Many life insurers are combining traditional policy information with third-party data to improve the underwriting process, increase operational efficiency, and gain a better understanding of historical mortality, morbidity, and lapsation experience.
The vast number of third-party datasets available makes it challenging to quantify the benefits of integrating the information into an insurance context. Doing so requires a framework that supports analytical and actuarial evaluations in tandem.
Analytical evaluations measure the strength of the statistical relationship between the traditional policy data and the third-party dataset. However, analytical evaluations do not capture the costs vs. benefits of integrating the data in an existing pricing basis or other actuarial contexts. Actuarial evaluation also provides the added insights necessary to quantify the additional benefits of third-party data.
About Third-Party Data
Traditional life insurance data includes age, gender, smoking status, sum assured, planned premium, policy year, and risk class. These types of data commonly influence actuarial assumptions about mortality and lapse potential and implicitly, the assumptions about the health and wealth of the individual.
Third-party data often provides variables with a more explicit relationship to health and wealth. Common types of third-party information are credit data, marketing data, electronic health and laboratory records, and health insurance claims. Behavioral data, such as criminal and motor vehicle records, also can be helpful.
Health-related data – such as pharmacy claims, medical records, and laboratory test results – allow for accelerated underwriting of individual life policies, i.e., underwriting policies without traditional, insurer-ordered paramedical exams. There is growing use of such data in group life insurance as well - click here to learn more.
Wealth-related data, such as credit variables (e.g., late payments) and marketing data (e.g., savings rate), provide more expansive views of wealth than sum assured and plan premium. Data analysis has also shown a correlation between credit data and mortality outcomes.
See also: The Power of Big Data: An RGA Case Study on Credit-Based Data and Accelerated Underwriting
Analytical evaluation of third-party data is critical to establishing a strong foundation for reliable assumptions. To quantify the strength of individual relationships among third-party data, existing risk factors, and analytical evaluations, actuaries and data scientists often use correlation measures. Questions such measures answer are:
- Do existing risk factors correlate to third-party data?
- How strong is the relationship between variables?
- Are the extreme values – whether very low or very high – in the third-party data more informative?
Traditional risk factors, such as underwriting class for mortality or sum assured and policy year for lapse, deserve examination relative to the third-party variables. For example, does disclosed smoker status correlate to a third-party smoker propensity variable? Non-obvious correlations should be examined as well, such as relationships between smoker propensity and sum assured.
When computing correlations, consider both stratified and non-stratified measures. For example, a third-party variable may have an uneven distribution (as shown in Figure 1), with most individuals having average scores and a few located at the extremes.
Figure 1: Smoker Propensity Distribution
A stratified sample (as shown in Figure 2) evens out the distribution. If the correlation increases, it means that extreme values are more informative.
Figure 2: Smoker Propensity Distribution (stratified)
Actuarial evaluations quantify the total impact of using third-party data on existing pricing and risk segmentation. The key question is: What is the potential gain from integrating this data into the expected basis? It is also important to ask if the information is sufficient to justify using specific third-party data.
For actuarial evaluations, lift charts are helpful because they reveal trends for segmenting mortality, morbidity, or longevity across the spectrum of third-party data attributes.
To build a lift chart, start with an existing actuarial experience study using typical fields such as the expected benefits the insurance company expects to pay. If it is a mortality study, use the number of deaths and append the third-party data attributes. Returning to the example of smoker propensity, actuaries can calculate the sum of actual deaths over the sum of expected deaths within each level. The more upward the line is in a lift chart, the greater the predictive value.
Figure 3: Smoker Propensity Lift (shows 5x lift)
Lift charts are useful for models built using multiple third-party variables. For example, a model that factors in the propensity to smoke with socioeconomic and behavioral factors can be used to predict the actual number of deaths or morbidity outcomes, such as incidence rates by each dataset. The evaluation can be especially helpful for underwriting group life insurance. After conducting in-force and experience analysis, actuaries can move on to integrating the results into the pricing for risk segmentation.
Combining different types of third-party data with traditional insurance information through predictive models can provide deeper insight to improve the underwriting process, enhance operational efficiency, and gain a better understanding of historical mortality, morbidity, and lapsation experience. The more thorough the third-party data evaluation, the greater its value for helping insurers improve operational efficiency and customer experience.