Using a Statistical Framework and AI Techniques to Enhance Basic Actuarial Assumptions Part 2: Application to Accelerated Underwriting

In Part 1 of this paper, the authors described a statistical model that can be used to combine various risk factors from independent external datasets in order to enhance a base actuarial assumption like mortality or morbidity. There are many practical applications for this methodology and in Part 2 of this paper, the authors will describe in detail how it can be used to enhance an accelerated underwriting model for life insurance.

Motivation

Accelerated underwriting (sometimes denoted by simplified underwriting) is the latest development by life insurers to minimize the time between application for life insurance and issuance of the policy. While accelerated underwriting practices vary between life insurers, they have some common traits:

• Absence of any medical underwriting—blood tests, bodily fluids, etc.
• Process is typically automated and a decision is made to issue a policy based on the policyholder’s responses and any additional information (e.g., motor vehicle violations, prescription drug use, etc.) that an insurer has access to.
• Most accelerated underwriting models have an upper limit (e.g., $1M) on the policy face amount that will be issued and above this face amount, full underwriting is required.

While accelerated underwriting helps in the sales process and an insurer’s competitive position, it does have its risks. If the accelerated underwriting is too rigid, it could eliminate acceptable risks, which could impact sales revenue. On the other hand, if the accelerated underwriting process is too liberal, it could negatively impact claims experience. It is critical, therefore, that the accelerated underwriting algorithm is statistically and actuarially rigorous and minimizes the risk of false rejection or false acceptance of policyholders being underwritten.

Accelerated Underwriting Assumptions

The following are the risk factors and the various risk levels that will be considered in accelerated underwriting that will be used to adjust the base select and ultimate mortality rates.

The choice of risk factors is based on a review of the existing literature that shows that underwriting mortality is impacted by varying degrees by each of these factors. These mortality adjustment factors to the baseline life expectancy (LE) are only available separately for each factor and there is no single dataset that captures the LE impact for combinations of these factors. Thus, the technique described in Part 1 of this paper lends itself nicely to create the accelerated underwriting algorithm.

Value of Adjustment Factors

The risk factor LE adjustment values have been estimated from various sources in the literature. The information has been provided as a factor that is multiplied by the baseline life expectancy for males or females. The baseline LE uses the select-and-ultimate VBT non-smoker mortality rates. The risk factors in the literature are further adjusted to ensure that the predicted aggregate adjustment factor is within reasonable bounds. To illustrate, a risk factor value of 1.05 for a given male risk factor means that with the existence of this risk factor, the male life expectancy will be increased by 1.05. A similar reasoning holds for a risk factor less than 1 that results in a reduction in the base life expectancy.

Once the risk adjustment value and various levels of the risk factors have been established, the technique described in Part 1 of this article can be used to develop the multiple linear regression predictive model. The multiple linear regression model for males and females is as follows:

Aggregate Adjustment Factor on Life Expectancy

=0.9976 + 0.0182 * Marital Status + 0.0525 * Income Level + 0.0029 * Sleep Hours + 0.0039 * Sleep Disturbances + 0.0373 * Number of Chronic Conditions + 0.0249 * Vegetarian + 0.0699 * Food Diversity of + 0.0216 * Drinking + 0.0103 * Occupation + 0.0347 * Education + 0.0700 * Driving History + 0.0154 * Stress + 0.0009 * Prescription Drugs + 0.0349 * Exercise + 0.0196 * BMI + 0.0369 * Family History

Accelerated Underwriting Model Results

For a given combination of risk factors at various levels, the accelerated underwriting model provides the following results:

• The baseline life expectancy (LE) based on age and gender and using the select-and-ultimate VBT table.
• The modified LE using the various risk factors and risk factor levels obtained from the underwriting process.
• The aggregate mortality adjustment to the select-and-ultimate VBT table using the modified LE.
• The standard deviation of the baseline LE.
• The best-case scenario LE and the worst-case scenario LE.

The model output could be used to assist an underwriter in the decision-making process to accept or reject a policyholder. For instance:

• If the modified LE falls within one standard deviation of the baseline LE, the policy should be issued with standard rates.
• If the modified LE exceeds one standard deviation of the baseline LE, the policy should be issued with preferred rates.
• If the modified LE falls below one standard deviation of the baseline LE, the policy should be rejected or be subject to full underwriting.

The actual decision-making process may not necessarily be this prescriptive and will vary by each insurer’s own risk tolerance level. But our accelerated underwriting model does provide a well-established and rigorous decision-making process on how to assess and quantify the mortality risk of a newly underwritten policyholder based on accelerated underwriting.

Illustrations

Consider a new policy that is underwritten for a 30-year-old male non-smoker. Without any additional information, the estimated baseline LE is 53.41 years from the male non-smoker VBT table, and the standard deviation of the baseline LE is 11.15 years.

Using the additional information from the various risk factors, the best-case scenario LE is 70.67 years and the worst-case scenario LE is 15.10 years.

Now consider 2 policyholder scenarios:

Scenario A—Favorable underwriting scenario:

The policyholder is married, has a high-income level, graduate degree, managerial and professional occupation, no stress, no sleep disturbances, clean driving record, no family history, no prescription drugs, moderately active exercise, diversified food diet, long-term vegetarian diet and a BMI of 23.

Based on the accelerated underwriting model, we have the following output:

• The baseline LE for a 30-year-old male non-smoker is 53.41 years.
• The modified LE based on the assumed risk factors is 68.80 years.

Scenario B—Unfavorable underwriting scenario:

The policyholder is widowed, has a medium-income level, high school education, routine occupation, high stress, 3+ traffic violation, family history of heart disease, uses prescription drugs, never exercises, unbalanced food diversity diet, non-vegetarian diet and a BMI of 30.

Based on the accelerated underwriting model, we have the following output:

• The baseline LE for a 30-year-old male non-smoker is 53.41 years.
• The modified LE based on the assumed risk factors is 29.54 years.
  
  

Observations and Conclusions

While this is only one illustration on how our modeling approach can be used to enhance base actuarial assumptions, it is an important application because accelerated underwriting is a growing area of practice by most major insurers. Our modeling approach provides a statistical approach that is transparent, consistent, logical, easy to implement and easy to understand. On an ongoing basis, using the established and structured methodology we have created will allow a company to better track its accelerated underwriting practices and adjust assumptions as needed to better manage mortality experience.

Statements of fact and opinions expressed herein are those of the individual authors and are not necessarily those of the Society of Actuaries, the newsletter editors, or the respective authors’ employers.