Application of an adaptive Monte Carlo algorithm to mixed logit estimation

Fabian Bastin, Cinzia Cirillo, Philippe Toint

Résultats de recherche: Contribution à un journal/une revueArticle

77 Downloads (Pure)

Résumé

This paper presents the application of a new algorithm for maximizing the simulated likelihood functions appearing in the estimation of mixed multinomial logit (MMNL) models. The method uses Monte Carlo sampling to produce the approximate likelihood function and dynamically adapts the number of draws on the basis of statistical estimators of the simulation error and simulation bias. Its convergence from distant starting points is ensured by a trust-region technique, in which improvement is ensured by locally maximizing a quadratic model of the objective function. Simulated data are first used to assess the quality of the results obtained and the relative performance of several algorithmic variants. These variants involve, in particular, different techniques for approximating the model's Hessian and the substitution of the trust-region mechanism by a linesearch. The algorithm is also applied to a real case study arising in the context of a recent Belgian transportation model. The performance of the new Monte Carlo algorithm is shown to be competitive with that of existing tools using low discrepancy sequences. © 2005 Elsevier Ltd. All rights reserved.
langue originaleAnglais
Pages (de - à)577-593
Nombre de pages17
journalTransportation Research Part B : Methodological
Volume40
Numéro de publication7
Les DOIs
étatPublié - 1 août 2006

Empreinte digitale

simulation
Belgian
substitution
performance
Substitution reactions
Sampling
trend

Citer ceci

@article{33565eaa4bc74693a16df8893d493b2d,
title = "Application of an adaptive Monte Carlo algorithm to mixed logit estimation",
abstract = "This paper presents the application of a new algorithm for maximizing the simulated likelihood functions appearing in the estimation of mixed multinomial logit (MMNL) models. The method uses Monte Carlo sampling to produce the approximate likelihood function and dynamically adapts the number of draws on the basis of statistical estimators of the simulation error and simulation bias. Its convergence from distant starting points is ensured by a trust-region technique, in which improvement is ensured by locally maximizing a quadratic model of the objective function. Simulated data are first used to assess the quality of the results obtained and the relative performance of several algorithmic variants. These variants involve, in particular, different techniques for approximating the model's Hessian and the substitution of the trust-region mechanism by a linesearch. The algorithm is also applied to a real case study arising in the context of a recent Belgian transportation model. The performance of the new Monte Carlo algorithm is shown to be competitive with that of existing tools using low discrepancy sequences. {\circledC} 2005 Elsevier Ltd. All rights reserved.",
keywords = "Mixed logit, adaptive sampling, transportation modelling, trust-region methods, discrete choices",
author = "Fabian Bastin and Cinzia Cirillo and Philippe Toint",
note = "Copyright 2008 Elsevier B.V., All rights reserved.",
year = "2006",
month = "8",
day = "1",
doi = "10.1016/j.trb.2005.09.002",
language = "English",
volume = "40",
pages = "577--593",
journal = "Transportation Research Part B : Methodological",
issn = "0191-2615",
publisher = "Elsevier Limited",
number = "7",

}

Application of an adaptive Monte Carlo algorithm to mixed logit estimation. / Bastin, Fabian; Cirillo, Cinzia; Toint, Philippe.

Dans: Transportation Research Part B : Methodological, Vol 40, Numéro 7, 01.08.2006, p. 577-593.

Résultats de recherche: Contribution à un journal/une revueArticle

TY - JOUR

T1 - Application of an adaptive Monte Carlo algorithm to mixed logit estimation

AU - Bastin, Fabian

AU - Cirillo, Cinzia

AU - Toint, Philippe

N1 - Copyright 2008 Elsevier B.V., All rights reserved.

PY - 2006/8/1

Y1 - 2006/8/1

N2 - This paper presents the application of a new algorithm for maximizing the simulated likelihood functions appearing in the estimation of mixed multinomial logit (MMNL) models. The method uses Monte Carlo sampling to produce the approximate likelihood function and dynamically adapts the number of draws on the basis of statistical estimators of the simulation error and simulation bias. Its convergence from distant starting points is ensured by a trust-region technique, in which improvement is ensured by locally maximizing a quadratic model of the objective function. Simulated data are first used to assess the quality of the results obtained and the relative performance of several algorithmic variants. These variants involve, in particular, different techniques for approximating the model's Hessian and the substitution of the trust-region mechanism by a linesearch. The algorithm is also applied to a real case study arising in the context of a recent Belgian transportation model. The performance of the new Monte Carlo algorithm is shown to be competitive with that of existing tools using low discrepancy sequences. © 2005 Elsevier Ltd. All rights reserved.

AB - This paper presents the application of a new algorithm for maximizing the simulated likelihood functions appearing in the estimation of mixed multinomial logit (MMNL) models. The method uses Monte Carlo sampling to produce the approximate likelihood function and dynamically adapts the number of draws on the basis of statistical estimators of the simulation error and simulation bias. Its convergence from distant starting points is ensured by a trust-region technique, in which improvement is ensured by locally maximizing a quadratic model of the objective function. Simulated data are first used to assess the quality of the results obtained and the relative performance of several algorithmic variants. These variants involve, in particular, different techniques for approximating the model's Hessian and the substitution of the trust-region mechanism by a linesearch. The algorithm is also applied to a real case study arising in the context of a recent Belgian transportation model. The performance of the new Monte Carlo algorithm is shown to be competitive with that of existing tools using low discrepancy sequences. © 2005 Elsevier Ltd. All rights reserved.

KW - Mixed logit

KW - adaptive sampling

KW - transportation modelling

KW - trust-region methods

KW - discrete choices

UR - http://www.scopus.com/inward/record.url?scp=33645944677&partnerID=8YFLogxK

U2 - 10.1016/j.trb.2005.09.002

DO - 10.1016/j.trb.2005.09.002

M3 - Article

VL - 40

SP - 577

EP - 593

JO - Transportation Research Part B : Methodological

JF - Transportation Research Part B : Methodological

SN - 0191-2615

IS - 7

ER -