Econometrics of Duration Data with Macroeconomic Applications
Nicholas Kiefer, Professor in the Dept of Statistical Sciences, Cornell University, USA, will be the Distinguished Guest Professor.
Topics covered by ASSEE 2017 included
- Duration and Survival Analysis
- Review of problems leading to models based on censored distributions of durations.
- Hazard functions and distributions.
- Likelihood functions, bank crisis data, NBER business cycle data, and parametric estimation.
- Proportional hazard (semiparametric) models.
- Heterogeneity in duration distributions.
- Competing risk modeling.
- The Counting Process Formulation
- Counting processes and martingales.
- Parametric and nonparametric estimation.
- Semiparametric (Cox-type) estimation.
- Bayesian methods for duration analysis.
Many packages including R, SAS, STATA, Matlab, etc. provide programs for analysis of survival data. The course will use R. R is available from cran.r-project.org. Many tested and well-documented (but not guaranteed) packages are available on that website. The course will use "survival." For Bayesian analysis we will also need "mcmc" and "coda." A useful frontend for R is R-Studio, available at http://www.rstudio.com. Excercises will use NBER business cycle data, downloadable from http://www.nber.org/cycles.html
Recommended for an introduction and background is Kiefer (1988). See also Neumann (1999). A book with more coverage than the articles is Lancaster (1992). These references appeared as duration analysis via hazard modeling was new to economics. A key reference of duration analysis of business cycles is Diebold and Rudebusch (1999). The article Gomez-Gonzalez and Kiefer (2009) discusses the Colombian bank failure data and application used in these lectures. These references cite many others. On the counting process approach see Andersen et al. (1985), which is an insightful survey giving also a useful bibliography. The key, simple, article on residual analysis for specification checking in the Cox model is Schoenfeld (1982). The Cox model is so well known that the original reference is rarely given. It is Cox (1972).