Advanced econometrics: Panel data and duration models


Objectif

  • Refresher and further study of linear panel data models with large N and fixed T: strong and weak exogeneity, state dependence, properties and specification of correlated random effects models, clustering and quasi maximum likelihood approach.
  • Non-linear panel data models for large N and fixed T: focus on binary choice models (logit, probit), incidental parameters, estimation of parameters of interest (index and average marginal effects), sufficient statistics approach and (quasi) maximum likelihood. 
  • Duration Model: Discrete and continuous time duration models including the analysis of competing risks, mixed proportional hazard rate models and timing of events models.

For the panel data part, students are supposed to :
–          Understand the utility of panel data for identification of causal effect,
–          Understand well the difference between fixed and random effects estimators,
–          be able to compute such estimators,
–          understand how to build inference robust to serial correlation,
–          link identification assumptions with testable implications,
–          check the testable implications with the data,
–          deduce some properties of the data generating process from the comparison of various estimators,
–          program Stata for this purpose, be able to read and interpret Stata output,
–          be able to proove theoretical properties of the estimators (asymptotic properties),
–          understand the incicental parameter problem and the usual way to fix it,
–          choose the more relevant estimator depending on the case studied.

For the duration model part, students are supposed to :
– Understand the concept of duration analysis and hazard rates
– Understand the role of dynamic selection in duration models
– Derive the Maximum-Likelihood Estimators of different duration models
– Be able to estimate different types of duration models with Stata and interpret the results
– Understand and estimate competing risks models
 

Plan

1-Refresher and further study of linear panel data models with large N and fixed T: strong and weak exogeneity, state dependence, properties and specification of correlated random effects models, clustering and quasi maximum likelihood approach.

2-Non-linear panel data models for large N and fixed T: focus on binary choice models (logit, probit), incidental parameters, estimation of parameters of interest (index and average marginal effects), sufficient statistics approach and (quasi) maximum likelihood. 

3-Duration Model: Discrete and continuous time duration models including the analysis of competing risks, mixed proportional hazard rate models and timing of events models.

Références

  • Wooldridge : Econometric Analysis of Cross Section and Panel Data, second edition (mainly chapters 10, 11, 13, 15 and 22).
  • Lancaster (1990): The Econometric Analysis of Transition Data 
  • Cameron and Trivedi (2005): Microeconometrics, Methods and Applications, Chapters 17,18,19.