Mean residual life processes and associated submartingales

Basic Information

• Paper ID: 1807.05932
• Title: Mean residual life processes and associated submartingales
• Author: Antoine-Marie Bogso (University of Yaounde I, Department of Mathematics)
• Classification: math.PR (Probability Theory)
• Publication Date: October 9, 2018 (arXiv v4)
• Paper Link: https://arxiv.org/abs/1807.05932

Abstract

This paper employs the Madan-Yor argument to construct a class of associated submartingales for two-parameter processes ordered by increasing convex dominance. This class includes processes with MTP₂ integral survival functions. The author proves that the integral survival function of an integrable two-parameter process is MTP₂ if and only if it is TP₂ in each pair of parameters when the remaining parameters are fixed. Since there exist multiple two-parameter processes whose integral survival functions lack interval support, this result cannot be derived from known results. The MTP₂ property of certain MRL processes facilitates the demonstration of many other processes with the same property.

Research Background and Motivation

Problem Background

1. Success in the one-parameter case: Madan and Yor successfully used the Azéma-Yor solution to provide explicit associated martingales for each constant mean one-parameter mean residual life (MRL) process. One-parameter MRL processes are ordered by increasing convex dominance, and Kellerer's theorem guarantees the existence of associated martingales.
2. Challenges in the two-parameter case: Although the Madan-Yor argument remains valid in the two-parameter case, there is no corresponding version of Kellerer's theorem. Juillet recently proved that Kellerer's theorem fails in the two-parameter case and provided counterexamples.
3. Limitations of existing approaches:
   • The Madan-Yor argument does not apply to non-MRL peacock processes
   • Lack of general construction methods for two-parameter processes
   • Existing total positivity results apply only to functions with interval support

Research Motivation

1. Extending the theoretical framework: Extending one-parameter MRL theory to the two-parameter case, particularly for processes with MTP₂ integral survival functions
2. Resolving technical challenges: Proving equivalent characterizations of the MTP₂ property, even for functions lacking interval support
3. Constructing new process families: Utilizing the preservation of MTP₂ properties to construct more processes with the same property

Core Contributions

1. Theoretical breakthrough: Proves that the integral survival function of an integrable two-parameter process is MTP₂ if and only if it is TP₂ in each pair of parameters when the remaining parameters are fixed, extending the Karlin-Rinott result to cases without interval support
2. Construction method: Uses the Madan-Yor argument and the Cox-Hobson algorithm to construct associated submartingales for two-parameter MRL processes
3. Process family demonstration: Provides multiple families of two-parameter MRL processes with MTP₂ integral survival functions, including:
   • Biatomic MRL processes
   • Truncated transformation processes
   • Subordinated processes
   • Convolution processes
4. Treatment of non-MRL processes: Constructs associated submartingales for certain non-MRL but increasingly convex dominance ordered processes

Methodology Details

Task Definition

Study the MRL ordering properties of two-parameter probability measure families $(μ_t, t ∈ ℝ²₊)$ and construct associated submartingale processes. Main tasks include:

• Characterizing equivalent conditions for MRL ordering
• Proving equivalence of MTP₂ properties
• Constructing associated submartingales

Core Concepts

1. Two-parameter MRL Process Definition

A two-parameter MRL process is an integrable probability measure family $(μ_t, t ∈ ℝ²₊)$ whose Hardy-Littlewood function family $(Ψ_{μ_t}, t ∈ ℝ²₊)$ is pointwise non-decreasing, where:

$Ψ_{μ_t}(x) = \begin{cases} \frac{1}{μ_t([x,+∞[)} \int_{[x,+∞[} y μ_t(dy) & \text{if } x < r_{μ_t} \\ x & \text{if } x ≥ r_{μ_t} \end{cases}$

2. MTP₂ Function Definition

For $n ≥ 2$, a function $f$ is MTP₂ on $I = \prod_{i=1}^n I_i$ if for each $(x,y) ∈ I × I$: $f(x ∧ y)f(x ∨ y) ≥ f(x)f(y)$

3. Integral Survival Function

For an integrable process $μ = (μ_t, t ∈ ℝ²₊)$, its integral survival function is defined as: $C_μ(t,x) = \int_{[x,+∞[} (y-x)μ_t(dy)$

Main Theoretical Results

Theorem 2.6 (Characterization of MRL Ordering)

The family $(μ_t, t ∈ ℝ²₊)$ is non-decreasing in MRL ordering if and only if its integral survival function $C_μ$ satisfies: $∀ t_1 ≤ t_2, ∀ x_1 ≤ x_2, \det\begin{pmatrix} C_μ(t_1,x_1) & C_μ(t_1,x_2) \\ C_μ(t_2,x_1) & C_μ(t_2,x_2) \end{pmatrix} ≥ 0$

Theorem 2.8 (Equivalence of MTP₂ Property)

Let $(μ_t, t ∈ ℝ²₊)$ be an integrable process and $C_μ$ be its integral survival function. If $C_μ$ is TP₂ in each pair of variables when the remaining variables are fixed, then $C_μ$ is MTP₂.

Cox-Hobson Algorithm

For integrable probability measure families with positive mean, the Cox-Hobson algorithm provides a family of stopping times $(T_{μ_t}, t ∈ ℝ²₊)$ such that:

• $(B^+_{T_{μ_t} ∧ v}, v ≥ 0)$ is uniformly integrable
• The distribution of $B_{T_{μ_t}}$ is $μ_t$

When the family $(T_{μ_t}, t ∈ ℝ²₊)$ is almost surely non-decreasing, $(B_{T_{μ_t}}, t ∈ ℝ²₊)$ forms a submartingale.

Experimental Setup

Theoretical Verification

This work is primarily theoretical, with verification through construction of concrete examples:

1. Biatomic process families: Constructs process families with parameters $ε ∈ (0,1)$ and $r ∈ ℝ$, verifying MTP₂ properties
2. Truncated transformations: Constructs two-parameter processes from one-parameter MRL processes via truncation
3. Subordination construction: Constructs new MRL processes using transition kernels of Markov processes

Concrete Example Analysis

Example 3.1 (Biatomic MRL Process)

For $ε ∈ (0,1)$, $r ∈ ℝ$, define the process:

• $μ^ε_{(t,0)} = μ^ε_{(0,t)} = δ_r$
• For $t = (t,t') ∈ ℝ^*_+ × ℝ^*_+$: $μ^ε_t = \frac{t'}{t+t'} δ_{r-(1-ε)t} + \frac{t}{t+t'} δ_{r+t'}$

Verifies that the integral survival function of this process is TP₂ in each pair of variables.

Experimental Results

Main Theoretical Achievements

1. MTP₂ equivalence: Successfully proves that the MTP₂ property of the integral survival function is equivalent to the TP₂ property in each pair of parameters when remaining parameters are fixed, with this result holding even for functions lacking interval support.
2. Process construction: Demonstrates multiple families of MRL processes with MTP₂ integral survival functions:
   • Biatomic processes (Proposition 3.1)
   • Truncated transformation processes (Propositions 3.5, 3.6)
   • Subordinated processes (Theorem 3.11)
   • Convolution processes (Proposition 3.14)
3. Non-MRL process treatment: Constructs associated submartingales for certain non-MRL but convex dominance ordered processes (Theorems 4.4, 4.6)

Specific Verification Results

Biatomic Process Verification

• Proves that the process family $(μ^ε_t, t ∈ ℝ²₊)$ is indeed an MRL process
• Verifies that its integral survival function $C_{μ^ε}$ is TP₂ in $(t,t')$
• Notes that this function lacks interval support, so the Karlin-Rinott result does not apply

Truncated Process Verification

• Constructs truncated process families starting from finitely supported measures
• Proves that these processes have MTP₂ integral survival functions
• Demonstrates how parameter $ε$ affects the mean structure of the process

Related Work

One-parameter Theoretical Foundation

1. Madan-Yor work: Establishes connections between MRL ordering and martingale properties, using the Azéma-Yor embedding solution
2. Kellerer's theorem: Proves that processes ordered by increasing convex dominance have submartingale marginal distributions
3. Cox-Hobson algorithm: Generalizes the Azéma-Yor embedding to non-centered target distributions

Two-parameter Extension Attempts

1. Juillet's work: Proves that Kellerer's theorem fails in the two-parameter case, introduces the concept of biatomic convex ordering
2. Total positivity theory: MTP₂ function theory by Karlin-Rinott and others, but limited to interval support cases

Advantages of This Work

1. Theoretical breakthrough: First to prove the MTP₂ equivalent characterization for integral survival functions lacking interval support
2. Construction method: Provides systematic construction methods for two-parameter MRL processes
3. Application extension: Applies methods to non-MRL processes, expanding the scope of applicability

Conclusions and Discussion

Main Conclusions

1. Theoretical contribution: Establishes a complete characterization of the MTP₂ property of integral survival functions for two-parameter processes, extending existing theory to non-interval support cases
2. Methodological contribution: Demonstrates the effectiveness of the Madan-Yor argument in the two-parameter case, providing systematic methods for constructing associated submartingales
3. Construction achievement: Provides rich families of processes with MTP₂ properties, laying the foundation for further research

Limitations

1. Technical constraints: The Cox-Hobson algorithm requires processes to have positive mean; handling general cases requires translation
2. Scope of applicability: Main results focus on MRL processes; applicability to more general process families is limited
3. Computational complexity: Practical construction of associated submartingales may involve complex stopping time calculations

Future Directions

1. Generalization research: Explore more general classes of two-parameter processes
2. Numerical methods: Develop numerical algorithms for practical computation of associated submartingales
3. Application extension: Apply theory to practical problems in financial mathematics and risk management

In-depth Evaluation

Strengths

1. Theoretical depth: Resolves an important theoretical problem, proving MTP₂ equivalence in the non-interval support case
2. Systematic methodology: Provides a complete theoretical framework with multiple construction methods
3. Technical innovation: Skillfully handles the technical challenges of non-interval support cases
4. Clear presentation: Well-structured paper with detailed proofs and clear exposition

Weaknesses

1. Limited applications: Primarily theoretical work lacking practical application examples
2. Numerical verification: No numerical experiments provided to verify theoretical results
3. Computational complexity: Does not discuss the complexity and feasibility of practical computation

Impact

1. Theoretical contribution: Provides important supplements to two-parameter stochastic process theory
2. Methodological value: Offers new tools and perspectives for related research
3. Foundation for future work: Establishes basis for further theoretical and applied research

Applicable Scenarios

1. Stochastic process theory: Theoretical research on two-parameter processes
2. Financial mathematics: Multidimensional risk modeling and derivative pricing
3. Probability theory: Further development of MRL ordering and total positivity theory

References

This paper cites 24 important references, including:

• Madan & Yor (2002): Pioneering work on MRL ordering and martingale properties
• Karlin & Rinott (1980): Classical literature on MTP₂ function theory
• Cox & Hobson (2006): Important advances in the Skorokhod embedding problem
• Juillet (2016): Recent developments in two-parameter process theory