1. Sensitivity Analysis and Worst-Case Analysis
– Making use of netting effects when
designing insurance contracts
Marcus C. Christiansen
September 6, 2009
IAA LIFE Colloquium 2009 in Munich, Germany
2. Risks in life insurance
present value of future benefits and premiums depends on ...
(a) random pattern of states (Xt )
-
r
r
r
r
r Xt
t0
d
i
a
→ biometrical risk:
unsystematic biometrical risk (diversifiable)
systematic biometrical risk (non-diversifiable)
(b) investment return / interest rate ϕ(t)
→ financial risk (non-diversifiable)
(c) ...
3. Risks in life insurance
present value of future benefits and premiums depends on ...
(a) random pattern of states (Xt )
-
r
r
r
r
r Xt
t0
d
i
a
4. Risks in life insurance
present value of future benefits and premiums depends on ...
(a) random pattern of states (Xt )
-
r
r
r
r
r Xt
t0
d
i
a
→ biometrical risk:
• unsystematic biometrical risk (diversifiable)
5. Risks in life insurance
present value of future benefits and premiums depends on ...
(a) random pattern of states (Xt )
-
r
r
r
r
r Xt
t0
d
i
a
→ biometrical risk:
• unsystematic biometrical risk (diversifiable)
• systematic biometrical risk (non-diversifiable)
(b) investment return / interest rate ϕ(t)
→ financial risk (non-diversifiable)
(c) ...
6. Risks in life insurance
present value of future benefits and premiums depends on ...
(a) random pattern of states (Xt )
-
r
r
r
r
r Xt
t0
d
i
a
→ biometrical risk:
• unsystematic biometrical risk (diversifiable)
• systematic biometrical risk (non-diversifiable)
(b) investment return / interest rate ϕ(t)
1970 1980 1990 2000
0
2
4
6
8
10
→ financial risk (non-diversifiable)
(c) ...
7. Getting rid of the non-diversifiable risks?
(i) benefits guaranteed
Policyholder Third PartyInsurer
8. Getting rid of the non-diversifiable risks?
(i) benefits guaranteed
Policyholder Third PartyInsurer
(ii) benefits according to surplus
e.g. with profits policies such as DC pension plans
Policyholder Third PartyInsurer
9. Getting rid of the non-diversifiable risks?
(i) benefits guaranteed
Policyholder Third PartyInsurer
(ii) benefits according to surplus
e.g. with profits policies such as DC pension plans
Policyholder Third PartyInsurer
• contrary to customer demand (in Germany)!
10. Getting rid of the non-diversifiable risks?
(i) benefits guaranteed
Policyholder Third PartyInsurer
(ii) benefits according to surplus
e.g. with profits policies such as DC pension plans
Policyholder Third PartyInsurer
• contrary to customer demand (in Germany)!
(iii) benefits guaranteed by third party
e.g. reinsurance, securitisation, ...
Policyholder Third PartyInsurer
11. Getting rid of the non-diversifiable risks?
(i) benefits guaranteed
Policyholder Third PartyInsurer
(ii) benefits according to surplus
e.g. with profits policies such as DC pension plans
Policyholder Third PartyInsurer
• contrary to customer demand (in Germany)!
(iii) benefits guaranteed by third party
e.g. reinsurance, securitisation, ...
Policyholder Third PartyInsurer
• enough capacity for risk transfer?
12. Getting rid of the non-diversifiable risks?
(iv) reality (in Germany): compositions of (i) – (iii)
Policyholder Third PartyInsurer
13. Getting rid of the non-diversifiable risks?
(iv) reality (in Germany): compositions of (i) – (iii)
Policyholder Third PartyInsurer
• still a significant load of systematic risks for the insurer !
14. Getting rid of the non-diversifiable risks?
(iv) reality (in Germany): compositions of (i) – (iii)
Policyholder Third PartyInsurer
• still a significant load of systematic risks for the insurer !
Question A: effect of policy design on risk load of the insurer ?
Policyholder Third Party
⇐ §§ ⇒
Insurer
15. Getting rid of the non-diversifiable risks?
(iv) reality (in Germany): compositions of (i) – (iii)
Policyholder Third PartyInsurer
• still a significant load of systematic risks for the insurer !
Question A: effect of policy design on risk load of the insurer ?
Policyholder Third Party
⇐ §§ ⇒
Insurer
(v) netting effects
e.g. survival benefits vs. death benefits
±§§
§§
§§
Policyholder Third PartyInsurer
16. Getting rid of the non-diversifiable risks?
(iv) reality (in Germany): compositions of (i) – (iii)
Policyholder Third PartyInsurer
• still a significant load of systematic risks for the insurer !
Question A: effect of policy design on risk load of the insurer ?
Policyholder Third Party
⇐ §§ ⇒
Insurer
(v) netting effects
e.g. survival benefits vs. death benefits
±§§
§§
§§
Policyholder Third PartyInsurer
Question B: which combinations give strong netting effects ?
18. Effect of changes of the valuation basis on premiums
and reserves
qualitative results:
Lidstone (1905), Norberg (1985), Hoem (1988), Ramlau-Hansen (1988),
Linnemann (1993), ...
19. Effect of changes of the valuation basis on premiums
and reserves
qualitative results:
Lidstone (1905), Norberg (1985), Hoem (1988), Ramlau-Hansen (1988),
Linnemann (1993), ...
quantitative results:
Bowers et al. (1987, 1997, constant interest), Dienst (1995, yearly invalidity),
Kalashnikov and Norberg (2003, general single parameter), C. and Helwich
(2008, yearly interest & yearly mortality),
20. Effect of changes of the valuation basis on premiums
and reserves
qualitative results:
Lidstone (1905), Norberg (1985), Hoem (1988), Ramlau-Hansen (1988),
Linnemann (1993), ...
quantitative results:
Bowers et al. (1987, 1997, constant interest), Dienst (1995, yearly invalidity),
Kalashnikov and Norberg (2003, general single parameter), C. and Helwich
(2008, yearly interest & yearly mortality),
f(x + ∆x) f(x) + x f, ∆x
interpret x f = f (x) as sensitivity of f at x
21. Effect of changes of the valuation basis on premiums
and reserves
qualitative results:
Lidstone (1905), Norberg (1985), Hoem (1988), Ramlau-Hansen (1988),
Linnemann (1993), ...
quantitative results:
Bowers et al. (1987, 1997, constant interest), Dienst (1995, yearly invalidity),
Kalashnikov and Norberg (2003, general single parameter), C. and Helwich
(2008, yearly interest & yearly mortality),
f(x + ∆x) f(x) + x f, ∆x
interpret x f = f (x) as sensitivity of f at x
• general approach for all valuation basis parameters and
with continuous time ?
22. General sensitivity analysis (C., 2008)
LET
Vt,a(ϕ, µad , µai , ...) := prospective reserve at time ’t’ in state ’a’
with a valuation basis (ϕ, µad , µai , ...) whose entries are either
(a) vectors of yearly interest rates, mortality probabilities, etc.
(b) or intensity functions for interest, mortality, etc.
(c) or cumulative intensity functions for interest, mortality, etc.
23. General sensitivity analysis (C., 2008)
LET
Vt,a(ϕ, µad , µai , ...) := prospective reserve at time ’t’ in state ’a’
with a valuation basis (ϕ, µad , µai , ...) whose entries are either
(a) vectors of yearly interest rates, mortality probabilities, etc.
(b) or intensity functions for interest, mortality, etc.
(c) or cumulative intensity functions for interest, mortality, etc.
THEN there exists a generalized first-order Taylor expansion
Vt,a(ϕ + ∆ϕ, µad + ∆µad , ...) Vt,a(ϕ, µad , ...)
+ ϕVt,a , ∆ϕ + µad
Vt,a , ∆µad + ...
24. General sensitivity analysis (C., 2008)
LET
Vt,a(ϕ, µad , µai , ...) := prospective reserve at time ’t’ in state ’a’
with a valuation basis (ϕ, µad , µai , ...) whose entries are either
(a) vectors of yearly interest rates, mortality probabilities, etc.
(b) or intensity functions for interest, mortality, etc.
(c) or cumulative intensity functions for interest, mortality, etc.
THEN there exists a generalized first-order Taylor expansion
Vt,a(ϕ + ∆ϕ, µad + ∆µad , ...) Vt,a(ϕ, µad , ...)
+ ϕVt,a , ∆ϕ + µad
Vt,a , ∆µad + ...
with generalized gradients / sensitivities
ϕVt,a (s) = −1s>t · v(t, s) ·
k
P(Xs = k|Xt = a) · Vs−,k
µad Vt,a (s) = 1s>t · v(t, s) · P(Xs− = a |Xt = a) · S@Rad (s)
25. Example
endowment insurance with disability waiver
30
1
252015
0,6
0,2
0,4
5
-0,2
0 35
0,8
0
10
(a,d)
(i,d)
(i,a)
ϕ
(a,i)
time
sensitivities of
the prospective reserve
at time t = 0 in state a =’active’
26. Example: mortality sensitivities
policyholder at time t = 0 is a 30 year old male
10 20 30 40 50 60 70
K2
K1
0
1
pure endowment ins.
(i,d)
annuity ins.
Φ
temporary life ins.
time
mortality sensitivities of
the prospective reserve
at time t = 0 in state a =’active’
30. Conclusion for Tool 1: Sensitivity Analysis
Pros
• yields graphic images of risk structures
Cons
31. Conclusion for Tool 1: Sensitivity Analysis
Pros
• yields graphic images of risk structures
• sensitivity is linear with respect to combinations of policies
Cons
32. Conclusion for Tool 1: Sensitivity Analysis
Pros
• yields graphic images of risk structures
• sensitivity is linear with respect to combinations of policies
very helpful tool to find netting effects
Cons
33. Conclusion for Tool 1: Sensitivity Analysis
Pros
• yields graphic images of risk structures
• sensitivity is linear with respect to combinations of policies
very helpful tool to find netting effects
Cons
• disregards the variability of the arguments
34. Conclusion for Tool 1: Sensitivity Analysis
Pros
• yields graphic images of risk structures
• sensitivity is linear with respect to combinations of policies
very helpful tool to find netting effects
Cons
• disregards the variability of the arguments
• no real-valued risk measure
35. Conclusion for Tool 1: Sensitivity Analysis
Pros
• yields graphic images of risk structures
• sensitivity is linear with respect to combinations of policies
very helpful tool to find netting effects
Cons
• disregards the variability of the arguments
• no real-valued risk measure
how to compare policies (with respect to their systematic
mortality risk) ?
41. Comparing / Measuring policies
IDEA (risk measure for systematic mortality risk):
use the SCR according to the standard formula in Solvency II
systematic mortality risk = Lifelong & Lifemort
42. Comparing / Measuring policies
IDEA (risk measure for systematic mortality risk):
use the SCR according to the standard formula in Solvency II
systematic mortality risk = Lifelong & Lifemort
• reflects the practical consequences for the insurer
43. Comparing / Measuring policies
IDEA (risk measure for systematic mortality risk):
use the SCR according to the standard formula in Solvency II
systematic mortality risk = Lifelong & Lifemort
• reflects the practical consequences for the insurer
• approximation of the risk measure V@R
44. Comparing / Measuring policies
IDEA (risk measure for systematic mortality risk):
use the SCR according to the standard formula in Solvency II
systematic mortality risk = Lifelong & Lifemort
• reflects the practical consequences for the insurer
• approximation of the risk measure V@R
• cost-of-capital method: proportional to some (hypothetical)
market price
45. Comparing / Measuring policies
IDEA (risk measure for systematic mortality risk):
use the SCR according to the standard formula in Solvency II
systematic mortality risk = Lifelong & Lifemort
• reflects the practical consequences for the insurer
• approximation of the risk measure V@R
• cost-of-capital method: proportional to some (hypothetical)
market price
BUT risk approximation by the standard formula is too rough
here !
46. Calculation of Lifemort and Lifelong
∆NAV = changes in the net value of assets and liabilities due to ...
40 45 50 55 60 65
0,000
0,005
0,010
0,015
0,020
+10% mortality shock
−25% longevity shock
mortality rate
± mixed shock
47. Calculation of Lifemort and Lifelong
∆NAV = changes in the net value of assets and liabilities due to ...
40 45 50 55 60 65
0,000
0,005
0,010
0,015
0,020
+10% mortality shock
−25% longevity shock
mortality rate
± mixed shock
• Do we really study the crucial scenarios?
48. MARKT/2505/08, TS.XI.B.3 & TS.XI.C.3
For those contracts that provide benefits both in case of death
and survival, one of the following two options should be chosen
[...]:
1 Contracts [...] should not be unbundled. [...] the mortality
scenario should be applied fully allowing for the netting
effect provided by the ’natural’ hedge between the death
benefits component and the survival benefits component.
[...]
49. MARKT/2505/08, TS.XI.B.3 & TS.XI.C.3
For those contracts that provide benefits both in case of death
and survival, one of the following two options should be chosen
[...]:
1 Contracts [...] should not be unbundled. [...] the mortality
scenario should be applied fully allowing for the netting
effect provided by the ’natural’ hedge between the death
benefits component and the survival benefits component.
[...]
2 All contracts are unbundled into 2 separate components:
one contingent on the death and other contingent on the
survival of the insured person(s). [...]
50. MARKT/2505/08, TS.XI.B.3 & TS.XI.C.3
For those contracts that provide benefits both in case of death
and survival, one of the following two options should be chosen
[...]:
1 Contracts [...] should not be unbundled. [...] the mortality
scenario should be applied fully allowing for the netting
effect provided by the ’natural’ hedge between the death
benefits component and the survival benefits component.
[...]
2 All contracts are unbundled into 2 separate components:
one contingent on the death and other contingent on the
survival of the insured person(s). [...]
51. MARKT/2505/08, TS.XI.B.3 & TS.XI.C.3
For those contracts that provide benefits both in case of death
and survival, one of the following two options should be chosen
[...]:
1 Contracts [...] should not be unbundled. [...] the mortality
scenario should be applied fully allowing for the netting
effect provided by the ’natural’ hedge between the death
benefits component and the survival benefits component.
[...] → not the crucial scenarios
2 All contracts are unbundled into 2 separate components:
one contingent on the death and other contingent on the
survival of the insured person(s). [...] → no netting effect
55. Optimization problem
If we focus on the liabilities only then ∆NAV = ∆V0,a .
Problem Find scenario µad with
V0,a(µad ) = max V0,a(µad ) LowBound ≤ µad ≤ UppBound
56. Optimization problem
If we focus on the liabilities only then ∆NAV = ∆V0,a .
Problem Find scenario µad with
V0,a(µad ) = max V0,a(µad ) LowBound ≤ µad ≤ UppBound
first-order Taylor approximation
V0,a(µad + ∆µad ) = V0,a(µad ) + µad
V0,a , ∆µad + Remainder
where µad V0,a (s) = 1s>0 · v(0, s) · P(Xs− = a |X0 = a) · S@Rad (s)
57. Optimization problem
If we focus on the liabilities only then ∆NAV = ∆V0,a .
Problem Find scenario µad with
V0,a(µad ) = max V0,a(µad ) LowBound ≤ µad ≤ UppBound
first-order Taylor approximation
V0,a(µad + ∆µad ) = V0,a(µad ) + µad
V0,a , ∆µad + Remainder
where µad V0,a (s) = 1s>0 · v(0, s) · P(Xs− = a |X0 = a) · S@Rad (s)
Conclusion 1 For all s
sign µad
V0,a(s) = sign S@Rad (s)
58. Optimization problem
If we focus on the liabilities only then ∆NAV = ∆V0,a .
Problem Find scenario µad with
V0,a(µad ) = max V0,a(µad ) LowBound ≤ µad ≤ UppBound
first-order Taylor approximation
V0,a(µad + ∆µad ) = V0,a(µad ) + µad
V0,a , ∆µad + Remainder
where µad V0,a (s) = 1s>0 · v(0, s) · P(Xs− = a |X0 = a) · S@Rad (s)
Conclusion 1 For all s
sign µad
V0,a(s) = sign S@Rad (s)
Conclusion 2 Because of the maximality property of µad :
µad
V0,a , ∆µad ≤ 0 for all LowBound ≤ µad + ∆µad ≤ UppBound
59. Solution of the optimization problem
Hence
µad =
UppBoundad : sign S@Rad > 0
LowBoundad : sign S@Rad < 0
& Thiele’s rekursion/differential/integral equation
THEOREM (C., 2008)
The Worst-Case rekursion/differential/integral equation has a
unique solution Vt,a that is maximal.
Conclusion
solution Vt,a =⇒ S@Rad =⇒ µad
60. Solution of the optimization problem
Hence
µad =
UppBoundad : sign S@Rad > 0
LowBoundad : sign S@Rad < 0
& Thiele’s rekursion/differential/integral equation
= Worst-Case rekursion/differential/integral equation
61. Solution of the optimization problem
Hence
µad =
UppBoundad : sign S@Rad > 0
LowBoundad : sign S@Rad < 0
& Thiele’s rekursion/differential/integral equation
= Worst-Case rekursion/differential/integral equation
THEOREM (C., 2008)
The Worst-Case rekursion/differential/integral equation has a
unique solution Vt,a that is maximal.
62. Solution of the optimization problem
Hence
µad =
UppBoundad : sign S@Rad > 0
LowBoundad : sign S@Rad < 0
& Thiele’s rekursion/differential/integral equation
= Worst-Case rekursion/differential/integral equation
THEOREM (C., 2008)
The Worst-Case rekursion/differential/integral equation has a
unique solution Vt,a that is maximal.
Conclusion
solution Vt,a =⇒ S@Rad =⇒ µad
63. Example 1
Task Design a combination of
1×pure endowment ins. + β× temporary life ins.
64. Example 1
Task Design a combination of
1×pure endowment ins. + β× temporary life ins.
Question Which ratio
death benefit : survival benefit = β : 1
leads to the lowest systematic mortality risk ?
65. Example 1
Task Design a combination of
1×pure endowment ins. + β× temporary life ins.
Question Which ratio
death benefit : survival benefit = β : 1
leads to the lowest systematic mortality risk ?
10 20 30
K0,2
0
0,2
0,4
0,6
0,8
sensitivity: temp. life ins.
sensitivity: pure endowm. ins.
66. Example 1
Task Design a combination of
1×pure endowment ins. + β× temporary life ins.
Question Which ratio
death benefit : survival benefit = β : 1
leads to the lowest systematic mortality risk ?
10 20 30
K0,2
0
0,2
0,4
0,6
0,8
0 1 2 3
0
0,002
0,004
0,006
0,008
0,010
0,012
0,014
0,016
0,018
sensitivity: temp. life ins.
sensitivity: pure endowm. ins.
systematic mortality risk
β : 1
67. Example 1
Task Design a combination of
1×pure endowment ins. + β× temporary life ins.
Question Which ratio
death benefit : survival benefit = β : 1
leads to the lowest systematic mortality risk ?
10 20 30
K0,2
0
0,2
0,4
0,6
0,8
0 1 2 3
0
0,002
0,004
0,006
0,008
0,010
0,012
0,014
0,016
0,018
sensitivity: temp. life ins.
sensitivity: pure endowm. ins.
systematic mortality risk
β : 1
68. Example 2
Task Design a combination of
1×disability ins. + β× temporary life ins.
Question Which ratio
death benefit : yearly disability benefit = β : 1
leads to the lowest systematic mortality risk ?
69. Example 2
Task Design a combination of
1×disability ins. + β× temporary life ins.
Question Which ratio
death benefit : yearly disability benefit = β : 1
leads to the lowest systematic mortality risk ?
0 1 2 3 4
0
0,01
0,02
0,03
synchronised changes of (a, d) and (i, d)
β : 1 β : 1
70. Example 2
Task Design a combination of
1×disability ins. + β× temporary life ins.
Question Which ratio
death benefit : yearly disability benefit = β : 1
leads to the lowest systematic mortality risk ?
0 1 2 3 4
0
0,01
0,02
0,03
synchronised changes of (a, d) and (i, d)
β : 1 β : 1
71. Example 2
Task Design a combination of
1×disability ins. + β× temporary life ins.
Question Which ratio
death benefit : yearly disability benefit = β : 1
leads to the lowest systematic mortality risk ?
0 1 2 3 4
0
0,01
0,02
0,03
0 1 2 3 4
0
0,01
0,02
0,03
synchronised changes of (a, d) and (i, d)
independent changes of (a, d) and (i, d)
β : 1 β : 1
72. Example 3
Task Design a combination of
1×annuity ins. + β× temporary/whole life ins.
Question Which ratio
death benefit : yearly annuity benefit = β : 1
leads to the lowest systematic mortality risk ?
73. Example 3
Task Design a combination of
1×annuity ins. + β× temporary/whole life ins.
Question Which ratio
death benefit : yearly annuity benefit = β : 1
leads to the lowest systematic mortality risk ?
0 10 20 30 40
0
0,1
0,2
0,3
0,4
temporary life ins.
β : 1 β : 1
74. Example 3
Task Design a combination of
1×annuity ins. + β× temporary/whole life ins.
Question Which ratio
death benefit : yearly annuity benefit = β : 1
leads to the lowest systematic mortality risk ?
0 10 20 30 40
0
0,1
0,2
0,3
0,4
temporary life ins.
β : 1 β : 1
75. Example 3
Task Design a combination of
1×annuity ins. + β× temporary/whole life ins.
Question Which ratio
death benefit : yearly annuity benefit = β : 1
leads to the lowest systematic mortality risk ?
0 10 20 30 40
0
0,1
0,2
0,3
0,4
0 10 20 30 40
0
0,1
0,2
0,3
0,4
temporary life ins.
whole life ins.
β : 1 β : 1
76. Example 3
Task Design a combination of
1×annuity ins. + β× temporary/whole life ins.
Question Which ratio
death benefit : yearly annuity benefit = β : 1
leads to the lowest systematic mortality risk ?
0 10 20 30 40
0
0,1
0,2
0,3
0,4
0 10 20 30 40
0
0,1
0,2
0,3
0,4
temporary life ins.
whole life ins.
β : 1 β : 1
77. literature
• Christiansen, M.C. (2008): A sensitivity analysis concept
for life insurance with respect to a valuation basis of infinite
dimension. Insurance: Mathematics and Economics 42,
680-690.
• Christiansen, M.C. (2008): A sensitivity analysis of typical
life insurance contracts with respect to the technical basis.
Insurance: Mathematics and Economics 42, 787-796.
• Christiansen, M.C., Helwich, M. (2008): Some further
ideas concerning the interaction between insurance and
investment risks. Blätter der DGVFM 29 (2), 253-266.
• Christiansen, M.C. (2008): Biometrical worst-case and
best-case scenarios in life insurance. Preprint. Available at
http://www.uni-rostock.de/∼christiansen/