Predictive likelihood for Bayesian model selection and averaging

Tomohiro Ando, Ruey Tsay

Research output: Contribution to journalArticle

23 Citations (Scopus)

Abstract

This paper investigates the performance of the predictive distributions of Bayesian models. To overcome the difficulty of evaluating the predictive likelihood, we introduce the concept of expected log-predictive likelihoods for Bayesian models, and propose an estimator of the expected log-predictive likelihood. The estimator is derived by correcting the asymptotic bias of the log-likelihood of the predictive distribution as an estimate of its expected value. We investigate the relationship between the proposed criterion and the traditional information criteria and show that the proposed criterion is a natural extension of the traditional ones. A new model selection criterion and a new model averaging method are then developed, with the weights for the individual models being dependent on their expected log-predictive likelihoods. We examine the performance of the proposed method using Monte Carlo experiments and a real example, which concerns the prediction of quarterly growth rates of real gross domestic product in the G7 countries. Out-of-sample forecasts show that the proposed methodology outperforms other methods available in the literature.

Original languageEnglish
Pages (from-to)744-763
Number of pages20
JournalInternational Journal of Forecasting
Volume26
Issue number4
DOIs
Publication statusPublished - 2010 Oct

Fingerprint

Model averaging
Model selection
Bayesian model
Predictive distribution
Estimator
Expected value
Gross domestic product
Out-of-sample forecasting
Information criterion
Methodology
Model selection criteria
Prediction
Individual model
Asymptotic bias
G-7 countries
Monte Carlo experiment

Keywords

  • Finance
  • Model averaging
  • Model selection
  • Predictive likelihood

ASJC Scopus subject areas

  • Business and International Management

Cite this

Predictive likelihood for Bayesian model selection and averaging. / Ando, Tomohiro; Tsay, Ruey.

In: International Journal of Forecasting, Vol. 26, No. 4, 10.2010, p. 744-763.

Research output: Contribution to journalArticle

Ando, Tomohiro ; Tsay, Ruey. / Predictive likelihood for Bayesian model selection and averaging. In: International Journal of Forecasting. 2010 ; Vol. 26, No. 4. pp. 744-763.
@article{e6f8b01b3a17421893b931b8aec483ef,
title = "Predictive likelihood for Bayesian model selection and averaging",
abstract = "This paper investigates the performance of the predictive distributions of Bayesian models. To overcome the difficulty of evaluating the predictive likelihood, we introduce the concept of expected log-predictive likelihoods for Bayesian models, and propose an estimator of the expected log-predictive likelihood. The estimator is derived by correcting the asymptotic bias of the log-likelihood of the predictive distribution as an estimate of its expected value. We investigate the relationship between the proposed criterion and the traditional information criteria and show that the proposed criterion is a natural extension of the traditional ones. A new model selection criterion and a new model averaging method are then developed, with the weights for the individual models being dependent on their expected log-predictive likelihoods. We examine the performance of the proposed method using Monte Carlo experiments and a real example, which concerns the prediction of quarterly growth rates of real gross domestic product in the G7 countries. Out-of-sample forecasts show that the proposed methodology outperforms other methods available in the literature.",
keywords = "Finance, Model averaging, Model selection, Predictive likelihood",
author = "Tomohiro Ando and Ruey Tsay",
year = "2010",
month = "10",
doi = "10.1016/j.ijforecast.2009.08.001",
language = "English",
volume = "26",
pages = "744--763",
journal = "International Journal of Forecasting",
issn = "0169-2070",
publisher = "Elsevier",
number = "4",

}

TY - JOUR

T1 - Predictive likelihood for Bayesian model selection and averaging

AU - Ando, Tomohiro

AU - Tsay, Ruey

PY - 2010/10

Y1 - 2010/10

N2 - This paper investigates the performance of the predictive distributions of Bayesian models. To overcome the difficulty of evaluating the predictive likelihood, we introduce the concept of expected log-predictive likelihoods for Bayesian models, and propose an estimator of the expected log-predictive likelihood. The estimator is derived by correcting the asymptotic bias of the log-likelihood of the predictive distribution as an estimate of its expected value. We investigate the relationship between the proposed criterion and the traditional information criteria and show that the proposed criterion is a natural extension of the traditional ones. A new model selection criterion and a new model averaging method are then developed, with the weights for the individual models being dependent on their expected log-predictive likelihoods. We examine the performance of the proposed method using Monte Carlo experiments and a real example, which concerns the prediction of quarterly growth rates of real gross domestic product in the G7 countries. Out-of-sample forecasts show that the proposed methodology outperforms other methods available in the literature.

AB - This paper investigates the performance of the predictive distributions of Bayesian models. To overcome the difficulty of evaluating the predictive likelihood, we introduce the concept of expected log-predictive likelihoods for Bayesian models, and propose an estimator of the expected log-predictive likelihood. The estimator is derived by correcting the asymptotic bias of the log-likelihood of the predictive distribution as an estimate of its expected value. We investigate the relationship between the proposed criterion and the traditional information criteria and show that the proposed criterion is a natural extension of the traditional ones. A new model selection criterion and a new model averaging method are then developed, with the weights for the individual models being dependent on their expected log-predictive likelihoods. We examine the performance of the proposed method using Monte Carlo experiments and a real example, which concerns the prediction of quarterly growth rates of real gross domestic product in the G7 countries. Out-of-sample forecasts show that the proposed methodology outperforms other methods available in the literature.

KW - Finance

KW - Model averaging

KW - Model selection

KW - Predictive likelihood

UR - http://www.scopus.com/inward/record.url?scp=77956264822&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=77956264822&partnerID=8YFLogxK

U2 - 10.1016/j.ijforecast.2009.08.001

DO - 10.1016/j.ijforecast.2009.08.001

M3 - Article

VL - 26

SP - 744

EP - 763

JO - International Journal of Forecasting

JF - International Journal of Forecasting

SN - 0169-2070

IS - 4

ER -