on the valuation of
insurance, Ph.D. dissertation, University of Illinois.
Almagro, Manuel, Sonlin, Stephen M., 1995. An Approach to Evaluating Asset Allocation
Strategies for Property/Casualty Insurance Companies, Casualty Actuarial Society Discussion
Paper Program, 55-79.
Australian Prudential Regulation Authority, 2000. Draft Prudential Standard on Capital Adequacy
for General Insurers (released on July 28, 2000). The document is available at
http://www.apra.gov.au/mediareleases
Beard, R. E., 1964. Some Thoughts on the Solvency of Insurance Companies, Algemene
Reinsurance Companies Jubilee Letter No. 1.
Berketi, Alexandra, 1998. Allowing for Insurance Companies’ Liabilities in Mean-Variance
Models, PhD Thesis, Herriot-Watt University, UK.
Bohra, R., Weist, T., 2001. Preliminary Due Diligence of DFA Insurance Company, Casualty
Actuarial Society Forum, 25-58.
Browne, M. J. , Hoyt, R. E., 1995. Economic and Market Predictors of Insolvencies in the
Property-Liability Insurance Industry, The Journal of Risk and Insurance, 62(2), 309-327.
Browne, M. J., Carson, J.M., Hoyt, R. E., 1999. Economic and Market Predictors of Insolvencies
in the Life-Health Insurance Industry, The Journal of Risk and Insurance, 66(4), 643-659.
Browne, M. J., Carson, J. M., Hoyt, R. E., 2001. Dynamic financial models of life insurers, North
American Actuarial Journal, 5, 11-26.
Burkett, J. C., McIntyre, T. S., Sonlin, S. M., 2001. DFA Insurance Company Case Study, Part I:
Reinsurance and Asset Allocation, Casualty Actuarial Society Forum, 59-98.
Canadian Institute of Actuaries, 1991. Standard of Practice on Dynamic Solvency Testing for Life
Insurance Companies (in effect January 1, 1992). This document is available at
http://www.actuaries.ca/publications/sop_e.html
Canadian Institute of Actuaries, 1993. Standard of Practice on Dynamic Solvency Testing for
Fraternal Benefits Society (in effect January 1, 1994). This document is available at
http://www.actuaries.ca/publications/sop_e.html
Canadian Institute of Actuaries, 1998. Standard of Practice on Dynamic Capital Adequacy Testing
(in effect January 1, 1999). This document is available at
http://www.actuaries.ca/publications/sop_e.html
Canadian Institute of Actuaries, 1999. Education note: standard of practice on dynamic capital
adequacy testing. This document is available at
http://www.actuaries.ca/publications/1999/9930e.pdf
Carińo et al., 1994. The Russel-Yasuda Kasai Model: An Asset Liability Model for a Japanese
55
 Insurance Company using Multi-Stage Stochastic Programming, Interfaces, Vol. 24, 29-49.
Carińo, D. R., Ziemba,W. T., 1998. Formulation of the Russel-Yasuda Kasai Financial Planning
Model, Operations Research, Vol. 46(4), 433-449.
Carińo, D. R., Myers, D. H., Ziemba,W. T., 1998. Concepts, Technical Issues, and Uses of the
Russel-Yasuda Kasai Financial Planning Model, Operations Research, Vol. 46(4), 450-462.
Casualty Actuarial Society, 1995. Dynamic financial analysis handbook. This document is
available at http://www.casact.org/coneduc/specsem/98dfa/dfahdbk.pdf
Casualty Actuarial Society, 2000. Dynamic financial analysis research book. This document is
available at http://www.casact.org/research/dfa/index.html
Chan, K. C., Karolyi, G. S., Longstaff, F. A., and Sanders, A. B., 1992, An Empirical comparison
of Alternative Models of the Short-Term Interest Rate, Journal of Finance, 47, 1209-1227.
Christofides, S., 2000. Discussion of Corporate Decisions in General Insurance: Beyond the
Frontier by M. P. Cumberworth, A. N. Hitchcox, W. D. McConnell and A. D. Smith, British
Actuarial Journal, Vol. 6(II), 259-296.
Christofides, S., Smith, A. D., 2001. DFA- The Value of Risk, Casualty Actuarial Society Forum,
153-194.
Consigli, Giorgio, 1998. Dynamic Stochastic Programming for Asset and Liability Management,
PhD Thesis, University of Essex, UK.
Correnti, S. and Sweeney, J. C., 1994. Asset-Liability Management and Asset Allocation for
Property and Casualty Companies- The Final Frontier, Proceedings of the 4th Actuarial
Approach for Financial Risks International Colloquium, Vol. 2, 907-918.
Correnti, S, Nealon, P. A. , Sonlin, S M., 1996. Decomposing Risk to Enhance ALM and Business
Decision Making for Insurance Companies, Proceedings of the 6th Actuarial Approach for
Financial Risks International Colloquium, 443-472.
Correnti, S., Sonlin, S.M., Isaac, D.B., 1998. Applying a DFA Model to Improve Strategic
Business Decisions, Casualty Actuarial Society Forum, 15-51.
Cumberworth, M. P., Hitchcox, A. N., McConnell, W. D., Smith, A. D., 2002: Corporate
Decisions in General Insurance: Beyond the Frontier, British Actuarial Journal, 6 (II) , 259-
296.
D’Arcy, S. P., Gorvett, R. W., 2004. The use of dynamic financial analysis to determine whether
an optimal growth rate exists for a property-liability insurer. The Journal of Risk and
Insurance. 71, 583-615.
D’Arcy, S. P., Gorvett, R. W., Herbers, J. A., Hettinger, T. E., Lehmann, S. G. , Miller ,M. J.,
1997. Building a Public Access PC-Based DFA Model, Casualty Actuarial Society Forum, 1-
40.
D’Arcy, S. P., Gorvett, R. W., Hettinger, T. E., Walling, R.J., 1998. Using the Public Access DFA
Model: A Case Study, Casualty Actuarial Society Forum, 53-118.
Daykin et al., 1984. The solvency of general insurance companies, Journal of the Institute of
Actuaries 114, 279-336.
Daykin et al., 1987. Assessing the Solvency and Financial Strength of a General Insurance
56
 Company, Journal of the Institute of Actuaries, Vol. 114, Part 2, 227-325.
Daykin, C. D. , Bernstein, G. D., 1985. A Simulation Model to Examine Questions of Solvency in
the Light of Asset and Run-off Risks. The paper was presented to ASTIN Colloquium in
Biarritz in October 1985.
Daykin et al., 1990. Managing Uncertainty in a General Insurance Company, Journal of the
Institute of Actuaries, Vol. 117, Part 2, 173-277.
Daykin, C. D., Pentikäinen, T., Pesonen, M., 1994. Practical Risk Theory for Actuaries, fist
edition (Chapmean & Hall; UK).
Dynamic Solvency Testing Working Party, 1994. Dynamic Solvency Testing, in the Proceedings
of Current Issues in Life Assurance, Seminar, 2 November 1994, the Faculty and Institute of
Actuaries.
Faculty and Institute of Actuaries, 1996. Guidance Note 2: Financial Condition Reports. This
document is available at http://www.actuaries.org.uk/map/GN02V1-0.pdf.
Feldblum, Sholom, 1992. European Approaches to Insurance Solvency, Casualty Actuarial
Society Discussion Paper Program, 511-552.
Geoghegan, T. J. et al., 1992. Report on the Wilkie stochastic investment model. Journal of
Institute of Actuaries. 119, 173-228.
Gorvett, R. W., 1998. Dynamic Financial Analysis of Property-Liability Insurance Companies,
PhD Thesis, University of Illinois at Urbana-Champaign.
Hardy, M. R., 1993. Stochastic Simulation in Life Office Solvency Assessment, Journal of
Institute of Actuaries, Vol. 120, 131-151
Hardy, M. R., 1996. Simulating the Relative Solvency of Life Insurers, British Actuarial Journal,
Vol. 2, No. IV, 1003-1019.
Huber, P. P., 1996. A review of Wilkie’s stochastic asset model. British Actuarial Journal. 3, 181-
210.
Kaufman, A. M., Ryan, T. A. 2000., Strategic Asset Allocation for Multi-Line Insurers Using
Dynamic Financial Analysis, Casualty Actuarial Society Forum, 1-20.
Kaufmann, R., Gadmer, A., Klett, R., 2001. Introduction to Dynamic Financial Analysis, Astin
Bulletin, Vol. 31, (1), 213-249.
Kitts, A., 1990. Comments on a model of retail price inflation. Journal of Institute of Actuaries.
117, 407-413.
Kitts, A., 1990. Comments on a model of retail price inflation. Journal of Institute of Actuaries.
117, 407-413.
Krouse, C. G., 1970. Portfolio Balancing Corporate Assets and Liabilities with Special
Application to Insurance Management, Journal of Financial and Quantitative Analysis, Vol. 5,
Issue 1, 77-104.
Larsen, M. R. et al., 2004. Presenting DRM Results to Decision Makers: A Summary Report,
Casualty Actuarial Society Forum, 318-331.
Laster, D., Thorlacius, A. E., 2000. Asset-liability Management for Insurers, Swiss Re, Sigma No.
6/2000, 1-36.
57
Lee, P. J. and Wilkie, A. D., 2000. A comparison of stochastic asset models. Proceedings of the
10th AFIR Colloquium, 407-445.
Limb, A. P., 1984. The Solvency of Life Assurance Companies. A Report of Working Party to the
Faculty of Actuaries.
Lowe, S. P., Stanard, J. N., 1996. An integrated dynamic financial analysis and decision support
system for a property catastrophe reinsurer, Casualty Actuarial Society Forum, 89-118.
Macdonald, A. S., 1995. A Stochastic Evaluation of Solvency Valuations for Life Offices, PhD
Thesis, Herriot-Watt University, UK.
Mango, D. F., 1998. An application of game theory: property catastrophe risk load. Proceeding of
Casualty Actuarial Society.
Mango, D. F., Mulvey, J. M., 2000. Capital Adequacy and Allocation Using Dynamic Financial
Analysis, Casualty Actuarial Society Forum, 55-76.
Muir, M., Sarjant, S., 1997. Dynamic Solvency Testing. This paper was presented to the Staple
Inn Actuarial Society on 4 March 1997.
Mulvey, J. M., Thorlacius, A. E., 1998. The Towers Perrin Global Capital market Scenario
Generation System, in World Wide Asset and Liability Modelling, edited by W. Ziemba and
Mulvey J., first edition (Cambridge University Press; UK)
Nowman, K. B., 1997. Gaussian Estimation of Single-Factor Continuous Time Models of the
Term Structure of Interest Rates, Journal of Finance, 5, 1695-1706.
Oakden, D. J., Friedland, J., Périgny, I. 2001. Dynamic Capital Adequacy Survey, Tillinghast-
Towers Perrin. This document is available at
http://www.actuaries.ca/meetings/aa/2001/I_Perigny.pdf, accessed 23 February 2003.
Ong, A. S. K., 1995. Asset Allocation Decision Models in Life Insurance, Ph.D. Thesis, City
University, UK.
Pentikäinen, Rantala, 1982. Solvency of Insurers and Equalization Reserves (Insurance Publishing
Company; Helsinki).
Pentikäinen, T., 1988. On the Solvency of Insurers, in Cummins, J. D. and R. Derrig (editors):
Classical Insurance Solvency Theory, U.S.A.: Kluwer Academic Publishers.
Pentikäinen et al., 1989. Insurance Solvency and Financial Strength (Insurance Publishing
Company; Helsinki).
Philbrick, S., 1999. Brainstorms: capital allocation. Actuarial Review. February.
Ross, M. D., 1991. Modelling a With-Profits Life Office, Journal of the Institute of Actuaries, Vol.
116, 691-715.
Philbrick, S. W., Painter, R. A., 2001. DFA Insurance Company Case Study, Part II: Capital
Adequacy and Capital Allocation, Casualty Actuarial Society Forum, 99-151.
Rambaruth, G., 2003. A comparison of Wilkie-type stochastic investment models, Master
dissertation, City University, UK.
Ross, M. D., McWhirter, M. R., 1991. The Impact on Solvency and Policy Results of the
Valuation Regulations Restrictions on Equity Yields. Unpublished paper.
Ryan, J. P., 1984. Application of Simulation Techniques to Solvency Testing for a Non-life Office,
58
 Transactions of the 22nd International Congress of Actuaries, Vol. 3, 269-277.
Ryan, et al., 2001. Financial Condition Assessment. This paper was presented to the Institute of
Actuaries on 26 March 2001.
Shiu, Y. M., 2004a. Determinants of United Kingdom general insurance company performance,
British Actuarial Journal 10, 1079-1110.
Shiu, Y. M., 2004b. Dynamic Financial Analysis in the insurance industry: a study of current
practices, determinants of company performance, and application of Dynamic Financial
Analysis. Unpublished PhD Thesis, the University of Edinburgh, UK.
Shiu, Y. M., 2005. The determinants of solvency in the United Kingdom life insurance market,
Applied Economics Letters 12, 339-344.
Society of Actuaries, 1996. Dynamic Financial Condition Analysis Handbook (Society of
Actuaries; USA).
Sweeney, J. C. et al., 1998. Optimal Insurance Asset Allocation in a Multi-Currency Environment,
in World Wide Asset and Liability Modelling, edited by W. Ziemba and Mulvey J., first
edition (Cambridge University Press; U.K.)
Szkoda, S. et al., 1995. CAS Dynamic Financial Analysis Handbook, Release 1.0 (Final)
(Casualty Actuarial Society; U.S.A.).
Walling, R. J., Hettinger, T. E., Emma, C. C., Ackerman, S. 1999. Customising the Public Access
Model Using Publicly Available Data, Casualty Actuarial Society Forum, 239-266.
Wilkie, A.D., 1986. A Stochastic Investment Model for Actuarial Use, Transactions of Faculty of
Actuaries, Vol. 39, 341-403
Wilkie, A.D., 1995. More on a Stochastic Model for Actuarial Use, British Actuarial Journal, Vol.
1, Part V, 777-964
Witcraft, S. E., 1998. Profitability Targets: DFA Provides Probability Estimates, Casualty
Actuarial Society Forum, 273-302.

59