Measuring the Longevity Tail of the Pension Gap: A Longevity-at-Risk Framework for Annuity, Pension, and Risk-Transfer Markets

Technical Summary: Measuring the Longevity Tail of the Pension Gap

Problem Statement
The paper addresses a critical gap in pension and annuity risk management: the inability to measure and backtest the "longevity tail"—the risk of unexpectedly rapid mortality improvement. While standard mortality models (e.g., Lee–Carter, Cairns–Blake–Dowd) effectively project central mortality paths, they do not provide a compact, comparable, or liability-linked metric for adverse tail outcomes. Rising longevity widens the pensions gap not merely by increasing expected lifespans, but by introducing uncertainty regarding the rate of improvement. For annuity writers and defined-benefit plans, the adverse risk is not lower-than-expected mortality, but rather survival probabilities rising faster than assumed, thereby extending payment durations and inflating the present value of liabilities. Practitioners lack a tool to convert demographic tail risk into economic stress metrics suitable for pricing, solvency testing, and risk-transfer governance.

Methodology
The paper develops a Longevity-at-Risk (LaR) framework, defined as the upper-tail quantile of mortality improvement for a specific age, country, and forecast horizon. The methodology proceeds through four distinct stages:

1. Mortality Improvement Measurement: Using data from the Human Mortality Database (HMD) for eight developed economies (Canada, US, UK, France, Germany, Netherlands, Sweden, Japan) for ages 50–100, the study calculates mortality improvement ($MI$) as the negative log change in central death rates over horizons of 1, 5, 10, and 20 years.
2. Quantile Modeling (Composite Quantile LaR): Instead of modeling the conditional mean of mortality rates, the author employs Composite Quantile Regression to estimate the conditional distribution of mortality improvement directly. The model utilizes a rich set of predictors including:
   • Lagged domestic mortality improvement (1 and 2 periods).
   • Age, period, country, and cohort effects (grouped by decade).
   • Reference-population variables: A leave-one-country-out exposure-weighted mortality improvement index and a reference-population stress indicator.
   • Demographic state variables: Lagged life expectancy at age 65, old-age dependency ratio, and exposure-to-risk.
   • Volatility indicators: Five-year rolling standard deviations of domestic and reference mortality improvement.
3. Liability Translation: The estimated mortality improvement quantiles are converted into adverse mortality scenarios. These scenarios are then translated into Liability-based LaR, which measures the increase in annuity present value (PV) relative to a baseline. This step links demographic tail risk directly to balance-sheet stress.
4. Conditional Longevity-at-Risk (CoLaR): The framework introduces CoLaR to measure domestic tail risk conditional on cross-population stress. This captures the amplification of domestic longevity risk when a reference population is already in a state of adverse mortality improvement, acknowledging shared dynamics across developed economies.

Key Results
The study validates the framework using rolling-origin out-of-sample testing on a common seven-country sample (excluding Germany for the benchmark comparison due to data constraints in the specific benchmark models).

• Superior Calibration: The Composite Quantile LaR model significantly outperforms formal Lee–Carter and Cairns–Blake–Dowd benchmarks in tail calibration.
  • Coverage Error: LaR achieved an average absolute coverage error of 0.0235, compared to 0.1196 for Lee–Carter and 0.2124 for Cairns–Blake–Dowd.
  • Pinball Loss: LaR recorded the lowest average pinball loss (0.0070), indicating superior quantile forecast accuracy compared to the benchmarks (0.0161 and 0.0200, respectively).
• Determinants of Tail Risk: The regression results indicate that the drivers of adverse mortality improvement differ from central trends. Lagged mortality improvement shows persistence at longer horizons (5–20 years), and domestic volatility is a significant positive predictor of upper-tail outcomes. Reference-population stress significantly amplifies domestic tail risk at 5- and 10-year horizons.
• Liability Stress: The translation of LaR into annuity present value reveals economically meaningful stress. At the 95th percentile, mean PV stress rises from 2.11% at a 1-year horizon to 7.00% at a 10-year horizon. At the 99th percentile, stress increases from 2.80% to 7.93%. The model produces prudently conservative stress estimates, with exceedance rates generally below expected levels, particularly at longer horizons. Note that while the full analysis includes 20-year horizons, the formal benchmark comparison in Table 4 is restricted to 1, 5, and 10-year horizons due to the shorter realized validation period for the 20-year horizon.

Significance and Contributions
The paper makes three primary contributions to the literature on pensions and longevity risk:

1. Reframing the Pensions Gap: It treats the pensions gap not only as a savings shortfall but as a measurable tail-risk exposure embedded in lifetime-income guarantees. It shifts the focus from forecasting the "expected path" of mortality to measuring the "adverse tail" of mortality improvement.
2. Development of LaR and CoLaR: It introduces Longevity-at-Risk as a transparent, backtestable metric for mortality-improvement tail risk. Unlike existing quantile mortality models that focus on distributional estimation, this framework specifically targets the risk variable relevant to liability management. It further extends this with Conditional Longevity-at-Risk (CoLaR), providing a mechanism to assess how cross-population stress amplifies domestic tail risk.
3. Liability-Linked Risk Management: The paper demonstrates how to translate demographic tail quantiles into annuity present-value stress. This provides a practical tool for annuity pricing, pension risk transfer, longevity reinsurance, solvency stress testing, and risk-appetite governance. By linking demographic uncertainty directly to economic exposures, the framework offers a standardized stress metric that can be monitored over time to trigger pricing reviews or capital adjustments.

The author concludes that while standard mortality models remain indispensable for central forecasting, a liability-linked LaR framework offers a more suitable tool for managing the specific tail risks that drive annuity solvency and pension funding requirements. The approach complements existing stochastic mortality models by providing the actuarial tail-risk interpretation necessary for insurance-based retirement solutions.