Title: Coherent Measures of Risk
1Coherent Measures of Risk
- CAS Seminar on
- Dynamic Financial Analysis
- June 8, 2001
- Glenn Meyers
- Insurance Services Office, Inc.
2New Papers
- Coherent Measures of Risk
- Philippe Artzner, Freddy Delbaen, Jean-Marc Eber
and David Heath, Math. Finance 9 (1999), no. 3,
203-228 - http//www.math.ethz.ch/delbaen/ftp/preprints/Coh
erentMF.pdf - Coherent Measures of Risk - An Explanation for
the Lay Actuary - Glenn Meyers
- http//www.casact.org/pubs/forum/00sforum/meyers/C
oherent20Measures20of20Risk.pdf
3A List of Loss Scenarios
Define a measure of risk r(X) MaximumXi
4Subadditivity
r(XY) ? r(X)r(Y)
5Monotonicity
If X ? Y for each scenario, then r(X) ? r(Y)
6Positive Homogeneity
For all l ? 0 and random loss X, r(lX) lr(Y)
7Translation Invariance
For all random losses X and constants a r(Xa)
r(X) a
8Axioms for Coherent Measures of RiskSatisfied by
our example
- Subadditivity For all random losses X and Y,
- r(XY) ? r(X)r(Y)
- Monotonicity If X ? Y for each scenario, then
- r(X) ? r(Y)
- Positive Homogeneity For all l ? 0 and random
loss X - r(lX) lr(Y)
- Translation Invariance For all random losses X
and constants a - r(Xa) r(X) a
9Value at Risk/Probability of Ruinis not coherent
- violates subadditivity
10Standard Deviation Principle is not coherent -
violates monotonicity
11The Representation Theorem
- Let ? denote a finite set of scenarios.
- Let X be a loss associated with each scenario.
- A risk measure, ?, is coherent if and only if
there exists a family, ?, of probability measures
defined on ? such that
i.e. the maximum of a bunch of generalized
scenarios
12Probability Measures?The Easiest Example
- Let A Ai be the set of one element subsets of
W. Let Xi be the loss for ai.
13Probability Measures?The Next Easiest Example
- Let A Ai be the set of n element subsets of W.
Let Xw be the loss for w?W
14Proposed Measure of RiskTail Value at Risk
Value at Risk
Tail Conditional Expectation Tail Value at Risk
15Tail Value at Risk is the average of all losses
above the Value at Risk
16(No Transcript)
17TVaR and Expected Policyholder Deficit
The appeal of TVaR and EPD is that they both
address the question -- How bad is bad?
18Determine the Amount of Capital
- Decide on a measure of risk
- Tail Value at Risk
- Average of the top 1 of aggregate losses
- Note that the measure of risk is applied to the
insurers entire portfolio of losses. - Capital determined by the risk measure.
- C r(X) - EX
19Step 2Allocate Capital
- How are you going to use allocated capital?
- Use it to set profitability targets.
- How do you allocate capital?
- Any way that leads to correct economic decisions,
i.e. the insurer is better off if you get your
expected profit.
20Better Off?
- Let P Profit and C Capital. Then the insurer
is better off by adding a line/policy if
? Marginal return on new business ? return
on existing business.
21OK - Set targets so that marginal return on
capital equal to insurer return on Capital?
- If risk measure is subadditive then
- Sum of Marginal Capitals is ? Capital
- Will be strictly subadditive without perfect
correlation. - If insurer is doing a good job, strict
subadditivity should be the rule.
22OK - Set targets so that marginal return on
capital equal to insurer return on Capital?
If the insurer expects to make a return, e P/C
then at least some of its operating divisions
must have a return on its marginal capital that
is greater than e. Proof by contradiction If
then
23Ways to Allocate Capital 1
- Gross up marginal capital by a factor to force
allocations to add up. - Economic justification - Long run result of
insurers favoring lines with greatest return on
marginal capital in their underwriting. - Appropriate for stock insurers.
- I use it because it is easy.
24Ways to Allocate Capital 2
- Average marginal capital, where average is taken
over all entry orders. - Shapley Value
- Economic justification - Game theory
- Appropriate for mutual insurers
25Ways to Allocate Capital 3
- Line headed by CEOs kid brother gets the
marginal capital. Gross up all other lines. - Economic justification -
???
26Conclusion on Allocating Capital
- Axioms for coherent measures of risk do not
prescribe a unique allocation method. - Additional economic and/or fairness assumptions
are needed.
27Papers on Coherent Allocation
- Coherent Allocation of Risk Capital
- Michael Denault
- http//www.risklab.ch/ftp/papers/CoherentAllocatio
n.pdf - The Cost of Financing Insurance - V 2.0
- Glenn Meyers
- http//www.casact.org/pubs/forum/01spforum/meyers/
index.htm