Ricardo Daziano, Cornell University
"Bayesian statistical learning of preferences with applications to electric vehicle demand and use"
Abstract
Bayesian statistical learning of preferences with applications to electric vehicle demand and useBayesian estimation of discrete choice models offers several advantages over classical methods, particularly in handling parameter uncertainty and accommodating complex model structures. Bayesian approaches allow for flexible and robust inference, especially when dealing with models where maximum simulated likelihood estimation (MSLE) is computationally expensive or even prohibitive. The posterior distribution provides a full probabilistic characterization of the parameters, yielding more nuanced insights such as credible intervals, and posterior post-processing of parameters of interest. This talk will provide an overview of a Bayesian estimator of a discrete choice model with latent variables representing pro-environmental preferences. A case study on the adoption of electric vehicles, from a choice experiment in Germany focusing on the valuation of emission reductions, will highlight how Bayesian techniques enable efficient computation in this empirical context, especially in the presence of latent variables. The interaction between latent environmental concerns and willingness to pay for emission savings will be discussed, demonstrating the power of Bayesian methods in jointly estimating these latent preferences and their influence on decision-making. The presentation will conclude with a discussion of current research on enhancing the estimator through variational Bayes methods. Variational inference provides a scalable alternative to traditional Markov Chain Monte Carlo techniques by approximating the posterior distribution more efficiently. This approach significantly reduces computational time while maintaining a high degree of accuracy, particularly in large-scale models or those with high-dimensional latent variables. The potential of variational Bayes to further streamline Bayesian estimation in discrete choice models will be highlighted in the context of choices of smart electric vehicle charging.
Contact person: Anders Munk Nielsen.