Risk Loads for Insurers

The focus of prior discussion has been 5 methods of setting risk loads, and the pros and cons of each. Two additional
observations:
All of the methods are more similar than different, and if we use a common set of assumptions, theyre nearly
equivalent
None of the methods resolve several fundamental problems, and any risk loads derived from these methods must
contain a fair amount of subjectivity
Variance, Standard Deviation, Ruin Theory, and CAPM describe similar concepts. The other two methods (utility theory and
reinsurance method) have evident that they are also strongly related to the others.
Kreps equation for surplus supporting insurance variability of a given portfolio (assumed here to be the industry portfolio):
=
V = Surplus
S = Std Dev of the portfolio

z = standard normal percentile value associated with a given probability of ruin

R = Return in Dollars

This is actually produced by a ruin theory equation:

Pr + > + + + <
L = Loss (RV)
E = Expense
e = threshold probability of ruin corresponding to z
=
We standardize, to get
=

The marginal surplus required for a new risk (x) is:

=
Where the primes are as noted before, for the portfolio with x added.
Kreps solves for ( ):
= 2 + /( + )
= standard deviation of x
C = correlation coefficient of x and L
This last equation shows the increase in standard deviation if z is the last risk added to the portfolio (the formula is order
dependent). Measuring each risks surplus requirement based only on its marginal contribution will underestimate the total
surplus need of the portfolio.
We could call this a Ruin Theory approach, as ruin theory is the base of the equations. We can relate this method to the
others.
If we assume that x and L are independent, so C=0, and: =
variance of the new risk.

2
+

2
2

, so the marginal surplus is a function of the

Conversely, assume that x and L are completely dependent, so C=1, and: =

function of the standard deviation of the new risk.

2+
+

, so the marginal surplus is a

The last steps in both of these equations are valid because the standard deviation is small (relative to S).
Bault

Risk Load For Insurers

Page 1 of 4

In the most common situation, x is slightly correlated with L. In these cases, the marginal surplus will be a linear
combination of the variance and standard deviation related to the covariance.
Two important results:
1) The distinction between variance and standard deviation methods is somewhat artificial. Which method to use is a
function of the correlation between the new and existing risks, and the correct answer is a marginal risk approach
that incorporates the covariance
2) Marginal risk methods (including the variance and standard deviation as special cases) are closely related to a ruin
theory approach.
Let P be the premium associated with the industry portfolio, and p be the premium associated with the new risk. Then,
( )/ is the marginal contribution to the standard deviation of the return on premium from the new risks.
Using = 2 + /( + ) from above, and some algebra, we can show that


,

This result looks like a CAPM beta, and in fact, it is the formula proposed for beta by Feldblum.
If we set =

, there might be some disagreement, as this essentially stares that the variance in profit equals the variance

in loss. There are important sources of risk that arent derived from the loss distribution (inflation, investment, default,
parameter risk), which do need to be measured, but measuring them doesnt require measuring the variance of profits
directly. As long as x and L reflect these additional sources of risk, its appropriate to use as defined.
Covariance and ruin theory approaches usually dont reflect these additional sources of risk. Adjusting loss distributions to
reflect these sources is non-trivial. However, for most lines of business, why wouldnt the loss distribution be a reasonable
first approximation for measuring the variance of profits? Most of the components have very low variance compared to
losses. Parameter variance must be included, but its not clear that the best way to do this is by measuring calendar year
variance of profits directly.
The Return on Equity Equation:
= +

the target return on equity for risk x (and all risks)

the risk free rate of return obtained on the supporting surplus
the return on premium
the premium to surplus (or leverage) ratio appropriate for x, as long as

is the appropriate leverage ratio for the

industry.
We can verify this via the standard CAPM equation = + . We can solve for the return on premium by
subtracting the risk free rate and dividing by the portfolio leverage ratio to get:
=

This is equivalent to the ROE equation above, provided that means the same thing in both equations. The ROE equation
and the standard CAPM equation are two different approaches to the same question: How do we determine the needed profit
load?
Under CAPM:
Under the ROE Equation:

Each risk requires a different rate of return, but has a common leverage ratio
Target a common rate of return, but vary the leverage requirement, essentially scaling all
lines to the market return

Well call the ROE Equation the leverage ratio approach from here on out
Bault

Risk Load For Insurers

Page 2 of 4

Under CAPM, the industry leverage ratio is used for all lines of business, where under the leverage ratio approach, the
leverage varies by lines. If we adjust the surplus requirements by beta, then we can accept an equal rate of return on equity
across all lines. This return will equal the industry rate.
Another important result:
3) CAPM and the leverage approach, which are based upon the covariance method, are equivalent for computing
insurance profit loads.
If we allocate surplus in proportion to a lines beta, we obtain a profit load rule thats equivalent to the one produced by
CAPM. We also normalize the ROEs towards the industry average, letting us target a single ROE for all lines. We can get
the same answers by not allocating surplus, using the industry leverage for all lines, and varying the target ROE according to
CAPM.
According to CAPM and the leverage approach, the appropriate industry surplus is , but the first methods discussed state
that its actually . There is no adjustment that brings these approaches in line. Instead, we need to stop viewing
solvency as binary. We do this by adopting a more aggressive ruin constraint. If we require that the sum of loss and expense
minus profit may not exceed premium plus surplus, the needed surplus would now be , as indicated by the CAPM and
leverage approach. Essentially, this changes the ruin theory approaches to say that simply surviving isnt good enough.
The two final approaches are utility theory and the reinsurance method, neither of which has straightforward equations to
compare to the previous results.
The reinsurance method isnt even an independent method, its more of a check. Reinsurers are subject to the same market
forces as primary insurers, and assuming that marginal risk methods as cored for primary insurers, they should also work for
the reinsurers. This should help all insurers calibrate their estimates from other approaches. The problem with this method
is that it cant determine risk loads from scratch.
Utility theory is slightly more complicated. Theres no good way to determine exactly what utility function should be used
for determining investor preferences. CAPM requires that investors have utility functions of a certain form. If we consider
all of the assumptions required by CAPM except for this, and assume that the investors will value per the CAPM formula, it
implies that investors are risk-averse, because they demand higher returns for taking on more risk.
We could better price the risk if we had more information about the market utility function. We could come up with
functions so complex that CAPM no longer applies, but these functions would probably fit into the framework of something
more complex, like the Arbitrage Pricing Model, which is a generalization of CAPM. APM allows investors to use
information other than mean and variance statistics to price risk. The potential shift in approach:
Moment Based
CAPM
Simple Utility Functions
Simple Ruin Theory

Moments + other data

APM
Complex Utility Functions
Complex Ruin Theory

Simple = Tractable & Understandable

Tractability is less of an issue, as computing power increases, and research is more complete. Understandability could slow
progress slightly more.
Utility Function Results:
1) For a fairly large class of tractable utility functions, there is consistency with CAPM and related methods, so its not
important that we dont have a method to determine what utility function to use
2) Even if more complex utility functions might model market preferences more accurately, there are probably other
equivalent methods that would be used in practice.
We now have a single approach for setting profit loads that is objective, agrees with financial theory, and could be used in
practice. However, NONE of the five approaches deal with a few remaining issues.
Issue #1: What is the Industry Leverage Ratio?

is not known for the industry, and we cant rely on any given year end snapshot of equity to be free of distortion and
random fluctuations. We must select it, giving consideration to the amount of risk the market and company management are

Bault

Risk Load For Insurers

Page 3 of 4

willing to bear. The most CAPM can accomplish is to compute profit load relativities. CAPM doesnt require us to know
leverage ratios by line, but it does require that we know the overall leverage ratio (or the leverage ratio for one line).
Issue 2: Why Industry Leverage Over Company Leverage?
Feldblum says a small or moderate size insurer needs a slightly larger risk load than that indicated by the industry-wide
experience to pick up some of the specific risk. The assumption is that the risk of insolvency is higher, and thus the small
insurer offers a lower quality product and thus demands a lower premium.
Bault says that while an insurer may possess additional risks versus other companies, theres no clear reason why an insured
would pay for it. Shouldnt equivalent lines of business demand equal risk premiums in a competitive market? Using a
market equilibrium approach, shouldnt insureds pay only the competitive equilibrium risk charge for all interacting
companies? This means that the beta should be measured against the market return, as opposed to a companys overall
return.
In view of overall concerns for asset/liability management, shouldnt we measure risk against the entire market? The extent
to which the investment portfolio doesnt interact efficiently with the insurers underwriting book is another risk for which
insurers might not be willing to pay.
From a stockholders perspective, asset risk is an important component of an insurers beta, but these forms of investments
should pay their own returns, and the charges should not be charged back to the policyholder. Only the risks that arise from
the interaction of investment and underwriting that cant be diversified away should be included in the insurance risk loads.
If we measure risk against a company portfolio only, its possible that individual transactions could unduly influence the risk
calculation. Using a larger market base forces the risk measurement of individual contracts closer to the margin, which
equalizes risk charges, and better satisfies CAPM assumptions.
The CAPM Methodology has not addressed these issues.
Issue #3: How to Compute Covariance?
CAPM is a theory that applies specifically to financial securities, so we need to make some assumptions in order to adapt the
approach to measuring insurance risk. Empirical profit information is not the best starting point for this calculation. Theres
at least 2 problems that we need to solve:
1) Adjusting for reserve deficiencies and redundancies
2) Using discounted cash flows to allocated investment income.
Theres actually a third problem. In order to estimate by-line beta factors, well need historical operating ratios by line and in
total. However, we want an estimate of the current beta. One would assume this requires that our data be at current level, an
adjustment that involves much more than just rate changes and trends.
Bringing the data to current level has the effect of reducing the variance of historical loss ratios that resulted from shifting
conditions. These shifts reflect legitimate risks to the company and should be included in the cost of capital. By bringing
data to current level, obtains a good measure of process risk, but it does eliminate valid sources of parameter risk that
somehow must be measured and included.
Some of this perceived risk cant be passed on to insureds, like the risk due to deliberate under-pricing.
Instead of starting with calendar data, we can begin with a model of current accident year losses, adjusted for all current
conditions, and some measure of parameter risk. Models of this form probably already exist for pricing and/or reserving
purposes. We must also include a measure of this distributions covariance to the market. However, this is hard to get at, and
in most cases, we must ignore the covariance terms, and instead use a more practical simplification.
Its a choice between simplifying assumptions:
CAPM with calendar year data adjusted top-down as best as possible
Bottom Up with an accident year model, reflecting as many sources of risk as possible.
BOTH are approximations, and need more research.

Bault

Risk Load For Insurers

Page 4 of 4