Statistical inference with exchangeability and martingales

Area of Science:

• Statistics
• Probability Theory

Background:

• Review of exchangeability and its significance in Bayesian statistics.
• Highlighting the predictive nature of Bayesian models and symmetry assumptions.
• Examining existing bootstrap methods and Doob's martingale-based Bayesian inference.

Purpose of the Study:

• Introduce a novel parametric Bayesian bootstrap.
• Demonstrate the fundamental role of martingales in this new approach.
• Provide theoretical underpinnings and practical illustrations.

Main Methods:

• Reviewing exchangeability and Bayesian approaches.
• Analyzing the Bayesian bootstrap and Efron's parametric bootstrap.
• Developing a parametric Bayesian bootstrap using martingales.

Main Results:

• A new parametric Bayesian bootstrap method is proposed.
• Martingales are shown to be crucial for the methodology.
• Theoretical results and illustrative examples are presented.

Conclusions:

• The proposed parametric Bayesian bootstrap offers a new tool for Bayesian inference.
• Martingales are integral to understanding and applying this method.
• The work contributes to the ongoing discourse on Bayesian inference challenges and prospects.