Actuarial modelling using decrement models involves calculating the probability of an insured individual leaving a specific status—typically "active" or "alive"—due to various competing causes such as death, disability, or policy surrender. These models allow insurers to accurately price products and determine reserve requirements by evaluating how multiple risks interact over the lifetime of a policy.
Understanding the Fundamentals of Decrement Models
In the context of life insurance, a decrement is any event that causes a policy to cease or transitions it to a different state of benefits. While a simple mortality table focuses on a single decrement (death), modern life insurance products are far more complex. Actuarial science uses multiple decrement models to handle the reality that an individual might leave the "insured" pool for several reasons, and the occurrence of one event often precludes or changes the likelihood of another.
By 2026, the integration of high-frequency data and machine learning has refined these models, but the underlying mathematical framework remains rooted in probability theory and survival analysis. The primary goal is to derive the probability that an individual aged $x$ will exit the insured group within a specific time frame due to a specific cause $j$.
The Mathematical Framework of Multiple Decrements
The core of multiple decrement theory lies in the "force of decrement." For each cause $j$, there is an associated force of decrement, denoted as $\mu_x^{(j)}$. This represents the instantaneous rate of transition at age $x$ for cause $j$.
When multiple decrements are at play—such as death, accidental death, and permanent disability—the total force of decrement, $\mu_x^{(\tau)}$, is the sum of all individual forces:
$$\mu_x^{(\tau)} = \sum_{j=1}^m \mu_x^{(j)}$$
From this, actuaries derive the probability of survival from all causes combined, denoted as $p_x^{(\tau)}$. This survival probability is essential for determining the "expected" number of policies that will remain in force at any given time, which directly dictates the premium income the insurer can expect to receive.
Single vs. Multiple Decrement Tables
A common challenge in actuarial modelling is distinguishing between "crude" probabilities and "net" probabilities.
- Associated Single Decrement Tables (ASDT): These models describe a world where only one cause of decrement exists. For example, if we ignore lapses and disability, what is the "pure" probability of death? This is a theoretical construct used to isolate specific risks.
- Multiple Decrement Tables (MDT): These reflect the real-world scenario where all decrements act simultaneously. The probability of an individual dying (decrement 1) is influenced by the fact that they might lapse their policy (decrement 2) before death can occur.
In professional practice, actuaries often start with ASDTs (often provided by national regulatory bodies or reinsurers) and then "operate" on them to create a combined MDT. This process requires careful handling of the "independence" assumption—the idea that the occurrence of one decrement does not change the underlying probability of another. While mathematically convenient, 2026 industry standards emphasize that decrements are often dependent (e.g., a person with declining health is less likely to lapse their policy).
Key Decrements in Life Insurance Products
The complexity of a decrement model depends largely on the product being valued. Different insurance contracts trigger different "exits" from the model.
1. Mortality (Death)
The most fundamental decrement. In 2026, mortality modelling has moved beyond static tables to include longevity trends and the lingering impacts of periodic health crises. Mortality is usually split into "standard" and "accidental" in many life policies because accidental death often triggers a higher payout (double indemnity).
2. Withdrawal and Surrender (Lapse)
Lapses occur when a policyholder stops paying premiums or voluntarily terminates the contract. Unlike mortality, which is biological, lapse rates are behavioral. They are highly sensitive to:
- Economic Conditions: In high-interest-rate environments, policyholders may surrender older policies to seek better returns elsewhere.
- Policy Duration: Lapse rates are typically higher in the first two years of a policy and stabilize thereafter.
- Distribution Channel: Policies sold through direct-to-consumer digital platforms often show different lapse behavior than those sold via traditional brokers.
3. Disability and Morbidity
For riders or standalone health products, the decrement is the onset of a critical illness or total permanent disability (TPD). Modelling this requires "multi-state" models where an individual can move from "Active" to "Disabled" and potentially back to "Active" (recovery), or from "Disabled" to "Dead."
4. Maturity
This is a deterministic decrement. For endowment policies, the model must account for the fact that at a certain age or policy year, the contract ends with a guaranteed payout.
The Role of Decrement Models in Pricing and Reserving
The practical application of these models occurs during the valuation process. Actuaries use decrement probabilities to project future cash flows.
Premium Calculation
To set a premium, the actuary must ensure that the Present Value of Future Premiums (PVFP) equals the Present Value of Future Benefits (PVFB) plus expenses and profit margins.
- If the model underestimates lapses, the insurer might over-calculate the expected premium income.
- If the model underestimates mortality, the insurer will under-price the risk, leading to financial instability.
Reserving under IFRS 17 and Solvency II
As of 2026, international accounting standards like IFRS 17 require insurers to use "best estimate" assumptions for all decrements. This means that models cannot simply be conservative; they must be realistic.
- Best Estimate Liability (BEL): This is calculated using the multiple decrement probabilities to discount expected future claims.
- Risk Adjustment: An additional layer of capital is held to account for the uncertainty in these decrement assumptions.
| Decrement Type | Primary Driver | Impact on Liability | Typical Data Source |
|---|---|---|---|
| Mortality | Biological/Age | High (Payout) | Industry Tables / Internal Experience |
| Lapse | Economic/Behavioral | Moderate (Lost Premium) | Internal Company Data |
| Disability | Health/Occupational | High (Ongoing Payout) | Reinsurer Studies |
| Maturity | Contractual | Fixed (Known Date) | Policy Administration System |
| Expenses | Inflationary | Operational | Internal Accounting |
Data Graduation and Assumption Setting
A model is only as good as its inputs. Actuaries must "graduate" raw data to remove noise and ensure the probabilities follow a logical progression (e.g., mortality should generally increase with age).
Methods of Graduation
- Whittaker-Henderson: A mathematical technique that balances "fit" (how close the curve is to the data) with "smoothness" (how few jagged edges the curve has).
- Parametric Models: Using formulas like Gompertz-Makeham to describe mortality.
- Splines: Flexible polynomial functions used for complex lapse or morbidity curves.
In 2026, the use of "Predictive Analytics" has become standard for assumption setting. Instead of looking solely at historical averages, actuaries use generalized linear models (GLMs) to understand how variables like credit score, geographic location, and even digital engagement levels correlate with lapse and mortality decrements.
Transitioning to Multi-State Models
While multiple decrement models assume a "one-way" trip (once you lapse or die, you are out), multi-state models are a more advanced evolution used for complex products like Long-Term Care (LTC) or Income Protection.
In a multi-state model, an individual can exist in several states:
- Healthy
- Sick/Disabled
- Dead
The model allows for transitions between Healthy and Sick (and back again), while Death remains the "absorbing state." The transition intensities $(\alpha_{ij})$ are the multi-state equivalent of the force of decrement. Modelling these transitions requires solving systems of Kolmogorov differential equations, often performed via stochastic simulation in modern actuarial software.
Common Pitfalls in Decrement Modelling
Even with sophisticated software, several technical errors can compromise the integrity of a life insurance model.
1. Selective Symmetrization (The "Antiselection" Problem)
Antiselection occurs when the individuals who choose to stay in a policy have a different risk profile than those who leave. For example, if a policy becomes very expensive, healthy people might lapse (surrender), leaving only the "unhealthy" individuals in the pool. If the decrement model treats lapse and mortality as entirely independent, it will fail to capture the rising mortality rate of the remaining group.
2. Data Thinness
For older ages (95+) or rare diseases, there is often very little data. Actuaries must use "extrapolation" to extend decrement curves. Errors in extrapolation can lead to significant "tail risk," where the model fails to account for the high cost of the few remaining policyholders.
3. Economic Correlation
Lapse decrements are rarely static. In 2026, many models incorporate "Dynamic Lapse Assumptions." These formulas adjust the lapse rate based on the spread between the policy's guaranteed interest rate and the current market rate. Failing to model this can lead to a "run on the bank" scenario during periods of rapid interest rate hikes.
Computational Implementation and Software
The era of manual spreadsheets for multiple decrement tables is over. In 2026, actuarial platforms such as Milliman Mind, FIS Prophet, and Moody’s AXIS handle these calculations using high-performance computing.
The workflow typically involves:
- Extract-Transform-Load (ETL): Pulling policyholder data from administration systems.
- Experience Analysis: Comparing actual decrements from the past year against what the model predicted (Actual-to-Expected or A/E ratios).
- Assumption Update: Adjusting the $\mu_x^{(j)}$ values based on the A/E results.
- Projection: Running the model over a 50-to-100-year horizon to determine solvency and profitability.
FAQ
What is the difference between a force of decrement and a probability of decrement?
The force of decrement ($\mu$) is an instantaneous rate, similar to a speedometer reading at a specific moment. The probability of decrement ($q$) is the likelihood of the event happening over a discrete interval, such as one year. Actuaries use the force to derive the probability through integration.
Why do actuaries assume decrements are independent?
Independence simplifies the math significantly, allowing actuaries to calculate "total" survival by simply multiplying the survival probabilities of individual decrements. However, modern 2026 standards often require "correlation factors" or multi-state models to account for the reality that health and behavior are often linked.
How does IFRS 17 change how we model decrements?
Under older standards, actuaries often used "prudent" (conservative) assumptions. IFRS 17 requires "best estimate" assumptions with a separate, explicit "Risk Adjustment." This forces models to be more precise and data-driven, as the margin for error must be clearly quantified rather than hidden in the rates.
Can decrement models be used for Group Life insurance?
Yes, but the decrements differ. In Group Life, "termination of employment" is a major decrement that doesn't exist in individual life insurance. The model must account for the turnover rate of the specific company or industry being insured.
What happens if two decrements occur at the same time?
In a continuous-time model, the probability of two decrements occurring at the exact same instant is zero. In discrete-time models (e.g., annual steps), actuaries must use "uniform distribution of deaths" (UDD) or other interpolation assumptions to decide how to allocate decrements that occur within the same year.
Practical Recommendations
For professionals working on life insurance valuations in 2026, it is critical to move beyond static MDTs. Ensure your models incorporate dynamic lapse functions that react to the current economic environment. Furthermore, perform regular Experience Studies at a granular level; aggregating data too broadly can mask "pockets" of risk where specific demographics are experiencing higher-than-expected mortality or disability rates. Finally, validate the independence assumption—if your data shows that lapses decrease as health declines, your model must reflect this dependency to avoid underestimating future claims.
