Tradecraft MIKE FAILOR
Are We Prepared for Stochastic Modeling?
It is not the strongest of the species that survives, nor the most intelligent
that survives. It is the one that is the most adaptable to change.
—Commonly attributed to Charles Darwin
THE EXPANSION OF MODEL-BASED VALUATIONS (MBV) and
corresponding extension of stochastic modeling into life insurance
product lines appears to be inevitable. Even though two of the major regulations driving this trend (Solvency II and VM- 20) are having
implementation setbacks, MBV and stochastic modeling will continue
to proliferate.
While this ongoing trend may result
in more appropriate risk-based valuations, I would like to raise two concerns
as we continue with the transition. First,
we must recognize that actuarial and
statistical modeling skills must evolve
to make the transition from formulaic
to stochastic modeling. Second, companies must have the necessary controls in
place for the ongoing development and
prudent management of their actuarial
systems and models. Companies may not
realize the increasing significance of my
latter concern, but as I explain below,
this mind-set needs to change.
Actuarial and Stochastic
Modeling Skills
We know that our actuarial, statistical,
and technical skill sets must expand to
keep pace with evolving modeling requirements. While I don’t claim to be
an expert on stochastic modeling, I have
developed enough stochastic models to
appreciate some of the theoretical and
practical issues that may arise when
setting out to stochastically model life
insurance.
One of my stochastic modeling experiences involved the creation of a Monte
Carlo modeling system that simulated
30 years of cash flows on large portfolios
of medically underwritten life insur-
ance at the seriatim level. While doing
initial research, our team quickly real-
ized that there were plenty of resources
for stochastically modeling population,
annuitant, and pensioner mortality, but
very little for fully underwritten indi-
vidual life mortality. Without getting into
the details, we had to flesh out numerous
questions pertaining to model design.
Here are a few preliminary questions
that we needed to consider regarding
model construction:
■ ■ What is the best way to segment the
insured mortality business?
■ ■ How should stochastic mortality improvement be modeled?
■ ■ How should stochastic catastrophic
mortality be modeled?
■ ■ Should lapse rates be modeled
stochastically?
■ ■ Which statistical distributions (i.e.,
normal, gamma, etc.) should we
consider for each of the stochastic
variables, and how should the associated distribution parameters be
derived?
■ ■ Which stochastic variables should be
correlated, and how should the cor-
relation matrices be derived?
Other stochastic modeling projects
will generate different questions, and the
answers may depend on such things as
the credibility of the underlying experi-
ence as well as regulatory requirements.
Actuaries bring a lot of product ex-
pertise and analytical talent to the table.
However, our skill sets cannot remain
static. If actuaries are expected to ap-
ply stochastic modeling within their
respective disciplines, they need to be
well-grounded in stochastic modeling
theory and construction. They also need
a thorough understanding of the risk fac-
tors that drive real-world results.
There are varying degrees of stochastic modeling skills among our ranks.
Some actuaries are prepared to meet new
demands for stochastic model-based
approaches while others will need to
develop both the theoretical knowledge
and the ability to apply it within their
lines of business. Moreover, the actuarial profession needs to further commit to
the advancement of stochastic modeling
knowledge and skills beyond the more
established economic modeling (i.e., stochastic mortality).
We need to proactively identify and
acquire these skills in advance of industry demand. Consider, for example, the
spike in demand in the European Union
for modeling actuaries with Solvency II
knowledge. Many insurers struggled to
implement new software systems and to
develop the required actuarial models.
Will U.S. insurers experience a similar
resource crunch when new MBV regulatory frameworks are implemented? Will