Some continuity estimates for ruin probability and other ruin-related quantities

Basic Information

• Paper ID: 2511.12218
• Title: Some continuity estimates for ruin probability and other ruin-related quantities
• Author: Lazaros Kanellopoulos (University of the Aegean, Greece)
• Classification: math.PR (Probability Theory)
• Publication Date: November 18, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2511.12218

Abstract

This paper investigates continuity properties of ruin probability in the classical risk model. Continuity estimates for the deficit at ruin are derived using properties of contraction integral operators. These results are also applied to continuous-time surplus processes with diffusion perturbations, where the ruin probability can be expressed as a convolution of the compound geometric distribution K with the diffusion term. The paper derives continuity inequalities for K and proposes an iterative approximation method for ruin-related quantities, with results verified through numerical examples.

Research Background and Motivation

Core Problems

1. Stability Problem: In actuarial mathematics, actual risk parameters (such as claim arrival intensity λ and claim amount distribution F) are typically unknown and must be estimated using approximate parameters. How can one measure the impact of parameter perturbations on ruin probability and other output quantities?
2. Continuity Estimates: Seeking inequalities of the form ν(ψ_α, ψ_α̃) ≤ w(δ(α, α̃)), where δ measures the distance between input parameters and ν measures the distance between output functions.
3. Diffusion Perturbation Model: For risk models incorporating diffusion terms, existing literature lacks continuity analysis using probabilistic metrics.

Research Significance

• Theoretical Significance: Refines the stability analysis framework in risk theory and provides theoretical guarantees for parameter estimation error propagation
• Practical Value: Assists actuaries in assessing error bounds when computing ruin probability using approximate parameters
• Methodological Innovation: Applies Banach contraction principle and fixed-point theory to new areas of ruin theory

Limitations of Existing Methods

• Strong stability methods primarily apply to classical models without diffusion
• Explicit solutions exist only for special distributions (exponential, phase-type)
• Insufficient analytical tools for continuity analysis of diffusion-perturbed models

Core Contributions

1. Continuity Inequality for Ruin Probability (Theorem 1): For claim distributions with finite (γ+1)-th moment, establishes ruin probability continuity bounds based on metric ν_γ
2. Continuity Bounds for Deficit at Ruin (Theorem 2): Derives uniform metric bounds between two deficit-at-ruin distributions of the form: $\sup_{u≥0}|G(u,y) - \tilde{G}(u,y)| ≤ \frac{1}{c-λμ}[λQ_y(F,\tilde{F}) + |λ-\tilde{λ}|\tilde{μ}]$
3. Continuity of K(u) in Diffusion Models (Theorem 3): First continuity estimates for the compound geometric distribution K(u) in diffusion-perturbed models
4. Iterative Approximation Method (Lemma 1): Proposes explicit iterative approximation formula for K(u) based on Banach fixed-point theorem with clear convergence guarantees
5. Numerical Verification: Validates theoretical bounds through multiple examples (though bounds are relatively loose)

Detailed Methodology

Problem Formulation

Given two risk processes:

• Process 1: Parameters (λ, c, F), ruin probability ψ(u), deficit at ruin G(u,y)
• Process 2: Parameters (λ̃, c, F̃), ruin probability ψ̃(u), deficit at ruin G̃(u,y)

Objective: Establish inequalities of the form d_out(ψ, ψ̃) ≤ f(d_in(λ,F; λ̃,F̃))

where d_in and d_out are appropriately chosen metrics and f is an increasing function satisfying f(0)=0.

Theoretical Framework

1. Banach Space Construction

• D_γ Space: Functions h satisfying h_γ(x) = (1+x)^γ|h(x)| ∈ D_∞ (right-continuous left-limited functions)
• Metric: ν_γ(x,y) = ∫_0^∞ (1+t)^γ|x(t)-y(t)|dt
• Completeness: (D_γ, ν_γ) is a complete metric space

2. Contraction Operator

Define the integral operator: $Tx(u) = \frac{λ}{c}\left[\int_u^∞ F(t)dt + \int_0^u x(u-t)F(t)dt\right]$

Key Properties:

• T: D_γ → D_γ
• Contraction modulus: λM_γ^X/c < 1 (under net profit condition)
• Where M_γ^X = ∫_0^∞(1+t)^γF(t)dt = E(X+1)^(γ+1)-1/(γ+1)

3. Fixed-Point Method

• Ruin probability ψ is the unique fixed point of operator T: ψ = Tψ
• Utilizes triangle inequality: ν_γ(ψ,ψ̃) ≤ ν_γ(Tψ,Tψ̃) + ν_γ(Tψ̃,T̃ψ̃)
• First term controlled by contractivity, second term measures operator difference

Theorem 1 Details (Ruin Probability Continuity)

Statement: Assume EX^(γ+1) < ∞, EX̃^(γ+1) < ∞, then: $\int_0^∞(1+t)^γ|ψ(t)-\tilde{ψ}(t)|dt ≤ \frac{c}{c-λM_γ^X}\left[\frac{ν_{γ+1}(F,\tilde{F})}{γ+1} + ν_γ(F,\tilde{F})M_γ^L + \frac{|λ-\tilde{λ}|}{c}M_{γ+1}^{\tilde{X}}(1+M_γ^L)\right]$

Proof Outline:

1. Establish ν_γ(Tx,Ty) ≤ (λM_γ^X/c)ν_γ(x,y) (contractivity)
2. Estimate ν_γ(Tψ̃,T̃ψ̃), decomposed into I_1 (distribution difference) + I_2 (intensity difference)
3. For I_1, apply change of order of integration:
   • ∫0^t(1+u)^γdu terms yield ν{γ+1}(F,F̃)/(γ+1)
   • Convolution terms use inequality (1+z+t)^γ ≤ (1+z)^γ(1+t)^γ
4. Handle I_2 similarly

Special Case (γ=0): Reduces to bounds under Kantorovich metric

Theorem 2 Details (Deficit at Ruin)

Background: The deficit-at-ruin tail G(u,y) = ψ(u) - G(u,y) satisfies: $G(u,y) = \frac{λ}{c}\left[\int_0^u G(u-t,y)F(t)dt + \int_{u+y}^∞ F(t)dt\right]$

Statement: $\sup_{u≥0}|G(u,y)-\tilde{G}(u,y)| ≤ \frac{1}{c-λμ}[λQ_y(F,\tilde{F}) + |λ-\tilde{λ}|\tilde{μ}]$

where Q_y(F,F̃) = ∫_y^∞|F(t)-F̃(t)|dt

Proof Strategy:

• Use uniform metric ν_d(x,y) = sup_t|x(t)-y(t)|
• Fix y and view G(·,y) as a function of u
• Verify contractivity of operator T^z (modulus λμ/c)
• Estimate ν_d(T^zΓ̃^z, T̃^zΓ̃^z)

Theorem 3 Details (K(u) in Diffusion Models)

Model: U(t) = u + ct - S(t) + σB(t)

Key Quantity: K(u) = Pr(L_K > u), where L_K is the aggregate loss from record highs caused by claims

Structure: K(u) satisfies the improper integral equation: $K(u) = \frac{λμ}{c}\left[A(u) + \int_0^u K(u-t)dA(t)\right]$

where A(x) is the distribution function of L_o + L_c, L_o ~ H_1 (diffusion term), L_c ~ F_e (equilibrium distribution)

Continuity Bound: $\sup_u|K(u)-\tilde{K}(u)| ≤ \frac{1}{c-λμ}\left[λμ\left(\frac{c}{D}K(H_1,\tilde{H}_1) + \frac{|\tilde{D}-D|}{D} + \frac{K(F,\tilde{F})}{μ} + \frac{|\tilde{μ}-μ|}{μ}\right) + |λμ-\tilde{λ}\tilde{μ}|\right]$

Technical Challenge: Requires estimating ∫|a(t)-ã(t)|dt, where a is the density of the convolution of H_1 and H_2

Iterative Approximation (Lemma 1)

Iteration Scheme: K_n = T_dK_, K_0 ∈ K_d

Explicit Formula (K_0=k constant):

• n=1: K_1(u) = φ - (1-k)φA(u)
• n≥2: K_n(u) = φ - (1-k)φ^nA^{*n}(u) - (1-φ)∑_^{n-1}φ^iA^{*i}(u)

where φ = 1/(1+θ)

Convergence: Guaranteed by BFPT with ν_d(K_n,K) → 0

Special Case (Exponential claims, β=c/D): $K_n(u) = φe^{-βu}S_1(βu) + \sum_{m=2}^{n-1}φ^me^{-βu}[S_{2m-1}(βu)-S_{2m-3}(βu)] + ...$

where S_m(z) = ∑_^m z^r/r!

Experimental Setup

Numerical Example Design

Example 1 (Theorem 1 Verification):

• Distribution: X ~ Mixed exponential (1/2·Exp(5/4) + 1/2·Exp(5/6)), X̃ ~ Exp(1)
• Parameters: EX = EX̃ = 1, multiple (λ,c) combinations
• Metrics: γ = 0, 1

Example 2 (Theorem 2 Verification):

• Three distributions: X^(1) ~ Erlang(3,3), X^(2) ~ Exp(1), X^(3) ~ Mixed exponential
• Parameters: θ = 1 or 4, multiple (u,y) pairs
• Comparison: |G^(i)(u,y) - G^(j)(u,y)| versus theoretical bound DK2

Example 3 (Theorem 3 Verification):

• Distribution: X ~ Exp(3), X̃ ~ Mixed exponential
• Diffusion parameters: multiple (D,D̃) combinations
• Comparison: sup|K(u)-K̃(u)| versus bound DK3

Example 4 (Iterative Approximation):

• Distribution: X ~ Exp(β)
• Initial values: k = 0.0, 0.1, ..., 1.0
• Iteration count: n = 1,...,5
• Two parameter settings

Computational Methods

• Exact values: Utilizing known explicit solutions (exponential, Erlang cases)
• Numerical integration: Computing metrics and bounds
• Iterative computation: Using explicit formulas from Lemma 1

Experimental Results

Example 1 Results (Table 1)

Observations:

• γ=0, λ=5/6:
  • True distance: 0.0154 (c=3) → 0.0060 (c=7)
  • Theoretical bound DK1: 0.1211 → 0.0929
  • Bound looseness: Approximately 8-15 times
• γ=1 yields looser bounds (approximately 8-13 times)
• Bound slightly increases with λ
• Both true distance and bound decrease with increasing c

Conclusion: Bounds are conservative but qualitatively correct

Example 2 Results (Table 2)

Four Comparisons:

1. θ=1, X^(1) vs X^(2): True distance 0.0080-0.0832, bound 0.1547-0.2900
2. θ=4, X^(1) vs X^(2): True distance 0.0034-0.0309, bound 0.0387-0.0735
3. θ=4, X^(1) vs X^(3): True distance 0.0035-0.0328, bound 0.0413-0.0779
4. θ=4, X^(3) vs X^(2): True distance 0.0001-0.0023, bound 0.0027-0.0054

Findings:

• Increasing θ (safety loading) significantly reduces both distance and bound
• Distance first increases then decreases with y (maximum exists)
• Similar distributions (X^(3) and X^(2)) have small distances (< 0.003)
• Bound looseness: Approximately 2-10 times, tighter when θ=4

Author's Note: Bound looseness arises from requiring uniform validity over all u

Example 3 Results (Table 3)

Diffusion Parameter Impact:

• D=1, D̃=1/10: True distance 0.0854, bound 0.4837 (5.66 times)
• D=3, D̃=1/20: True distance 0.1334, bound 0.5254 (3.94 times)

Trends:

• Larger difference between D and D̃ yields larger distance
• Distance increases with D
• Bound looseness: Approximately 3-7 times

Ratio Analysis: |1-Ratio| column shows relative errors of 2.94-6.75

Example 4 Results (Tables 4-5)

Convergence Speed (Table 4, β=2, exact value K(1)=0.3325717):

• n=1: 0.2030-0.5000 (depends on k)
• n=3: 0.3316-0.3336 (error <0.001)
• n=5: Approximately 0.3326 (error <10^-5)

Initial Value Sensitivity:

• k=0.0, n=5 error: 0.0000008
• k=1.0, n=5 error: 0.0000007
• Conclusion: After 5 iterations, insensitive to initial value

Table 5 (β=3/2, exact value K(1)=0.6573777):

• Similar rapid convergence
• n=5 error for all initial values <3×10^-6

Practical Value: 3-5 iterations achieve high-precision approximation

Related Work

Stability Methods

• Beirlant & Rachev (1987): First formalization of stability problems in actuarial mathematics
• Kalashnikov (2000): Strong stability methods for ruin probability
• Enikeeva et al. (2001): Continuity estimates for ruin probability
• Gordienko & Vázquez-Ortega (2016): Simple continuity inequalities using contraction mappings (directly extended in this paper)

Multidimensional and Markov-Modulated Models

• Benouaret & Aı̈ssani (2010): Strong stability for two-dimensional classical risk model
• Rusaityte (2001): Stability bounds for Markov-modulated + investment models
• Gordienko & Vázquez-Ortega (2018): Continuity inequalities for multidimensional renewal risk models

Contraction Principle Applications

• Gajek & Rudź (2018): Banach contraction principle for regime-switch models
• Jiang (2021): q-scale functions and ultimate ruin probability
• Sánchez & Baltazar-Larios (2020): Ruin probability approximation using BFPT
• Gajek & Rudź (2025): Fixed-point analysis for ruin problems

Diffusion Models

• Gerber (1970): Introduces diffusion-perturbed classical model
• Dufresne & Gerber (1991): Systematic study of three classes of ruin probabilities
• Tsai (2003, 2006): Structure and properties of K(u)
• Veraverbeke (1993): Asymptotic behavior analysis

Innovation of This Paper: First systematic application of probabilistic metric methods to diffusion-perturbed models

Conclusions and Discussion

Main Conclusions

1. Theoretical Contributions:
   • Establishes three main continuity inequalities (Theorems 1-3)
   • Provides computable error bounds
   • Extends contraction operator methods to diffusion models
2. Methodological Contributions:
   • Unified framework: selecting appropriate Banach spaces and metrics
   • Iterative approximation: explicit formulas with convergence guarantees
   • Numerical verification: multiple examples supporting theoretical results
3. Practical Significance:
   • Quantitative assessment of parameter estimation errors
   • Error control for approximation methods
   • Rapid computation of K(u) in diffusion models

Limitations

1. Bound Looseness:
   • Numerical examples show bounds 3-15 times larger than true distances
   • Reason: Requires uniform validity over all u and multiple triangle inequalities
   • Improvement potential: More refined analysis or adaptive bounds
2. Moment Conditions:
   • Theorem 1 requires EX^(γ+1) < ∞
   • Excludes certain heavy-tailed distributions (e.g., Pareto with α≤γ+1)
   • Alternative methods needed for heavy-tailed cases
3. Numerical Example Limitations:
   • Primarily uses exponential family distributions (with explicit solutions)
   • Limited verification for general distributions
   • High-dimensional cases not addressed
4. Diffusion Model Assumptions:
   • Theorem 3 assumes D≥D̃ and μ≥μ̃
   • Necessity of these assumptions insufficiently discussed
   • Treatment when assumptions violated not provided
5. Computational Complexity:
   • Bound computation involves multiple integrals
   • Practical feasibility for actual applications not thoroughly discussed

Future Directions

1. Theoretical Improvements:
   • Seeking tighter bounds
   • Handling heavy-tailed distributions
   • Extension to multidimensional models
2. Method Extensions:
   • Other risk model types (e.g., Lévy processes)
   • Time-dependent parameters
   • Dependent claims
3. Applied Research:
   • Parameter estimation from real data
   • Robustness analysis
   • Optimal approximation strategies

In-Depth Evaluation

Strengths

1. Mathematical Rigor:
   • Complete proof details
   • Clear mathematical framework
   • Explicit assumption statements
2. Methodological Innovation:
   • First systematic application of probabilistic metrics to diffusion models
   • Unified treatment of ruin probability and deficit at ruin
   • Explicit formulas for iterative approximation (Lemma 1)
3. Clear Structure:
   • Progression from simple to complex (classical → deficit → diffusion)
   • Each theorem accompanied by numerical verification
   • Good balance between technical details and intuitive explanation
4. Practical Value:
   • Provides computable error bounds
   • Fast convergence of iterative method (3-5 iterations)
   • Suitable for parameter sensitivity analysis
5. Literature Review:
   • Comprehensive coverage of related work
   • Clear positioning of contributions
   • Appropriate citations (32 references)

Weaknesses

1. Bound Effectiveness:
   • Numerical examples show relatively loose bounds (3-15 times)
   • May limit practical application value
   • Lacks tightness analysis (can bounds be improved?)
2. Numerical Experiments:
   • Primarily limited to exponential family (easy exact solutions)
   • Lacks Monte Carlo verification for general distributions
   • Computational cost not discussed
3. Theoretical Limitations:
   • Moment conditions exclude heavy-tailed cases
   • Theorem 3's one-sided assumptions (D≥D̃) insufficiently discussed
   • Limited guidance on optimal metric selection
4. Application Guidance:
   • No recommendations for choosing γ
   • Integration with parameter estimation not discussed
   • Insufficient comparison with other methods (e.g., strong stability)
5. Writing Details:
   • Some notation insufficiently explained (e.g., meaning of M^L_γ)
   • Theorem 2's remark mentions "uniformity" but not elaborated
   • Example 4 parameter selection motivation unclear

Impact Assessment

Theoretical Impact:

• Provides new analytical tools for ruin theory
• Extends contraction operator method applications
• May inspire research on tighter bounds

Practical Impact:

• Moderate: Bound looseness limits direct application
• Iterative method has practical value (fast convergence)
• Further work needed for risk management applications

Reproducibility:

• High: Complete numerical example parameters
• Detailed theoretical proofs
• Lacks code but formulas sufficient for implementation

Field Contribution:

• Fills gap in continuity analysis for diffusion models
• Provides systematic methodological framework
• Establishes foundation for subsequent research

Applicable Scenarios

Suitable Scenarios:

1. Parameter Sensitivity Analysis: Assessing parameter estimation error impact
2. Approximation Method Verification: Providing error bounds for numerical methods
3. Theoretical Research: Serving as benchmark for complex models
4. Teaching: Demonstrating contraction operator method applications

Unsuitable Scenarios:

1. Precise Error Control: Bounds may be overly conservative
2. Heavy-Tailed Distributions: Moment conditions not satisfied
3. High-Dimensional Complex Models: Method extension non-trivial
4. Real-Time Computation: Integral computation potentially slow

Improvement Recommendations:

• Combine with Monte Carlo methods for general cases
• Develop adaptive bounds (adjusted by specific parameters)
• Provide software implementation
• Integrate with machine learning for parameter estimation

Selected References

Foundational Literature:

• Asmussen & Albrecher (2010): Standard textbook on ruin probability theory
• Dufresne & Gerber (1991): Foundational work on diffusion-perturbed models

Methodology:

• Gordienko & Vázquez-Ortega (2016): Directly extended in this paper
• Gajek & Rudź (2018, 2025): Banach contraction principle applications

Diffusion Models:

• Tsai (2003, 2006): Structure and properties of K(u)
• Veraverbeke (1993): Asymptotic behavior

Overall Assessment: This is a mathematically rigorous and methodologically innovative theoretical paper that provides new analytical tools for ruin theory. Its main contribution lies in systematically applying probabilistic metric methods to diffusion-perturbed models and providing computable continuity bounds. Although the numerical bounds are relatively loose, limiting direct application value, the theoretical foundation and methodological contributions are significant. The paper is suitable for publication in specialized journals in actuarial science or applied probability and holds important reference value for researchers in this field. Future work should focus on improving bound tightness and extending to more general models.