Bayesian Model Comparison with the Hyvärinen Score: Computation and Consistency

Stephane Shao, Pierre E. Jacob, Jie Ding, Vahid Tarokh

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

The Bayes factor is a widely used criterion in model comparison and its logarithm is a difference of out-of-sample predictive scores under the logarithmic scoring rule. However, when some of the candidate models involve vague priors on their parameters, the log-Bayes factor features an arbitrary additive constant that hinders its interpretation. As an alternative, we consider model comparison using the Hyvärinen score. We propose a method to consistently estimate this score for parametric models, using sequential Monte Carlo methods. We show that this score can be estimated for models with tractable likelihoods as well as nonlinear non-Gaussian state-space models with intractable likelihoods. We prove the asymptotic consistency of this new model selection criterion under strong regularity assumptions in the case of nonnested models, and we provide qualitative insights for the nested case. We also use existing characterizations of proper scoring rules on discrete spaces to extend the Hyvärinen score to discrete observations. Our numerical illustrations include Lévy-driven stochastic volatility models and diffusion models for population dynamics. Supplementary materials for this article are available online.

Original languageEnglish (US)
JournalJournal of the American Statistical Association
DOIs
StatePublished - Jan 1 2019

Fingerprint

Model Comparison
Bayesian Model
Bayes Factor
Scoring
Likelihood
Sequential Monte Carlo Methods
Discrete Observations
Strong Regularity
Model Selection Criteria
Stochastic Volatility Model
State-space Model
Diffusion Model
Population Dynamics
Parametric Model
Logarithm
Logarithmic
Bayesian model
Model comparison
Model
Alternatives

Keywords

  • Bayes factor
  • Model selection
  • Noninformative prior
  • SMC
  • State-space model

Cite this

Bayesian Model Comparison with the Hyvärinen Score : Computation and Consistency. / Shao, Stephane; Jacob, Pierre E.; Ding, Jie; Tarokh, Vahid.

In: Journal of the American Statistical Association, 01.01.2019.

Research output: Contribution to journalArticle

@article{7662220e17b146a38b25730073dd152d,
title = "Bayesian Model Comparison with the Hyv{\"a}rinen Score: Computation and Consistency",
abstract = "The Bayes factor is a widely used criterion in model comparison and its logarithm is a difference of out-of-sample predictive scores under the logarithmic scoring rule. However, when some of the candidate models involve vague priors on their parameters, the log-Bayes factor features an arbitrary additive constant that hinders its interpretation. As an alternative, we consider model comparison using the Hyv{\"a}rinen score. We propose a method to consistently estimate this score for parametric models, using sequential Monte Carlo methods. We show that this score can be estimated for models with tractable likelihoods as well as nonlinear non-Gaussian state-space models with intractable likelihoods. We prove the asymptotic consistency of this new model selection criterion under strong regularity assumptions in the case of nonnested models, and we provide qualitative insights for the nested case. We also use existing characterizations of proper scoring rules on discrete spaces to extend the Hyv{\"a}rinen score to discrete observations. Our numerical illustrations include L{\'e}vy-driven stochastic volatility models and diffusion models for population dynamics. Supplementary materials for this article are available online.",
keywords = "Bayes factor, Model selection, Noninformative prior, SMC, State-space model",
author = "Stephane Shao and Jacob, {Pierre E.} and Jie Ding and Vahid Tarokh",
year = "2019",
month = "1",
day = "1",
doi = "10.1080/01621459.2018.1518237",
language = "English (US)",
journal = "Journal of the American Statistical Association",
issn = "0162-1459",
publisher = "Taylor and Francis Ltd.",

}

TY - JOUR

T1 - Bayesian Model Comparison with the Hyvärinen Score

T2 - Computation and Consistency

AU - Shao, Stephane

AU - Jacob, Pierre E.

AU - Ding, Jie

AU - Tarokh, Vahid

PY - 2019/1/1

Y1 - 2019/1/1

N2 - The Bayes factor is a widely used criterion in model comparison and its logarithm is a difference of out-of-sample predictive scores under the logarithmic scoring rule. However, when some of the candidate models involve vague priors on their parameters, the log-Bayes factor features an arbitrary additive constant that hinders its interpretation. As an alternative, we consider model comparison using the Hyvärinen score. We propose a method to consistently estimate this score for parametric models, using sequential Monte Carlo methods. We show that this score can be estimated for models with tractable likelihoods as well as nonlinear non-Gaussian state-space models with intractable likelihoods. We prove the asymptotic consistency of this new model selection criterion under strong regularity assumptions in the case of nonnested models, and we provide qualitative insights for the nested case. We also use existing characterizations of proper scoring rules on discrete spaces to extend the Hyvärinen score to discrete observations. Our numerical illustrations include Lévy-driven stochastic volatility models and diffusion models for population dynamics. Supplementary materials for this article are available online.

AB - The Bayes factor is a widely used criterion in model comparison and its logarithm is a difference of out-of-sample predictive scores under the logarithmic scoring rule. However, when some of the candidate models involve vague priors on their parameters, the log-Bayes factor features an arbitrary additive constant that hinders its interpretation. As an alternative, we consider model comparison using the Hyvärinen score. We propose a method to consistently estimate this score for parametric models, using sequential Monte Carlo methods. We show that this score can be estimated for models with tractable likelihoods as well as nonlinear non-Gaussian state-space models with intractable likelihoods. We prove the asymptotic consistency of this new model selection criterion under strong regularity assumptions in the case of nonnested models, and we provide qualitative insights for the nested case. We also use existing characterizations of proper scoring rules on discrete spaces to extend the Hyvärinen score to discrete observations. Our numerical illustrations include Lévy-driven stochastic volatility models and diffusion models for population dynamics. Supplementary materials for this article are available online.

KW - Bayes factor

KW - Model selection

KW - Noninformative prior

KW - SMC

KW - State-space model

UR - http://www.scopus.com/inward/record.url?scp=85056782562&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85056782562&partnerID=8YFLogxK

U2 - 10.1080/01621459.2018.1518237

DO - 10.1080/01621459.2018.1518237

M3 - Article

AN - SCOPUS:85056782562

JO - Journal of the American Statistical Association

JF - Journal of the American Statistical Association

SN - 0162-1459

ER -