Some traditional one-way pricing methods do not factor in effects of correlations in the data leading to
mispricing of relevant risks.
Not to be
confused with General Linear Models, Generalized Linear Models utilizes a
Maximum Likelihood estimation method and performs better with larger samples.
Generalized Linear Model is a predictive modeling techniques that can derive a correlation between existing modeled events (Mortality, surrender value) with additional factors identified (age, location, & etc) and provides a better understanding of pricing based on a segmented basis (pricing returns via area).
MULTI-DIMENSIONAL MODELS
One-way mortality analysis methods has the tendency to exclude key factors causing premiums charged to disregard inherent risks in the pool of policyholders.
GLM allows insurers to analyze few factors simultaneously. By including
more variables and capturing the correlations between them, insurers are able
to quantify the actual risk levels of the modeled events.
The
factors can range from internal data (customers’ age and location) to external
data (investment markets & geographical lifestyles).
MORTALITY ANALYSIS
There are concerns on time trends in the data of some mortality analysis when there
is a need to consider and quantify overall mortality improvements assumptions
GLM can
utilize a “calendar year of exposure” as a factor allowing more years of data
to be captured producing a more detailed trend. An example would be an
interaction between healthcare improvements and calendar year of exposure.
RETENTION ANALYSIS
With the
ability to account for large number of factors, GLM are also able to identify
new key factors affecting policy lapses and surrender decisions.
By
providing a correlation in the data and interaction between factors, insurers
are able to identify effects accurately and provide insight on correlations.
Examples are products duration against different agents or different commission
structures.
We will look at the results of one-way analysis and its effectiveness in detail in subsequent postings.
