Bayesian Full Information Analysis of Simultaneous Equation Models Using Integration by Monte Carlo

1. Front Matter

   Pages N2-VI

   PDF

2. Introduction

   • Luc Bauwens

   Pages 1-3

3. The Statistical Model

   • Luc Bauwens

   Pages 4-7

4. Bayesian Inference: The Extended Natural-Conjugate Approach

   • Luc Bauwens

   Pages 8-20

5. Selection of Importance Functions

   • Luc Bauwens

   Pages 21-34

6. Report and Discussion of Experiments

   • Luc Bauwens

   Pages 35-66

7. Extensions

   • Luc Bauwens

   Pages 67-69

8. Conclusion

   • Luc Bauwens

   Pages 70-70

9. Back Matter

   Pages 71-117

   PDF

In their review of the "Bayesian analysis of simultaneous equation systems", Dr~ze and Richard (1983) - hereafter DR - express the following viewpoint about the present state of development of the Bayesian full information analysis of such sys­ tems i) the method allows "a flexible specification of the prior density, including well defined noninformative prior measures"; ii) it yields "exact finite sample posterior and predictive densities". However, they call for further developments so that these densities can be eval­ uated through 'numerical methods, using an integrated software packa~e. To that end, they recommend the use of a Monte Carlo technique, since van Dijk and Kloek (1980) have demonstrated that "the integrations can be done and how they are done". In this monograph, we explain how we contribute to achieve the developments suggested by Dr~ze and Richard. A basic idea is to use known properties of the porterior density of the param­ eters of the structural form to design the importance functions, i. e. approximations of the posterior density, that are needed for organizing the integrations.

• Analysis
• Information
• Integration
• Likelihood
• calculus
• statistical model

• C.O.R.E., Louvain-La-Neuve, Belgium

  Luc Bauwens

Policies and ethics