Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
www.emeraldinsight.com/1526-5943.htm
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
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.
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
where:
Corri, j
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 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
rf
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
Backtesting
the SCR for
longevity risk
JRF
13,4
314
Backtesting
the SCR for
longevity risk
315
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.
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
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
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.