Longevity Gets Longer
This report by Longevity Underwriting explores in detail the results of the TwentyFirst life settlement Actual-to- Expected data.
Longevity Holdings is excited to report the details from TwentyFirst’s life settlement Actual-to-Expected report. In aggregate, the Actual-to-Expected ratio for the TwentyFirst (21st) LE business has been 103.5%, 103.8% and 100.8% for the 13-year, 10-year, and 5-year exposure periods analyzed.
In calculating the Actual-to-Expected ratios for the 10- and 13-year periods, we used our 2018 mortality tables and crediting and debiting (see Current Underwriting Methodology ) as if they were in effect at that time.
Introduction
TwentyFirst has been providing life expectancy estimates since 1998, primarily for the life settlement market. The cornerstone technology has been a proprietary software system that allows our underwrit- ers the tools to make rules-based decisions with consistent and reproducible results. The front-end of the software technology involves turning medical record information into structured data through a mix of automation and expert review.
The current process and methodology do not use unmonitored artificial intelligence. Relying on artificial intelligence (AI) to decipher mortality risks from electronic health records without professional review presents a multifaceted set of risks.
Primarily, there’s the concern of data accuracy and quality, as AI models heavily depend on the reliability and completeness of electronic health records, which can be susceptible to errors, omissions, or biases. The potential for over-reliance on AI-generated predictions could lead to complacency among secondary market professionals, diminishing the role of clinical judgment and human expertise. Moreover, the complexity of AI models makes their decision-making processes less transparent, raising issues of interpretability and accountability, and posing challenges in gaining trust from participants in the life settlement market.
In the realm of modern underwriting and risk assessment, the integration of advanced data analytics and expert underwriting assessment is crucial for accurate mortality estimates. Credits and debits are calibrated using advanced data analytic tools to optimize for predicting mortality expectations.
The combined efforts of carefully curated and consistent underwriting decisions along with a properly calibrated mortality model has been yielding accurate results.
This report dives into the data, methods, and Actual-to-Expected results for the TwentyFirst life expectancy (LE) product used in the life settlement market.
Data Set
The comprehensive LE dataset includes information on nearly 100,000 insured seniors and approximately 50,000 observed mortalities. This study is based on a subset of this data focusing on three exposure periods ending June 2023, encompassing a 13-year study, 10-year study and 5-year study. This allows for identifying recent trends in the underlying data.
The scatterplot below shows LEs issued over the past 5 years. Age of the insured at time of underwriting is along the x-axis and life expectancy estimates are along the y-axis. As one would expect, the general relationship shows longer life expectancies for younger insured lives and shorter life expectancy for older in
sureds.
The top edge of the scatterplot data represents the healthiest cohort of life expectancies (LE) for each age.
The bottom edge of the data shows the lowest LE estimates for various ages, representing heavily impaired lives with high mortality ratios. Doctor Reviews were excluded from the scatterplot below. Doctor Reviews are typically completed for late-stage cancer or severe neurological conditions and the LE typically ranges between 24 and 72 months.
The vertical dispersion illustrates the vital importance of underwriting for assessing individual mortality expectations. For example, the LE for an 80-year-old insured may range between 27 and 195 months, a difference of roughly 14 years. The professional judgment from experienced underwriters and doctors provides the vital step of the life expectancy calculation process.
Automated decision engines may misinterpret the nuance of comments found in doctor’s notes and create false positive indications of impairments that should not be considered for additional mortality risk.
Methodology
The base mortality tables used for this study are the proprietary ITM TwentyFirst 2018 Non-Tobacco and the ITM TwentyFirst 2018 Tobacco tables. This mortality table set is currently being used to estimate survival curves for calculating life expectancy estimates and is appropriate for performing actual to expected analysis on this dataset.
There is both historical and projected mortality improvement in the model ranging from 0.0% to 1.5% per year dependent on gender and age. The method for applying underwriting assessments is to exponentiate the mortality factor on the survival curve for each projected year. Additional modifications are made using age-based and durational-based vectors for items such as anti-selection, impairment patterns, lifestyle factors, and cancer decay factors. This is the Current Underwriting Methodology used today for calculating the life expectancy (LE) product.
The TwentyFirst underwriting software tracks over 400 medical conditions with additional variant factors based on activity levels, family history, and height/weight or body mass index. The underwriting process identifies the presence of each condition by reviewing all relevant pages of the medical records to make informed professional determinations of conditions. The results from underwriting may include both debits and credits tied to medical conditions, lifestyle factors, family history and other categories. The majority of medical conditions are debits which increase mortality, which the net effect would be to lower the life expectancy estimate. Roughly 40 indicators are credits, which the net effect would be to lengthen mortality for a longer life expectancy. Some flagged conditions currently drive no credits and debits, but we track them for observation purposes.
The two graphs show the percentage of cases with specific number of debits and credits attributable to an underwriting. The above graph highlights the frequency of debit conditions identified on life expectancy reports (LE) in the last 5 years, which averaged 15.1 debits per case. Some debits are minimal, like prior alcohol misuse and joint replacements, while other debits, like Huntington’s Disease and cirrhosis of the liver, can have material impact on the LE. These debits are occasionally offset by credits when present in the same case
The 21st LE algorithm takes a multivariate approach to adjusting the expected mortality assumption when calculating the life expectancy. This includes offsetting debits with credits when identified during underwriting.
The above graph shows the distribution of credits identified in LEs over the past 5 years, which averaged 4.9 credits per case. Examples of credits are family history of longevity and good exercise tolerance.
The underlying mortality assumptions and rating methodology from underwriting must be well integrated for accurate estimates. This study analyzes the combined results of the mortality table, underwriting decisions, and the calibrated debits and credits used to determine the expected survival curve used to calculate life expectancy (LE) estimates.
Insured deaths are identified by using the CertiDeath service provided by PBI, a Longevity Holdings company. Incurred but not reported (IBNR) has been included on the detailed analysis section of this report. The range of IBNR factors are 3.5%-5.0% which vary based on the state the insured lives when the LE was performed due reporting lags for certain states. For the entire 13-year, 10-year and 5-year datasets, the IBNR factor is 3.7%, 3.8%, and 4.4%, respectively.
The 5-year analysis used the Current Underwriting Methodology that has been in effect since implementation in 2018, otherwise known as the historical A-to-E mortality basis. The Current Underwriting Methodology was developed based on observations that prior tables and credit/debit methodology were predicting higher mortality rates than were actually observed. The 13- and 10-year Actual-to-Expected ratios would be materially lower if calculated using the tables and crediting and debiting methodology in effect at the time of the applicable underwriting. The 10-year and 13-year analysis used the modified A-to-E mortality basis. The 21st LE model used the underwriting assessment unchanged from the original LE date and used the current factors and methods to create the survival curve estimates used for the 10-year and 13-year A-to-E analysis.
Multiple underwritings on a single insured life are present for roughly 30% of the lives in the 21st dataset. The methodology used for this A-to-E study is the fractional method, where the exposure is weighted based on time elapsed since each underwriting. Only 1 actual death is tabulated for each insured life, both for insureds with a single underwriting or multiple underwritings.
Disclaimer: A-to-E results may not be indicative of future results.
Overall Results
The overall Actual-to-Expected ratio for the TwentyFirst (21st) LE business has been 103.5%, 103.8% and 100.8% for the 13-year, 10-year, and 5-year exposure periods analyzed. This includes an incurred but not reported (IBNR) adjustment. Without the IBNR adjustment, the gross A-to-E for same periods are 99.7%, 99.8% and 96.4% respectively. The remainder of graphs will use the A-to-E with IBNR. In calculating
the Actual-to-Expected ratios for the 10- and 13-year periods, we used our 2018 mortality tables and crediting and debiting (the Current Underwriting Methodology) as if they were in effect at that time.
The black line with tick marks indicates the 95% confidence interval around the Actual-to-Expected metric. The more death events in the dataset, the tighter the 95% confidence interval. Some of the underlying detailed A-to-E graphs may show wide 95% confidence intervals due to smaller amounts of deaths in the subset of data.
Detailed Results
The following section will detail the Actual-to-Expected results by underlying categories such as gender, smoker status, age, and mortality factor applied from underwriting assessments.
Male and Female results show similar patterns to overall with results close to expected with 95% confidence intervals inclusive of 100%. Male results have averaged +2% higher A-to-E’s than Females in all periods. Smoker cohort showed A-to-E lower than 100% for all periods, dropping to 95% over the past 5 years. Smokers remain a small portion of the data, only 5% of the LEs underwritten the past 5 years. The 95% confidence interval contains 100% meaning actual mortality results have not yet materially deviated from expected mortality. The non-smoker results mirror the aggregate A-to-E results.
The overall pattern for A-to-E by age shows higher than expected mortality for younger ages less than 75 and results within expectations for ages above 75. The larger width of the confidence intervals seen in ages < 70 and ages > 95 categories indicate less data points than other buckets. This is expected where sparce data exists outside of key ages.
Ages 80 & up are showing consistent results within 98-100% over the 13- and 10-year periods. The 5-year period shows a slight dip of 1.3-3.5% vs. the 10-year period. The confidence interval contains 100% for all ages above age 80 except for the 5-year period of age 90 to 95. The age 90-95 band for 5-year exposure period has an average LE of 3 1⁄2 years, meaning more than half of the LEs have not crossed over the 50% threshold for expected deaths. We are mindful not to make decisions on mortality assumption adjustments until the dataset becomes fully credible. LE providers with longer track records and larger datasets have the advantage of using data to credibly inform their mortality assumptions.
Although ages under 70 are less than 10% of all LEs underwritten, this grouping shows a heighted A-to-E of 122% to 128%. The 95% confidence interval does not contain 100% meaning that there is a statistically significant difference between the observed mortality data and expected mortality rates for all three exposure periods analyzed. Further commentary is included in the conclusion section.
The mortality factor results shown below are consistently within 99-101% for mortality factors between 3 and 30 for the 13-year and 10-year exposure periods. The 5-year period results are showing some volatility which can be expected for low LE count totals. Cohorts with mortality factors above 1.5 have 95% confidence intervals containing 100% meaning the observed data is within acceptable range for expected mortality.
The A-to-E ratio for mortality factors between 1.5 and 3 are showing 105-106% with the bottom range
of the confidence interval at 102%. This indicates expected mortality is outside the 95% confidence interval for the observed data and will be monitored closely. Similarly, but in the opposite direction, the A-to-E ratio for mortality factors under 1.5 is 101.8% and 99.6% for the 13-year & 10-year periods but is only 79.0% for the 5-year period with the upper range of confidence interval at 95.2%. This indicates the model is expecting too many deaths for insureds with credits and/or minimal debits in the past 5 years of LE certificates. This new development will be analyzed and monitored closely.
Conclusions
Overall, the TwentyFirst results continue to deliver consistent overall Actual-to-Expected results closeto 100%.
The 21st LE results in aggregate are 103.5%, 103.8% and 100.8% for the 13-year, 10-year, and 5-year exposure periods analyzed, applying the 2018 mortality tables and credit/debiting methodology. Subcategories demonstrating divergence from expectations will be monitored closely and reviewed for potential model revisions in the future.
The specific areas of focus will be the following cohorts:
• Under age 75
• Ages 85 & up with mortality factors between 3 and 7
• Mortality factors under 1.5
We are seeing more deaths than expected for insureds under age 75 suggesting that our LEs may be too long at the younger ages.
The age 75 & younger cohort represents 21% of LEs written in the past 5 years.
Ages 85 and above with mortality factors between 3 and 7 are showing fewer deaths than expected indicating our LEs may be too short for certain underwriting conditions.
The cohort of age 85+ with mortality factor between 3 & 7 represent about 16% of total LEs written in the past 5 years.
The recent 5-year results for mortality factors under 1.5 indicate fewer actual deaths than expected suggesting our LEs may be too short. This appears to be due to the preferred underwriting cases with mortality factors under 1 driving lower-than-expected results. This cohort represents about 12% of LEs issued in the past 5 years.
TwentyFirst is one of a few life expectancy providers with over 20 years of data and expertise in this market. TwentyFirst’s consistent approach to underwriting along with the data science team’s advanced analytic calibration of debits and credits is a unique solution to a complex problem. The results stand for themselves.
Please contact info@21stLE.com for more information on ordering the 21st LE Product.
ABOUT THE AUTHOR
JOHN LYNCH, ASA, MAAA is a credentialed Actuary with Longevity Holdings focusing on life underwriting and actuarial analysis.
John has a diversified background including over 17 years as an actuary with roles such as pricing, hedging and valuation for life & annuity products at a top 10 life insurer; pricing structured settlement products, universal life with secondary guarantee products and life insurance underwriting research at a top 5 property & casualty insurer; and a lead actuarial developer for a third-party service provider offering custom life insurance illustrations for 10 of the top 20 life insurancecompanies. He holds a B.S. in Mathematics from Villanova University (2005).
John is managing the assumptions for Fasano Associates LE and TwentyFirst Services LE products. John can be reached at jlynch@longevity.inc
ABOUT LONGEVITY UNDERWRITING
Longevity Holdings owns two independent operations providing unique approaches to life underwriting. TwentyFirst Services provides a data analytic centric approach to blending underwriting decisions with calibrated debits to perform mortality projections. Fasano Associates focuses on physician centric medical assessments to apply debits with a proprietary mortality table calibrated from 20 years of life settlement experience. The Actual-to-Expected ratios for both companies consistently are within 98-102% year after year. Learn more at longevity.inc
