Article from

The Financial Reporter

March 2019
Issue 116
PBR Simplified Methods Stochastic Reserve for Principle-Based Reserves (PBR). The
PBR Simplified Methods research team consisted of Steve

Project Strommen, Brian Hartman, Chris Davis, Therese DeWitt and

myself. The project was designed to provide four deliverables:
1. a multi-risk scenario generator (MRSG) for all material
By Mark Birdsall
assumptions; 2. objectivity measure(s) for material best esti-
mate assumptions; 3. an actuarial report detailing the results
of the project; and 4. recommendations for approximations
techniques by product type, together with a sample regulatory
demonstration.

S
The team has provided these deliverables to the SOA. The proj-
ection 2G of Valuation Manual Section 20 (VM-20)
ect report, including a sample regulatory demonstration, and
states the following: “A company may use simplifications,
the MRSG will be available on the SOA website.
approximations and modeling efficiency techniques to
calculate the net premium reserve (NPR), the deterministic
reserve and/or the stochastic reserve required by this section if This article provides a brief description of deliverables 1, 2
the company can demonstrate that the use of such techniques and 4, including testing results with respect to the two recom-
does not understate the reserve by a material amount, and the mended approximation methods for a sample universal life with
expected value of the reserve calculated using simplifications, secondary guarantees (ULSG) product.
approximations and modeling efficiency techniques is not less
than the expected value of the reserve calculated that does not MULTI-RISK SCENARIO GENERATOR
use them.” The MRSG produces actual rates for the equity and interest
risks and actual-to-expected (A/E) factors (also called “actu-
The Society of Actuaries (SOA) funded a research project al-to-tabular factors” or A/T) for the other material risks. The
to explore possible methods of approximating the VM-20 user first creates a risk definition file.

Figure 1
Defining the Risks

MARCH 2019 THE FINANCIAL REPORTER | 25

 PBR Simplified Methods Project

In the upper left-hand corner of Figure 1 (pg. 25), the user sum of the cumulative shocks. The pop and creep scenarios have
starts with the equity and interest risks and adds material risks the same sum of shocks over the first 20 years (240 months). In
determined by sensitivity testing, actuarial judgment or other the pop scenarios, the shocks are higher initially and then grade
methods. For the equity and interest risks, the scenarios are off towards zero. In the creep scenarios, the shocks are the same
generated using the SOA/AAA economic scenario generator in every month.
embedded in the MRSG.
The user has the option to select the binomial distribution or
The probability levels for the shocks to the material risks are to create a user-defined function. In the development of the
indicated in the lower left-hand corner, which reflect extreme
generator, Brian Hartman developed an approach that permits
scenarios that could be used to develop target surplus (e.g., 99.9
the use of the binomial distribution for any material risk with
percent and 0.1 percent). The 84.1 percent and 15.9 percent
a range from zero to one, such as lapse and mortality. The
scenarios could be used to represent moderately adverse scenar-
resulting distribution covers both estimation error and adverse
ios. Note that the user may input respective percentile levels.
deviation, as the valuation manual requires.
“Pop Up (Down)” and “Creep Up (Down)” are different meth-
ods for producing the scenarios at the selected probability levels. A user-defined function could be created when no relevant com-
For some risks, such as lapses, the adverse direction is not always pany or industry experience study is available, but the user has
clear, so deviations in both directions are tested. access to other information, such as credit spreads and default
rates.
The pop and creep scenarios differ in the pattern of shocks that
are used to create them. The PBR Simplified Methods research The choice of single-year versus lifetime scenarios impacts the
team defined the severity of a scenario as being measured by the frequency of change of the A/E factors for the specified risk.

Figure 2
Spread Net of Default Risk

26 | MARCH 2019 THE FINANCIAL REPORTER

 experience could be credibility blended with relevant industry
experience to establish the central estimate assumptions for
a material risk, including consideration of possible trends in
both relevant company and relevant industry experience. The
central estimate assumptions should be updated regularly to
not miss trends and provide a standard of comparison for
the actual base assumptions used in the cash flow projection
model. Four assumption objectivity measures are defined here.

Measure 1: Actual to Expected Ratios for the Material

Assumptions
The central estimate assumptions provide the denominators
for A/E ratios for the material assumptions for ULSG. The
actual modeling assumptions (without margins) used as the
basis for the cash flow projections provide the numerators in
the A/E calculations. If the anticipated experience assumptions
are set equal to the central estimate assumptions, the A/E ratios
for all the material risks equal one. The A/E ratio provides a
measure of the deviation of modeling assumptions to the cen-
tral estimate assumptions for each material risk.

Measures 2, 3: Margin Impact, Percent Statutory

Margin Impact
An anticipated experience reserve is defined as a deterministic
In Figure 2 (pg. 26), note that the spread net of default risk reserve from the cash flow projection model using the antic-
employs a user-defined function that provides a single value for ipated experience assumptions. Similarly, a central estimate
the first 20 projection years rather than varying year by year. assumption reserve is defined for this demonstration as a
After 20 years, the value reverts to the 50th percentile value of deterministic reserve using the central estimate assumptions.
1.02. For this situation, pop and creep scenarios do not apply.
The percentile points shown in Figure 2 for the user-defined The anticipated experience reserve minus the central estimate
function were based on an American Academy of Actuaries assumption reserve for a block of business equals the margin
presentation to the NAIC. Note the graph of this distribution impact. For this purpose, the statutory margin equals the
is shown in the lower right-hand corner of Figure 2. The distri- reported statutory reserve for a block of business minus the
bution of A/E ratios is skewed to the right, as one may expect. anticipated experience reserve. The percent statutory margin
impact equals the margin impact divided by the statutory
OBJECTIVITY OF ASSUMPTIONS margin.
Using the MRSG, the probability distributions of material
assumptions are objectively defined for economic scenarios and If the anticipated experience assumptions equal the central
for mortality and lapses based on company experience studies, estimate assumptions, the margin impact equals $0 and the
relevant industry experience studies, and other relevant infor- percent statutory margin impact equals 0 percent. Otherwise,
mation sources. these measures provide the combined impact of using modeling
assumptions different than the central estimate assumptions.
In measuring assumption objectivity, the concept of central esti- In a sense, these measures can be considered as measuring the
mate assumptions is used. Central estimate assumptions refer to degree to which the margin has been reduced to cover the use
assumptions for the material risks to develop baseline assump- of other modeling assumptions.
tions for modeling those material risks in cash flow projection
models. Where relevant company experience for a material risk Measure 4: Percent Aggregate Reserve Margin Impact
is 100 percent credible, the relevant company experience for that Using the MRSG to produce deterministic scenarios for each
material risk would be the central estimate assumption, includ- material risk at the 84th percentile of the distribution for that
ing consideration of possible trends in the experience. When risk, an aggregate reserve margin is calculated, including a
there is less than 100 percent credibility, the relevant company covariance adjustment, using either a square root formula

MARCH 2019 THE FINANCIAL REPORTER | 27

 PBR Simplified Methods Project

(like the statutory life risk-based capital formula) or a covari- Sensitivity testing at the 88th percentile and the 12th percentile
ance matrix. The percent aggregate margin impact equals the of the economic scenarios demonstrated that enhanced RSM
margin impact divided by the aggregate reserve margin. If the continued to work well as an approximation method as eco-
anticipated experience assumptions equal the central estimate nomic conditions change.
assumptions, the percent aggregate reserve margin impact
equals 0 percent. Otherwise, this measure provides the com- REGULATORY DEMONSTRATION
bined impact of using modeling assumptions different than the When available, see the project report on the SOA website for
central estimate assumptions. As above, this measure quantifies a sample regulatory demonstration for ULSG. The regulatory
the reduction in margin due to using other modeling assump- demonstration for level term would contain similar elements.
tions for one or more material risks.
In the testing for this project, the prevailing CRVM reserves
APPROXIMATION METHODS FOR THE for par whole life and accumulation UL were much higher than
VM-20 STOCHASTIC RESERVE any of the modeled reserves calculated. It should be noted that
Four products were tested for this project: level term, ULSG, modeled reserves reflect the level of expected profitability of
accumulation UL, and par whole life. Using the MRSG, the the respective products; the higher the expected profitability
research team developed what we called “fully stochastic scenar- of a product, the lower the modeled reserves. The regulatory
ios,” with all material risks varying according to the distributions demonstrations for these products simply need to provide
defined in the risk definition file. The regulatory guardrail evidence that the modeled reserves are less than the CRVM
against mortality improvement was observed. The conditional reserve. The goal is to minimize the extra work of PBR while
tail expectation (CTE) 70 reserve based on 1,000 fully stochastic still providing useful information for a company’s risk analysis.
scenarios was the standard of comparison for the approximation
methods. Two methods of approximating the PBR stochas- OTHER USES
tic reserve for these products provided useful estimates that The PBR Simplified Methods Project has resulted in the
remained stable over three successive valuation years. development of new tools that can be used not only for approx-
imation methods for the VM-20 stochastic reserves but also for
Method 1: Limited Number of Fully Stochastic other purposes.
Reserves plus CTE 70 Standard Deviation
The CTE 70 standard deviation is based on a 2005 article by VM-20 requires the development of margins for all material
John Manistre and Geoffrey Hancock in the North American assumptions in both the deterministic and stochastic reserve
Actuarial Journal titled “Variance of the CTE Estimator.” We calculations. The MRSG could be a useful tool for the objective
refer to the sum of the CTE 70 stochastic reserve and the CTE development of individual margins that reflect the amount of
error adjustment (i.e., standard deviation) as the “adjusted sto- relevant experience underlying the specific material assump-
chastic reserve.” For all four products, the adjusted stochastic tions. The process of calculating an aggregate margin reflecting
reserve was greater than the CTE 70 stochastic reserve based the covariance of the material risks can be used to calibrate the
on 1,000 fully stochastic scenarios for each of the numbers of individual margins to avoid the stacking problem of just adding
scenarios run (30, 50, 100 and 200) and for each of the three up individual margins.
successive valuation dates tested. The degree of conservatism in
the adjusted stochastic reserve over the stochastic reserve based Migrating asset adequacy analysis into a similar multi-risk
on 1,000 scenarios varied based on the number of scenarios. modeling structure using the MRSG and assumption objectiv-
ity measures would improve the analysis of company risk and
Method 2: ULSG Central Estimate Reserve Plus the consistency of reserve measures among blocks of business.
Aggregate Margin at the 88th Percentile (Enhanced Since the material risks in the MRSG are user-defined (except
RSM) for economic scenarios), the generator could be used for the
The enhanced representative scenarios method (RSM) is a asset adequacy analysis of all long-tailed lines of business.
method to approximate the PBR stochastic reserve by projecting Ranking of insurance risks could then be accomplished on a
cash flows using separate deterministic scenarios for each mate- legal entity basis.
rial risk at a selected probability level (e.g., 88th percentile) to
calculate an aggregate margin, which is then added to the central Many smaller companies use a percentage of RBC as a proxy
estimate reserve. For ULSG at December 2016, the aggregate for target surplus that is needed for pricing and capital alloca-
margin was about 157.8 percent of the central estimate reserve. tion purposes. There is no theoretical meaning to a multiple of
This result reflects the number of material risks associated with RBC but may simply represent a rule of thumb with respect to
ULSG, as well as the degree of variability in those risks. rating agency requirements to achieve certain ratings. With the

28 | MARCH 2019 THE FINANCIAL REPORTER

MRSG, extremely adverse scenarios for each material risk can CONCLUSION
be selected and the results combined to produce a target sur- For level term and ULSG, VM-20 defines an NPR that is
plus level that reflects the company’s specific risks rather than generally lower than the prior CRVM reserve for those prod-
an industry-wide average. Free surplus can also be computed ucts. For both products, reduced numbers of fully stochastic
consistently, and due to the excessive levels of statutory reserves scenarios produced a CTE 70 reserve that, when augmented
for some products, the value of free surplus may be larger than by the standard deviation of the CTE estimator, produced
previously thought. a good approximation of the CTE 70 reserve based on 1,000
fully stochastic scenarios. In addition, the enhanced RSM at the
GAAP reserve requirements currently require the addition of 88th percentile produced good approximations for the CTE 70
provisions for adverse deviations. The MRSG could be used reserve based on 1,000 scenarios.
to produce those margins on an objective basis at a probability
level consistent with GAAP. For the par whole life and accumulation UL products, the
current definition of CRVM serves as the NPR. This reserve
The VM-20 stochastic exclusion test (SET) is focused on eco- level does not reflect company experience and, for the prod-
nomic scenarios for interest rate and equity risk. The SET with uct designs tested and the assumptions used in the cash flow
a threshold of 6 percent may not differentiate well between projections, far exceeds the level of statutory conservatism
products with different risk profiles. It may be that the MRSG targeted by the NAIC’s Life Actuarial Task Force. For this sit-
could be used to refine the SET as part of the NAIC’s feedback uation, the incremental work to calculate the modeled reserves
loop. Regardless, a company could run tests using the MRSG for PBR may not be useful. However, using the tools developed
to determine whether stochastic reserves should be calculated, in this research project, enhanced work may be accomplished
either as part of the PBR reserves or as a part of asset adequacy in asset adequacy analysis that could set the stage for future
analysis. PBR developments, such as revised NPR calculations for these
two product types. As noted, less-profitable products would
The company’s use of the assumption objectivity measures is produce modeled reserves that could be higher than the NPR
entirely voluntary. In situations where reviewers (such as inde- (or CRVM) floor.
pendent auditors, regulators or other interested third parties)
may need assurance that assumptions used are appropriately set,
the voluntary submission of assumption objectivity measures
could help minimize time-consuming communications and Mark Birdsall, FSA, MAAA, FCA, MBA, is vice president
at Lewis & Ellis. He can be reached at mbirdsall@
increase trust for both current and future projects. Rate increase lewisellis.com.
filings for long-term care and state examinations of PBR are two
areas that could benefit from using this approach.

MARCH 2019 THE FINANCIAL REPORTER | 29