Course synopsis:
This course covers the new generation of discrete choice models and the simulation techniques that have allowed their development. It starts with the traditional models that can be estimated without simulation, namely, logit and nested logit. It then turns to the more powerful and flexible models that utilize simulation, including probit and mixed logit. Accept-reject, smoothed accept-reject, GHK, and direct simulators of the choice probabilities are developed and compared. Methods for taking draws from densities are explored, including several variance-reduction techniques such as antithetics, systematic sampling, and Halton draws. The course examines traditional maximum likelihood estimation as a step toward simulation-assisted estimation, including maximum simulated likelihood, method of simulated moments, and method of simulated scores. Recent advances in Bayesian procedures are covered, including the use of Markov Chain Monte Carlo methods (Gibbs sampling and the Metropolis-Hastings algorithm.) It shows how these Bayesian procedures can be useful for complex behavior that is not as amenable to classical methods and how the results of the Bayesian procedures can be interpreted from a classical perspective.

Caveats:
You will need to have RealPlayer 8 installed as a browser application to view the videos, and a system with enough memory to buffer the videostream in large chunks. Your system will also need a sound card if you want to listen to the audio portion of the lectures. Not recommended for people with slow modem connections.

If you have any questions, please send email to