Paper: arXiv 2611.18660

Authors: Stefan Gerhold, Friedrich Hubalek, Manuel Schranzhofer

Abstract

For an insurance portfolio modelled by a large number of i.i.d. losses, the average loss converges to its expected value by the law of large numbers, reflecting the diversification effect of pooling risks. Policy cancellations reduce the portfolio size and thereby weaken this effect. We quantify the resulting increase in risk by applying risk measures to the average loss before and after the cancellation of policies. For a portfolio of size \(n\), we show that the impact of cancelling a single contract is of order \(n^{-3/2}\), while the cancellation of a fixed proportion of contracts has an effect of order \(n^{-1/2}\). We further extend the proportional cancellation regime to heterogeneous portfolios with different cancellation rates and risk characteristics across segments. These asymptotic results provide a quantitative basis for assessing surrender fees and cancellation costs from the insurer’s perspective. We complement the theoretical analysis with numerical results for representative life insurance products, illustrating the magnitude of cancellation costs and the effect of portfolio aggregation. The two regimes require different techniques. For the cancellation of a fixed proportion of contracts, we employ central limit theorems in Wasserstein distance. For the cancellation of a single contract, we combine Kantorovich-Rubinstein duality with higher-order distributional approximations based on the Edgeworth and Cornish-Fisher expansions.

Complexity vs Empirical Score

  • Math Complexity: 9.0/10
  • Empirical Rigor: 4.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: This paper presents a highly mathematical and theoretical analysis of policy cancellation risk, employing advanced techniques like Wasserstein distance and Edgeworth expansions. While the theoretical framework is robust and novel, the empirical validation is limited to numerical illustrations rather than extensive backtesting or real-world data analysis, placing it firmly in the ‘Lab Rats’ quadrant. The novelty lies in quantifying the probabilistic effect of portfolio shrinkage on risk.

Research Flowchart

  flowchart TD
    A["Research Goal: Quantify risk increase due to policy cancellations"] --> B{"Key Methodology Steps"};
    B --> C1["Regime 1: Fixed proportion cancellation (order n^-1/2)"];
    B --> C2["Regime 2: Single contract cancellation (order n^-3/2)"];
    C1 --> D1["Method: Central Limit Theorem in Wasserstein distance"];
    C2 --> D2["Method: Kantorovich-Rubinstein duality + Edgeworth/Cornish-Fisher expansions"];
    D1 & D2 --> E["Data/Inputs: i.i.d. losses, Portfolio size 'n', Cancellation rates"];
    E --> F["Computational Processes: Numerical results for life insurance products"];
    F --> G["Key Findings: Asymptotic quantification of cancellation costs, Basis for surrender fees/cancellation costs, Effect of portfolio aggregation illustrated"];

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper develops asymptotic results quantifying how policy cancellations increase portfolio risk and provides numerical illustrations for life‑insurance products.

Source sentence
For a portfolio of size $n$, we show that the impact of cancelling a single contract is of order $n^{-3/2}$, while the cancellation of a fixed proportion of contracts has an effect of order $n^{-1/2}$. We show that for a large class of risk measures — including VaR and TVaR — the impact of cancelling a single contract is of order n−3/2, whereas the cancellation of a fixed proportion of contracts leads to order n−1/2.

Assumes 6

  • Losses of individual policies are independent and identically distributed. model
    Source sentence
    For an insurance portfolio modelled by a large number of i.i.d. losses, the average loss converges to its expected value by the law of large numbers, reflecting the diversification effect of pooling risks.
  • Cancellations are independent of the loss amounts. model
    Source sentence
    Finally, we show that the first-order asymptotics in the proportional cancellation regime remain unchanged when cancellations are modelled stochastically by independent Bernoulli random variables, provided that the cancellations are inde-pendent of the losses (see Theorem 4.8).
  • The loss distribution has finite moments sufficient for Edgeworth expansions. statistical
    Source sentence
    Note that finiteness of some moments, as assumed in Theorem 3.8 and subsequent results, is no severe restriction.
  • Risk measures considered are Lipschitz continuous on Lp for p∈(1,2). model
    Source sentence
    We also discuss the scope and limitations of our approach, including the class of risk measures considered.
Show 2 more
  • Gaussian losses are used as a blueprint for the analysis. model
    Source sentence
    In Section 2, we introduce the basic notation and discuss Gaussian losses, which serve as a blueprint for the subsequent analysis.
  • A constant annual cancellation probability (e.g., 1%, 5%, 10%) is assumed in the numerical examples. data
    Source sentence
    To illustrate the impact of cancellations, we consider a constant annual cancellation probability θ ∈{1%, 5%, 10%} over the term of the contract.

Data

  • Synthetic life‑insurance loss data synthetic unknown
    Source sentence
    We complement the theoretical analysis with numerical results for representative life insurance products, illustrating the magnitude of cancellation costs and the effect of portfolio aggregation. In particular, applying Theorems 4.1 and 4.3 to the homogeneous portfolio setting and Theorems 4.5 and 4.6 to the aggregated portfolio setting, we consider the quantities [ CCR = 1 −θ θ  σ n1/2 1 −(1 −θ)1/2 (1 −θ)1/2 

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • Cancelling a single contract changes the risk measure by an amount of order n⁻³ᐟ². theoretical
    Source sentence
    For a portfolio of size $n$, we show that the impact of cancelling a single contract is of order $n^{-3/2}$, while the cancellation of a fixed proportion of contracts has an effect of order $n^{-1/2}$. We show that for a large class of risk measures — including VaR and TVaR — the impact of cancelling a single contract is of order n−3/2, whereas the cancellation of a fixed proportion of contracts leads to order n−1/2.
  • Cancelling a fixed proportion of contracts changes the risk measure by an amount of order n⁻¹ᐟ². theoretical
    Source sentence
    For a portfolio of size $n$, we show that the impact of cancelling a single contract is of order $n^{-3/2}$, while the cancellation of a fixed proportion of contracts has an effect of order $n^{-1/2}$. We show that for a large class of risk measures — including VaR and TVaR — the impact of cancelling a single contract is of order n−3/2, whereas the cancellation of a fixed proportion of contracts leads to order n−1/2.
  • The cancellation cost per cancelling policyholder is of order n⁻¹ᐟ². theoretical
    Source sentence
    In both cases, the cancellation cost per cancelling policyholder is of order n−1/2.
  • For a new‑business cohort of 50 000 pure endowment contracts, relative cancellation costs are roughly 0.01 % to 1 % of the reserve. other figure from elsewhere in the paper
    Source sentence
    For pure endow-ments, the resulting relative cancellation costs are of the order of 0.01% to 1% of the reserve.
  • For term‑life contracts of the same size, relative cancellation costs range from about 1 % to 100 % of the reserve, depending on contract duration. other
    Source sentence
    For term life insurance, the corresponding relative cancellation costs are sub-stantially larger, reaching values of the order of 1% to 100% of the reserve, depending on the contract duration.

Checked by others

No other paper in the archive has been found to support or contradict these findings. Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-17 · same card as JSON · A language model chose these items: check the source sentence before relying on one.