The current issue and full text archive of this journal is available at

www.emeraldinsight.com/1526-5943.htm

Backtesting the solvency capital

requirement for longevity risk

Backtesting
the SCR for
longevity risk

Mariarosaria Coppola
Department of Theories and Methods of Human and Social Sciences,
Federico II University, Naples, Italy, and

Valeria DAmato
Department of Statistics and Economics, University of Salerno, Salerno, Italy

309
Received January 2012
Revised April 2012
Accepted May 2012

Abstract
Purpose The determination of the capital requirements represents the first Pillar of Solvency II.
The main purpose of the new solvency regulation is to obtain more realistic modelling and assessment
of the different risks insurance companies are exposed to in a balance-sheet perspective. In this
context, the Solvency Capital Requirement (SCR) standard calculation is based on a modular approach,
where the overall risk is split into several modules and submodules. In Solvency II, standard formula
longevity risk is explicitly considered. The purpose of this paper is to look at the backtesting approach
for measuring the consistency of SCR calculations for life insurance policies.
Design/methodology/approach A multiperiod approach is suggested for correctly calculating
the SCR in a risk management perspective, in the sense that the amount of capital necessary to meet
company future obligations year by year until the contract will be in force has to be assessed. The
backtesting approach for measuring the consistency of SCR calculations for life insurance policies
represents the main contribution of the research. In fact this kind of model performance is generally
specified in the VaR validation analysis. In this paper, this approach is considered for testing the ex
post performance of SCR calculation methodology.
Findings The backtesting framework is able to measure, from time to time, if the insurer has
allocated more or less capital to support his in-force business, with adverse effects on free reserves and
profitability or solvency.
Practical implications The paper shows that the forecasting performance is an important aspect
to assess the effectiveness of the model, a poor performance corresponding to a biased allocation of
capital.
Originality/value The backtesting approach for measuring the consistency of SCR calculations
for life insurance policies represents the main contribution of the research. In fact this kind of model
performance is generally specified in the VaR validation analysis. Recently, Dowd et al. have proposed
it for verifying the goodness of mortality models and now, in this paper, this approach is considered
for testing the ex post performance of SCR calculation methodology.
Keywords Capital, Finance, Regulation, Risk analysis, Insurance companies, Life insurance, Solvency II,
Solvency capital requirement, Longevity risk, Iterative Lee Carter model, Life annuity portfolio, Backtest
Paper type Research paper

1. Introduction
The determination of capital requirements represents the first pillar of Solvency II.
In this framework the solvency capital requirement (SCR) is intended to be the amount
of capital that an insurer needs in order to remain viable in the market and maintain its
default probability below a certain level (CEIOPS, 2007). The main purpose of the new
solvency regulation is to obtain a more realistic modelling and assessment of the
different risks insurance companies are exposed to. According to this regulation the SCR

The Journal of Risk Finance

Vol. 13 No. 4, 2012
pp. 309-319
q Emerald Group Publishing Limited
1526-5943
DOI 10.1108/15265941211254444

JRF
13,4

310

calculation could rely on a standard formula, full internal models or partial internal
models coupled with some parts of the standard model.
The basic principal is that the SCR will be determined as the 99.5 percent value at
risk (VaR) of the available capital over one-year time horizon (Artzner et al., 1997).
Insurance companies are encouraged to enforce (stochastic) internal models since
they should provide a more accurate assessment of the insurance risks. Unfortunately
such models are rather expensive and sophisticated, therefore small and medium-size
companies could prefer to rely on the standard model, even if also larger companies
could prefer to implement a few modules for their (partial) internal models. For these
reasons the European Commission has furnished a standard model that insurance
companies are allowed to use for approximating the capital requirements. This standard
model is based on a modular approach: the overall risk is split into several risks
(modules) for each of them risk sub-modules are considered. Modules and submodules
SCRs are computed separately and then aggregated according to a pre-specified
correlation matrices.
The European Commission for calibrating this standard model has recently
published the technical specification of 5th Quantitative Impact Study-QIS5 (CEIOPS,
2010) which maybe represents the last opportunity for insurance company for
evaluating the capital amount necessary to satisfy the new solvency regulations.
Longevity risk, i.e. the risk that the trend of longevity improvements significantly
change in the future, is one of the main risks insurers or pension funds providers have
to front. Whereas in most industrialized countries the fall in benefits from public pay as
you go pension schemes, and in general the uncertainty in the public pension systems,
it is expected that the relevance of the longevity phenomenon will increase in the next
future. In agreement with these considerations, longevity risk is explicitly considered
in Solvency II standard formula as a submodule of the life underwriting risk module.
A wide literature has been recently interested in these issues. Some authors analyzed
capital requirement for certain portfolios but considering approaches different from the
one-year 99.5 percent VaR of Solvency II, for example Hari et al. (2008) and Olivieri and
Pitacco (2008). Others considered the impact and the significance of longevity risk on
annuity or pension fund portfolios but they did not relate to capital requirement under a
given solvency regime. More recently Borger (2010) analyzed the adequacy of the
longevity shock specified in QIS4 (CEIOPS, 2007) standard formula comparing the
resulting capital requirement to the VaR based on a stochastic mortality model. He found
structural shortcomings and proposed a modified longevity shock for Solvency II
standard model.
This paper refers exactly to the sub-module of longevity risk. A multiperiod
perspective for correctly calculating the SCRs is suggested, in the sense that the
amount of capital necessary to meet company future obligations has to be assessed
year by year till the contract will be in force.
The backtesting approach for measuring the consistency of SCR calculations for life
insurance policies represents the main contribution of the present research. The term
backtest is usually referred to any method of evaluating forecasts against subsequently
realized outcomes. Recently, Dowd et al. (2010) have proposed it for verifying the
goodness of mortality models and now this approach is considered for testing the ex post
performance of SCR calculation methodology, in the persepective of the predictability
tests based on the in-sample fit model.

Generally this kind of model performance is specified in the VaR validation

analysis.
In particular, the backtesting procedure was originally designed for evaluating the
accuracy of VaR model, in the context of the market risk management, as laid out by
the Basle Committee on Banking Supervision (Christoffersen, 2004).
The main shortcomings deriving from the VaR are inevitably connected to the
underlying assumptions of the VaR methodologies. There is a large body of literature
on the topic and its relative advantages and disadvantages. VaR, however, has been
criticized on two grounds.
First of all, the VaR violates the characterization of the coherent risk measure
(Artzner et al., 1997, 1999).
Furthermore, the standard VaR measure presumes that the underlying distribution
is Gaussian, while it tends be nonnormal and to have fat-tails (Huisman et al., 2001).
By the aforementioned considerations, our contribution is the exploration of new
tool for backtesting based on the simulation algorithm which generates incorrelated
values, according to the probability distribution of the complex multifactor
phenomenon under consideration, the SCR exposed to the longevity risk.
A statistical analysis, the so-called distribution fitting, is performed on the data, to
verify the conformity to a given theoretical model. Finally, the goodness of fit is
calculated on the basis of the x 2-test.
The paper is organised as follows: in Section 2 the longevity phenomenon is
investigated, in Section 3 the SCR calculation in a multiperiod approach is discussed
according to the QIS5 guidelines, in Section 4 the backtesting framework for measuring
the consistency of SCR calculations for life insurance policies is proposed, Section 5
provides graphical analysis and numerical evidences. Section 6 concludes.
2. Longevity risk
It is very challenging to capture the tendency of the future mortality pattern, in
particular at retirement ages when the rectangularization phenomenon and the random
marked fluctuations are combined.
The risk connected to the mortality trend comes out in different ways. As concerns
the accidental component, one individual may live longer than the average lifetime in
the reference population. It corresponds to possible deviations around expected
mortality rates. It is related to the individual position and it becomes negligible in
respect of the large portfolios because of the pooling effect.
As concerns the systematic one, the average lifetime of a population may differ from
what it is expected. It refers to the deviations from expected values, rather than around
them. It reveals its systematic nature. This component matches up with longevity risk
and it considers the aggregate mortality phenomenon. Its effect may be significant if
referred to portfolios of long duration life contracts such as pension annuities,
characterized by a multiplicity of payments. Therefore, the correct assessment of the
longevity risk first involves a stochastic representation of mortality for measuring the
possible impact on the future payments and on annual outflows.
Risk management tools for dealing with longevity risk include reinsurance
arrangements and alternative risk transfers as securitization and in particular
mortality-linked securities.

Backtesting
the SCR for
longevity risk
311

JRF
13,4

312

From the insurer point of view, the adoption of internal models addressing the
longevity risk is needed to operate an appropriate capital allocation policies.
3. The SCR for longevity risk: a risk management framework
The SCR calculation according to Solvency II standard formula is based on a modular
approach which allows to obtain the SCR summing the capital requirement
for operational risk (SCRop) and adjustment for the risk absorbing effect of technical
provisions and deffered taxes (SCRAdj) to the basic SCR, from herein BSCR (Butt and
Haberman, 2009).
The BSCR is the SCRs before any adjustment and it is computed combining, on the
basis of a pre-specified correlation matrix Corr, capital requirement for six main risk
categories (modules): market risk, health underwriting risk, default risk, life
underwriting risk, non-life underwriting risk, intangible assets risk, so it follows:
sX

Corr i;j SCRi SCRj SCRint angibles

1
BSCR
i;j

where:
Corri, j

item set out in row i and column j of the correlation matrix.

SCRi,SCRj capital requirements for the individual SCR risks according to the
rows and coloumns of the correlation matrix.
Each modules listed above consists of several submodules whose corresponding SCRs
are calculated aggregating the submodules SCRs according to a given correlation
matrix.
Given the purpose of this paper the longevity risk is focused. It represents a specific
submodule of Life underwriting risk module and covers the risk of losses or adverse
changes in value of insurance liabilities resulting from changes in level, in the trend or
in the volatility of mortality rates, where a decrease in death rate lead to an increase in
the value of the insurers liabilities. According to Solvency II standard formula capital
charge for longevity risk (SCRlong) should be calculated as the change in net assets
value (NAV) due to a longevity shock under a specific survival scenario at time t 0
(Butt and Haberman, 2009). Hence:
SCRlong DNAV jlongevity shock

The longevity shock is represented by a 20 percent permanent reduction of the

mortality rates for each age and contract linked to longevity risk. As specified in
Solvency II regulations the parameters and the assumptions used for SCR calculation
are calibrated to correspond to the VaR of the basic own funds of an insurance
or reinsurance undertaking subject to a confidence level of 99.5 % over a one year
period (Artzner et al., 1997).
It is worth stressing that CEIOPS (2010) defined the NAV as the difference between
the market value of assets and liabilities. As well known, the market value of liabilities
is difficult to determine, therefore it stated that it can be approximated by the so-called
technical provisions which consists of the best estimate liabilities (BEL) and risk
margin (RM).

The RM can be interpreted as loading for facing all residual risk in respect of those
met by the SCR. It is calculated via a cost of capital approach and in the case under
consideration, taking into account only the longevity risk, it results:
RM t

X CoC SCRlong; th
h$0

1 r f 2h

313

being:
CoC

the cost of capital rate.

rf

the risk free interest rate.

In order to solve the evident situation of circularity, CEIOPS (2010) specifies that for
SCR calculation liabilities, should not include RM. Therefore, we have:
NAV t At 2 BELt

where At represents the market value of assets at time t.

The Solvency II capital requirement are defined according to a balance-sheet
framework looking at the insurer obligations over one single year. In this paper, a risk
management perspective is considered, therefore capital requirements are evaluated at
the beginning of each year with respect to the duration of the contract. In this sense, the
study refers to a multiperiod approach where, given a specific scenario, the insurer
may estimate today, t 0, what will be the amount of capital necessary to meet its
future obligations year by year till the contract will be in force.
4. SCR backtesting approach
In the context under consideration, a backtest framework is set out to evaluate at a given
time t if the insurer has correctly calculated the SCR in order to front his future obligations.
Note that the term backtest is referred to any method of evaluating forecasts against
subsequently realized outcomes. Generally, this kind of model performance is specified
in the VaR validation analysis and recently Dowd et al. (2010) have proposed it for
verifying the goodness of mortality models.
A key element of backtesting that differentiates it from other forms of historical
testing is that backtesting calculates how a strategy would have performed if it had
actually been applied in the past. This requires the backtest to replicate the conditions
of the time in question in order to get an accurate result.
In the context under consideration, the backtesting framework is designed to
measure from time to time if the insurer has allocate more or less capital to support his
in force business, with adverse effects on free reserves and profitability or solvency.
As shown in Dowd et al. (2010) a good model should produce forecasts that perform
well out-of-the sample, as well as provide good fits to the historical data and plausible
forecasts ex ante.
To investigate the predictive power of the SCRlong model, a particular backtest is
proposed: the contracting horizon backtest.
The main steps of the test are the following:
(1) selection of the metric of interest;
(2) selection of lookback window;

Backtesting
the SCR for
longevity risk

JRF
13,4

314

(3) selection lookforward window; and

(4) comparison of forecasts with realized outcomes.
The metric of interest, in evaluating the capital amount necessary to satisfy the new
solvency regulations, is the SCRlong which is a complex multifactor value based on a
given financial and demographic scenario.
The lookback window is represented by the reference historical horizon, it is chosen
taking into account that if the window length is n and the evaluation time is t,
observations from years t 2 n to t 2 1 are used.
The lookforward window is represented by the prediction interval, that is the
horizon over which the forecasts are made.
For comparing forecasts with realized outcomes it is necessary to specify what
constitutes a fail or a pass result.
The accuracy of the projections is measured by contracting the backtest horizon
over the time. In other words a start date, i.e. the stepping-off year, and a fixed end
date are fixed. The SCRlong is estimated on the basis of the information available at the
starting time t up to the end date n. As the temporal horizon tends to the end date, the
SCRlong formula is re-estimate by using the same numbers of historical observations,
according to the best estimate for representing the mortality dynamics. Intuitively if
the forecasts are good the consecutive obtained projections should gradually converge
towards the actual outcome at the forecast year. The actual SCRlong is evaluated on the
basis of the mortality experience in respect at that date, the calculation being done on
the observed death counts. The empirical findings are summarized by graphs and
charts of prediction intervals computed on N path simulations.
5. Numerical applications
In this section, a portfolio of c 1,000 immediate life annuity is considered. It is
assumed that all the annuitants are aged 40 at the issue time t 0 and they will receive
an annual amount b 1 at time t 1, 2,. . . until death. As regards the financial
scenario an interest rate equal to 2 percent is supposed.
The lifetimes of the insureds are supposed to be identically distributed and
independent given any mortality assumptions.
According to formula (4) and taking into account that for immediate life annuities
current assets are not stricken by the shock scenario, formula (2) can be simplified as
follows:
2 BELt
5
SCRlong;t BELshock
t
where:
BELt

the expected value of future payments according to a best estimate

life table.

the expected value of future payments according to a life table whose

BELshock
t
mortality rates are 20 percent lower than those in the best estimate
table, as stated in QIS5.
The mortality best estimate is derived according to the iterative Lee Carter model
(Huisman et al., 2001) estimated on the Italian male population dataset from 1950 to
2003 in R environment by ILC package (Artzner et al., 1999).

A backtesting on mortality rates is performed following a contracting horizon

scheme (CEA, 2006). A 33 years lookback window is considered, from 1950-1983 to
1970-2003, the forecast date is fixed at 2004. The lookforward window, that is the
horizon over which the forecasts are made, beginning from 1984 is characterized by a
contracting length up to the final date 2004.
Figure 1 shows that the forecasted outcomes obtained on the basis of the different
datasets converge towards the actual outcome revealing the goodness of the adopted
mortality model.
The backtest results obtained on the basis of 10,000 Monte Carlo simulations show
that the mortality rates q40 forecasted for 2004 according to the different datasets
converge to the mortality rate realized in the year 2004.
In a risk management environment we consider a multiperiod approach for
calculating, according to formula (5) the SCRlong based on the Iterative Lee Carter Model
(CEIOPS, 2010; Artzner et al., 1999) and the actual SCRlong. Fixing a demographic
scenario, Table I shows a decreasing behaviour of SCRlong, due to the decreasing
outflows for benefits.
Roughly speaking, we can observe an underestimation of the best estimate values
(Lee Carter SCRlong) in respect of realized outcomes, the reason is that the realized
outcomes incorporate automatically the improvements in longevity. Unlike, in the
multiperiod perspective the demographic scenario at t 0 is fixed, so the estimate of
mortality rate is not modified in the time. So the crude multiperiod approach cannot be
considered for testing SCRlong. In a certain sense comparing SCRlong different for
construction an implicit mistake would be made. To avoid this situation a contracting
horizon backtest in a similar way as made for testing the mortality model is proposed,
now the metric of interest is represented by the SCRlong at the forecast date.

Backtesting
the SCR for
longevity risk
315

Backtest on simulated survival rate

median px, x = 40
0.968
0.966
0.964
0.962
0.96
0.958
0.956
0.954
1

9 10 11 12 13 14 15 16 17 18 19 20 21
number of datasets

LC SCRlong

10
20
30
40
50

311.022
4.5538
0.00036295
0
0

Figure 1.
Backtest on 10,000 Monte
Carlo simulations of the
survival rate qx, age
x 40

Actual SCRlong
883.773
810.4921
274.5006
7.4557
0.000236

Table I.
Multiperiod SCRlong
comparison between LC
values and actual ones

JRF
13,4

316

Figure 2 shows that the SCRlong converges towards the actual value, this result is
particular revealing because it indicates that the obtained estimates implicitly incorporate
the longevity improvements due to the updating process of the lookback window.
To improve the significance of the validation for each SCRlong value obtained by a
specific dataset 10,000 Monte Carlo simulations were made. In Figure 3 the median,
minimum and maximum values are reported, we superimposed the red circle which
indicates the actual SCRlong at final date 2004.
It is stated that if a model is adequate, one can be confident that the prediction intervals
contain the actual values comparing the forecasts against realized outcomes.

Backtesting SCR2004 on projections

736
728
720
712
704
696
688
680
672
664
656
648
640

Figure 2.
Backtest on SCRlong,
final date 2004

o
1

10 11 12 13 14 15 16 17 18 19 20 21

number of dataset
SCR2004

Simulated SCR
3,000
2,500
2,000
1,500
1,000
500
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
number of dataset

Figure 3.
Backtest on SCRlong,
10,000 simulations

mean values
maximum values

minimum values
Realized SCR

The models performance is reflected in the degree of consistency between the realized
outcomes and the prediction interval associated with each forecast. According to the
Figure 3 the realized SCRlong falls in the bandwidth between the maximum and
minimum simulated values.
6. Concluding remarks
In the Solvency II context, the insurance companies put aside the capital as regards the
different kind of risk recognised by the QIS5. In particular, in the regulation
framework, SCR is intended to be the amount of capital that an insurer needs in order
to remain viable in the market and maintain its default probability below a certain
level (Klein et al., 2009).
Nevertheless, the balance-sheet approach prescribed by the new international
guidelines could not adequately catch the effective impact of the longevity risk in the
risk management valuations, given its specificity and incidence on long-term portfolios.
In this paper, we propose a multiperiod approach for correctly calculating the SCRs.
The multiperiod perspective is in the sense that we have to assess the amount of capital
necessary to meet company future obligations year by year till the contract will be in
force. In this setting, the aim of our research is to set out a backtesting framework for
measuring the consistency of SCR calculation methodologies for life insurance
portfolios. Indeed, the forecasting performance is an important aspect to assess the
effectiveness of the model, a poor performance corresponding to a biased allocation of
capital. In this order of ideas, it is relevant to represent accurately the future mortality
trend, as well as to consider the right tools to test the effectiveness of the formula or
model which it is implemented to calculate the SCR. We set out a backtesting
framework to evaluate the ex post performance of SCRlong calculation methodology. In
particular, on the basis of the modular approach in QIS5, we focus on the longevity risk
which represents a specific submodule of Life underwriting risk module.
The backtesting framework is generally specified in the VaR validation analysis.
Dowd et al. (2010) have proposed it for testing the goodness of mortality models, the
basic idea is that forecast distributions should be compared against subsequently
realized mortality outcome: realized outcomes compatible with their forecasted
distributions mean that the forecasts and the models that generated them are good
ones; incompatible forecast distributions and realized outcomes, means that the
forecasts and models are poor.
In the present paper we proposed this approach for testing the ex post performance
of SCR calculation methodology. In our case, the metric of interest is represented by the
SCRlong which is a multifactor value based on a given financial and demographic
scenario. The approach we suggested is a very general one; we believe that it can be
extended to all cases in which the adequacy of the results obtained by applying a
predictive model has to be evaluated. The essential condition is that the backtesting
framework has to be constructed in an appropriate manner. So what does it mean by
appropriately selecting first of all the metric of interest and then the lookback and the
lookforward windows. Also essential is to take into account that the obtained results
are linked to the particular dataset and sample periods we refer to.
The backtesting procedures developed in this paper can be seen as a diagnostic
check on an internal risk model. Thus, for complementing the investigation, various
specific tools could be taken into account.

Backtesting
the SCR for
longevity risk
317

JRF
13,4

318

In particular, further studies on possible extensions of the framework illustrated

here could be addressed to other risk measures which are considered for determining
provisions and capital requirements of an insurer, in order to avoid insolvency.
In particular, the main viable alternative to the widely used VaR is represented by
the Tail value-at-risk (TailVaR), that is the expected value of the loss in those cases
where it exceeds the predefined confidence level. It is sometimes also mentioned as
conditional tail expectation (CTE), expected shortfall (ES) or expected tail loss. Thus,
the TailVaR is equal to the average loss a company will suffer in case of (extreme)
situations where losses exceed the predefined confidence level (of 99.5 percent) (CEA
working paper on the risk measures VaR and TailVaR, 2006).
Backtesting the ES would have the benefit of potentially increasing the information
provided in the risk measure to reject a misspecified risk model. It ought to be a
supplement upon the magnitude of violations which we should expect threshold
determined by VaR.
For further future development of our research we would like to implement other
backtesting procedure, in particular we would be very interested in implementing some
formal hypothesis test based on comparisons of realized outcomes against forecasts.

References
Artzner, P., Delbaen, F., Eber, J.-M. and Heath, D. (1997), Thinking coherently, RISK, Vol. 10
No. 11.
Artzner, P., Delbaen, F., Eber, J.-M. and Heath, D. (1999), Coherent measures of risk, Math. Fin.,
Vol. 9 No. 3, pp. 203-28.
Borger, M. (2010), Deterministic shock vs. stochastic value-at-risk an analysis of the
Solvency II standard model approach to longevity risk, Blatter der DGVFM, Vol. 31,
pp. 225-59.
Butt, Z. and Haberman, S. (2009), Ilc: A Collection of R Functions for Fitting a Class of Lee-Carter
Mortality Models Using Iterative Fitting Algorithms, Actuarial Research Paper, No.190,
Cass Business School, London.
CEA (2006), Working paper on the risk measures VaR and TailVaR.
CEIOPS (2007), QIS 4, Technical Specifications, available at: https://eiopa.europa.eu/
fileadmin/tx_dam/files/publications/submissionstotheec/CEIOPS-DOC-23-07%20QIS4%
20-%20Technical%20Specifications%20%20Rev.pdf
CEIOPS (2010), QIS 5, Technical Specifications, available at: https://eiopa.europa.eu/consultations/
qis/quantitative-impact-study-5/technical-specifications/index.html
Christoffersen, P. and Pelletier, D. (2004), Backtesting value-at-risk: a duration-based approach,
Journal of Financial Econometrics, Vol. 2 No. 1, pp. 84-108.
Dowd, K., Cairns, A.J.G., Blake, D.P., Coughlan, G., Epstein, D. and Khalaf-Allah, M. (2010),
Backtesting stochastic mortality an ex-post evaluation of multi-period ahead-density
forecasts, Pension Institute, North American Actuarial Journal, Vol. 14 No. 3, pp. 281-98.
Hari, N., De Waegenaere, A., Melenberg, B. and Nijman, T.E. (2008), Longevity risk in
portfolios of pension annuities, Insurance Mathematics and Economics, Vol. 42 No. 2,
pp. 505-18.
Huisman, R., Koedijk, K.G., Kool, C.J.M. and Palm, F. (2001), Tail-index estimates in small
samples, Journal of Business & Economic Statistics, Vol. 19 No. 2.

Klein, R.W. and Wang, S. (2009), Catastrophe risk financing in the United States and the
Eurpean Union: a comparative analysisn of alternative regulatory approaches, The
Journal of Risk and Insurance, Vol. 76 No. 3, pp. 607-37.
Olivier, A. and Pitacco, E. (2008), Solvency requirements for life annuity: some comparisons,
working paper, University of Parma, Parma.

Backtesting
the SCR for
longevity risk

Further reading
Lee, R.D. and Carter, L.R. (1992), Modelling and forecasting US mortality, Journal of the
American Statistical Association, Vol. 87, pp. 659-71.
Renshaw, A.E. and Haberman, S. (2003a), Lee-Carter mortality forecasting: a parallel
generalized linear modeling approach for England and Wales mortality projections,
Applied Statistics, Vol. 52, pp. 119-37.
Renshaw, A.E. and Haberman, S. (2003b), Lee-Carter mortality forecasting with age specific
enhancement, Insurance: Mathematics and Economics, Vol. 33, pp. 255-72.
Renshaw, A.E. and Haberman, S. (2003c), On the forecasting of mortality reduction factors,
Insurance: Mathematics and Economics, Vol. 32, pp. 379-401.

319

Corresponding author
Valeria DAmato can be contacted at: vdamato@unisa.it

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com

Or visit our web site for further details: www.emeraldinsight.com/reprints

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.