>From: "Fleming, Kirk" <AEGISLAN/AEGIS/FLEMINK%Aegis@MCIMAIL.COM>
>Subject: Measures of performance and risk
>To: CASNET <casnet@lists.casact.org>
>
>
>
>What do companies typically use as a measure of expected performance and
>level of risk? Also, how do they set their risk tolerance benchmark?
>
At the last DFA Seminar in Seattle, Larry Berger was taking a formal poll
of these questions. I wonder if he could post the results on CASNET.
>Steve Lowe and Jim Stanard coauthored a paper in the Spring 1996 Forum where
>they describe using the Asset/Liability Efficient Frontier (ALEF) together
>with DFA to make decisions. ALEF requires you to make decisions considering
>both risk and performance. For example, if you were selecting between
>different mixes of asset classes that each had the same risk, you would pick
>the one that maximized expected performance.
The risk load formula we use at ISO was derived from ALEF considerations.
The derivation starts by (1) assuming that an insurer will choose its
portfolio through ALEF, and then (2) calculating the risk load that results
when all insurers play the ALEF game.
I feel much more comfortable with (1) than (2). I consider (2) the best
thing to do when there is no good answer.
As I was publishing my original risk load paper, Phil Heckman proved that
my ALEF approach was equivalent to the Rodney Kreps' marginal capital
approach. Note that this result assumes that the needed capital is a
function of the variance. (One example of such a function is the square
root -- yielding the standard deviation.) Although I have not formally
checked this out, I suspect that a similar statement is true for the
various other versions of ALEF.
If I were running an insurance company I would be looking at the actions
that yielded me the best return on marginal surplus.
>
>What do companies use as measures of risk and performance?
>
>Examples of measures of performance might include:
> - return on assets.
> - return on surplus.
>
>Lowe and Stanard list the following as measures of risk:
> - probability of ruin over the next ten years.
> - probability of combined ratio above 110% next year.
> - expected policyholder deficit on current business.
> - probability of suffering a net decline in surplus of 20% or more at the
>end of three years.
> - probability of failing an RBC test at any point in the next five years.
> - probability of a ratings downgrade by AM Best.
> - probability of a combined ratio two points or more worse than the
>industry.
In theory -- these are all different approaches. I.e. one can probably
imagine scenerios where a decision by one standard will be different than a
decision by another standard.
I suspect that in practice, the decisions made by one standard will be be
similar to those made by a REASONABLE other standards. Some work needs to
be done here. Does anybody have something along this line to share?
>
> If companies don't think in risk benchmarking terms, I'd be interested in
>hearing that also.
>
So would I.
>Thanks in advance,
>
>Kirk Fleming
>
>
>--------------------------------------------------------------------------
>2 Message:0002 2
>--------------------------------------------------------------------------
>From: "Fleming, Kirk" <AEGISLAN/AEGIS/FLEMINK%Aegis@MCIMAIL.COM>
>Subject: Measuring risk
>To: CASNET <casnet@lists.casact.org>
>
>
>Another question about solving for risk tolerance benchmarks that companies
>use. I'd appreciate comments on this approach.
>
>Suppose you're looking at asset allocation strategies and you define risk in
>a certain way - - let's say probability of surplus declining 50% on a GAAP
>basis while taking no more risk than a peer group of companies on the same
>basis.
>
> Is it possible to calculate reasonably robust results of the peer group's
>implied probability of surplus declining 50% on a GAAP basis using
>historical statutory asset and liability data as the data source?
This certainly can be done in the case of catastrophes. To do this you
need a catastrophe model and the exposure distribution of your peer group
of companies.
I suspect this kind of analysis can be done with the collective risk model.
The big problem with this model is that most current versions do not
recognize correlation between the lines of business -- which can lead to a
significant understatement of the variablity. The CAS Committee on the
Theory of Risk is funding a project to address this problem and I am
hopeful that the results can be posted in the next few months.
>
>I could see a number of problems with this, for example, the usual
>deficiencies of statutory data, and companies making asset allocation
>decisions using completely different strategies. On the other hand, it's
>doable.
This reminds me of the old joke comparing an actuary with a member of the
Mafia. One can tell you how many people are going to die, and the other
can tell you who. Many have said that insurance company insolvency is due
to incompetent or dishonest management.
An ALEF analysis is interesting, but it will not always tell the whole
story.
>
>Do people have other approaches or is ALEF not considered in asset
>allocation decisions?
>
>Thanks again,
>
>Kirk Fleming
>
Visit the CAS Web Site at http://www.casact.org
===============================================
To subscribe or unsubscribe from CASNET:
Send an e-mail to caslists@lists.casact.org
Type in the body join casnet to subscribe
or leave casnet to unsubscribe.