Presentation / Talk

The estimation of multivariate extreme value models from choice-based samples

We consider an estimation procedure for discrete choice models in general and Multivariate Extreme Value (MEV) models in particular. It is based on a pseudo-likelihood function, generalizing the Conditional Maximum Likelihood (CML) estimator by Manski and McFadden (1981) and theWeighted Exogenous Sample Maximum Likelihood (WESML) estimator by Manski and Lerman (1977). We show that the property of Multinomial Logit (MNL) models, that consistent estimates of all parameters but the constants can be obtained from an Exogenous Sample Maximum Likelihood (ESML) estimation, does not hold in general for MEV models. We propose a new estimator for the more general case. This new estimator estimates the selection bias directly from the data. We illustrate the new estimator on pseudo-synthetic and real data.

Related material