[Paper Review] Conditional Markov Chains Part II: Consistency and Copulae

Tomasz R. Bielecki, Jacek Jakubowski+1 more|arXiv (Cornell University)|Jan 22, 2015

Bayesian Modeling and Causal Inference18 references3 citations

TL;DR

This paper introduces and analyzes strong and weak Markovian consistency and copulae for conditional Markov chains (CMCs) with finite state spaces, enabling the construction of multivariate CMCs with prespecified marginal laws and controlled dependence. The key contribution is a rigorous framework for modeling dependent dynamic systems under conditioning, with applications to credit risk and insurance pricing.

ABSTRACT

In this paper we continue the study of conditional Markov chains (CMCs) with finite state spaces, that we initiated in Bielecki, Jakubowski and Niewęgłowski (2015). Here, we turn our attention to the study of Markov consistency and Markov copulae with regard to CMCs, and thus we follow up on the study of Markov consistency and Markov copulae for ordinary Markov chains that we presented in Bielecki, Jakubowski and Niewęgłowski (2013).

Motivation & Objective

• To develop a framework for modeling multivariate conditional Markov chains (CMCs) with given marginal laws and controlled dependence structures.
• To extend the concepts of Markov consistency and copulae from ordinary Markov chains to CMCs, particularly in the context of doubly stochastic processes.
• To address practical modeling needs in finance and insurance where conditioning on reference information (e.g., credit ratings, unemployment status) is essential.
• To establish sufficient and necessary conditions for strong and weak Markovian consistency in CMCs, ensuring compatibility between conditional dynamics and marginal laws.
• To introduce and characterize strong and weak CMC copulae as tools for constructing dependent CMCs with specified dependence patterns.

Proposed method

• Introduces strong and weak Markovian consistency for CMCs via conditional transition fields, ensuring that the conditional law of the process remains Markovian under a given filtration.
• Derives algebraic conditions (e.g., Condition (ASM-k)) for strong Markovian consistency, analyzing their necessity and implications.
• Defines strong CMC copulae as joint laws that preserve conditional independence and Markov properties under the conditioning sigma-field.
• Proposes weak CMC copulae as a relaxation allowing dependence structures that are Markovian only in a weaker sense, not necessarily preserving full conditional independence.
• Applies the theory to construct CMCs with given marginal laws using copula-based dependence structures, validated through examples like conditionally independent, common jump, and perfect dependence copulae.
• Uses the tower property and conditional expectation decomposition to verify consistency and independence properties in the multivariate setting.

Experimental results

Research questions

• RQ1 Under what conditions is a conditional Markov chain Markovian under a given filtration, both strongly and weakly?
• RQ2 How can one construct a multivariate CMC such that its components have prespecified marginal laws and a desired dependence structure?
• RQ3 What is the relationship between weak and strong Markovian consistency in CMCs, and when does weak consistency imply strong consistency?
• RQ4 How do CMC copulae—specifically strong and weak ones—enable the modeling of dependent dynamic systems under conditioning?
• RQ5 Can the proposed framework be applied to real-world problems such as pricing unemployment insurance products with dependent individual dynamics?

Key findings

• Strong Markovian consistency is characterized by necessary and sufficient conditions on the transition kernels, with Condition (ASM-k) shown not to be necessary.
• Weak Markovian consistency is established via sufficient and necessary conditions, and it is shown that weak consistency does not imply strong consistency in general.
• A conditionally independent strong CMC copula is constructed such that components are independent given the conditioning sigma-field and preserve their marginal laws.
• A common jump strong CMC copula is proposed where components jump simultaneously with a common intensity, preserving the Markov property under conditioning.
• A perfect dependence strong CMC copula is constructed where components evolve identically, ensuring perfect dependence while maintaining Markov consistency.
• A weak CMC copula is exhibited that is not a strong CMC copula, demonstrating that weak consistency does not entail full conditional independence or strong Markov property.