Optimal (Re)insurance Contract Design

Part I: Introduction
C HENGGUO W ENG

Department of Statistics and Actuarial Science

University of Waterloo

READI Research Workshop - Module 2

Feb 22, 2019

Chengguo Weng([email protected]) – p. 1/18

 Outline

• Section I-1. Optimal (Re)insurance Models

• Section I-2. Risk Measures

• Section II-3. Topics in the Module

Chengguo Weng([email protected]) – p. 2/18

• Section I-1.
Optimal (Re)insurance Models

Chengguo Weng([email protected]) – p. 3/18

Section I-1. Optimal (Re)insurance Models
Reinsurance
• Reinsurance is an insurance on insurance.

• Why Reinsurance?
 Mitigate the risk exposure of an insurer and hence stabilize the underwriting
(or earnings) volatilities.

 Utilized by the insurer to avoid a large single loss, which might lead to the
insurer’s bankruptcy.

 A newly established insurance company can obtain the business expertise from
some reinsurance companies by relating to them through reinsurance contracts.

 Reinsurance also provides a mechanism allowing an insurance company to

increase its capacity to accept risks.

Chengguo Weng([email protected]) – p. 4/18

Section I-1. Optimal (Re)insurance Models
Reinsurance Models

Premium π0
Policyholders Insurance Company
(Insureds) (Insurer or Cedent)
Loss X

Π(f ): Reinsurance Premium

f (X): Ceded Claims
e.g. Expected value premium Π(f )
e.g. stop loss:
principle:
f (X) = (X − d)+ , d ≥ 0
Π(f ) = (1 + θ)E[f (X)]
e.g. quota-share:
Reinsurance Company (Reinsurer) f (X) = cX, 0 ≤ c ≤ 1
• Insurer’s retained risk: Rf (X) = X − f (X)
• Insurer’s total risk: Tf (X) = Rf (X) + Π(f ) = X − f (X) + Π(f )
• Tradeoff between the amount of loss retained and the reinsurance premium
payable to a reinsurer

Chengguo Weng([email protected]) – p. 5/18

 Two classic optimal reinsurance models
• The stop-loss reinsurance I ∗ (X) = (X − d∗ )+ with retention d∗
satisfying (1 + θ)E[I ∗ ] = π solves the following two classic
optimal reinsurance models




min Var R(X) = Var X − I(X)
s.t. 0 ≤ I(x) ≤ x, for all x ≥ 0, and (1 + θ)E[I] = π.


max Eu(W0 − X + I(X) − π)
s. t. 0 ≤ I(x) ≤ x, for all x ≥ 0, and (1 + θ)E[I] = π,

where W0 is the insurer’s initial capital, and u is her utility

function which is concave. The later model is due to Arrow
(1963).

Chengguo Weng([email protected]) – p. 6/18

 Optimality of stop-loss reinsurance
• Consider the stop-loss contract with

I ∗ (X) = (X − d)+ and R∗ (X) = X − I ∗ (X) = X ∧ d

where E[I] = π/(1 + θ) and E[R∗ ] = C := E[X] − π/(1 + θ).

• Decomposition of variance:

Var(R) = E[R2 ] − (E[R])2 = E[(R − d)2 + 2Rd − d2 ] − C 2

= E[(R − d)2 ] + 2Cd − d2 + C 2

• Note that 0 ≤ I(x) ≤ x implies 0 ≤ R(x) ≤ x; moreover,

2 2 2
   
E[(R − d) ] = E (R − d) I{X≤d} + E (R − d) I{X>d}
2 2
   
≥ E (X − d) I{X≤d} + E (d − d) I{X>d}
2
 
= E (X − d) I{X≤d}

∗ ∗
2
d)2+ ] R∗ )2+ ] 2
 
E[(R − d) ] = E[(R − − E[(d − = E (X − d) I{X≤d}
Chengguo Weng([email protected]) – p. 7/18
Section I-1. Optimal (Re)insurance Models
Risk Measure Minimization Reinsurance Models
• A plausible optimal reinsurance model:


minf ∈F ρ(Tf (X)) = ρ X − f (X) + Π(f (X))
s.t. Π(f (X)) ≤ π

 where ρ is a chosen risk measure, such as variance, VaR and CTE,

 F is the feasible set and often taken as

F = {f : 0 ≤ f (x) ≤ x for all x ≥ 0, both f and Rf are increasing}

• Complexity of solving the above risk measure minimization models:

 If the form of f is specified: technically tractable.
4 stop-loss f (x) = (x − d)+
4 quota-share f (x) = cx
 If a general f is considered: infinite dimensional problem, very challenging to
obtain the explicit solutions.
 Relies on the employed risk measure ρ and premium principle Π.
Chengguo Weng([email protected]) – p. 8/18
• Section I-2.
Risk Measures

Chengguo Weng([email protected]) – p. 9/18

Section I-2. Risk Measures
What is a risk measure?
• For some quantifiable risks we may model potential losses with a statistical
distribution
 Simple parametric model:
4 Normal, Lognormal, Pareto, Compound Poisson
 Functions of simple parametric model:
4 e.g., (ST − K)+ , where K is the strike price and ST denotes a random
stock price following a lognormal distribution
 Complex model:
4 e.g., An investment on a basket of assets (equities, bonds, etc. )
4 e.g., Total loss from operational risk
4 Often Monte Carlo simulation has do be adopted for quantification

Chengguo Weng([email protected]) – p. 10/18

Section I-2. Risk Measures
What is a risk measure? (Cont’d)
• Given the loss distribution of of a risk exposure, the next question we often ask in
RM is:
How much is the firm likely to lose if things go wrong?
• We need one number or at least a small set of numbers to indicate the risk
managers the degree to which the firm is subject to particular aspects of risk.
• Generally speaking, a risk measure ρ is a functional which maps the set of all risks
under consideration to the real line R to quantify potential losses arising from each
risk.

Chengguo Weng([email protected]) – p. 11/18

Section I-2. Risk Measures
Functions of risk measures?
• Calculation of insurance premium: Insurance premiums compensate an insurance
company for bearing the risk of the insured claims. The size of this compensation
can be viewed as a measure of the risk associated with the claims.
• Risk management metrics: Risk measures are important risk management tools in
monitoring and controlling the risk exposure; serve the internal risk management
purposes.
• Determination of risk capital requirement: One of the principal functions of risk
measures in the financial and insurance sectors is to determine the amount of
capital a financial institution needs to hold as a buffer against unexpected future
losses on its portfolio in order to satisfy a regulator, who is concerned with the
solvency of the institution.

Chengguo Weng([email protected]) – p. 12/18

Section I-2. Risk Measures
Functions of risk measures (Cont’d)
• Common risk measures in insurance are premium principles, which are used to
determine the price of an insurance policy:

 Expectation principle.
Π(Z) = (1 + θ)E[Z] with θ > 0.

 Standard deviation principle.

p
Π(Z) = E[Z] + β D[Z], where β > 0 and D[Z] denotes the variance of Z.

 Dutch principle.
Π(Z) = E[Z] + β E(Z − E[Z])+ with 0 < β ≤ 1.

 Exponential principle.
Π(Z) = β1 log E[exp(βZ)] with β > 0.

Chengguo Weng([email protected]) – p. 13/18

Section I-2. Risk Measures
Functions of risk measures (Cont’d)
• The risk measure ρ(L) can be interpreted as the amount of capital that should be
added to a position, say an investment portfolio, with loss given by L, so that the
position becomes acceptable to an external or internal risk controller.
 Position with ρ(L) ≤ 0 are acceptable without injection of capital. If
ρ(L) < 0, capital may even be withdrawn.
 If ρ(L) > 0 for some loss L, we expect ρ(L − ρ(L)) = 0.
 Associated with a risk measure is an acceptable set Aρ which is defined as

Aρ = {X ∈ G : ρ(X) ≤ 0} ,

where G denotes the set of all risks under consideration.

 Given an acceptable set A, we may define a risk measure as
ρA (L) = min{l ∈ R : L − l ∈ A}.

Chengguo Weng([email protected]) – p. 14/18

Section I-2. Risk Measures
VaR and CVaR
• The Value-at-Risk (VaR) of a loss variable Z at a confidence level α (often taken
as 95% or 99%), 0 < α < 1, is defined as

VaRα (Z) = inf{z ∈ R : Pr(Z > z) ≤ 1 − α}

= inf{z ∈ R : Pr(Z ≤ z) ≥ α}.

• The Conditional VaR (CVaR) of a loss variable Z at a confidence level α,

0 < α < 1, is defined as

1 w1
CVaRα (Z) = VaRs (Z)ds.
1−α α

 If Z is a continuous random variable, then CVaR has the following simple

representation:

CVaRα (Z) = E[Z|Z > VaRα (Z)] = E[Z|Z ≥ VaRα (Z)].

• CVaR is the expected loss given that the loss falls in the worst (1 − α)100% part of
the loss distribution. It reflects the magnitude of the risk on the tail.
Chengguo Weng([email protected]) – p. 15/18
Section I-2. Risk Measures
Expectile
• The expectile or α-expectile of a loss variable Z with E[Z 2 ] < ∞ at a confidence
level α ∈ (0, 1), denoted by E(Z; α), is defined as the following minimizer:
h i h i
2 2
E(Z; α) = arg min α E (Z − m)+ + (1 − α) E (m − Z)+ , (1)
m∈R

where (x)+ = max(x, 0).

• It can be shown that a number E(Z; α) ∈ R solves the above optimization problem
if and only if
   
α E (Z − E(Z; α))+ = (1 − α) E (E(Z; α) − Z)+ ,

which is equivalent to
  2α − 1
E(Z; α) = E[Z] + β E (Z − E(Z; α))+ with β = . (2)
1−α

• We take equation (2) as the definition of α-expectile.

• Coherent & Elicitability Gneiting(2011)
Chengguo Weng([email protected]) – p. 16/18
• Section I-3.
Optimal (Re)insurance Models

Chengguo Weng([email protected]) – p. 17/18

Section I-3. Topics in the Module
Topics in the Module
• CVaR Minimization and Extentions
• CDF formulation and Solution under Distortion Risk Measures
• Empirical Reinsurance Models and Applications to Agricultural Sector
• Index Insurance Design
• Optimal Partial Hedging by Minimizing VaR and CVaR

Chengguo Weng([email protected]) – p. 18/18