The idea for reading group emerged when I was asked to
comment an applied political
economy paper using a Bayesian approach to select variables. More
precisely, the
econometric question in this paper was to select a model when the list
of
variables was longer than the number of observations. A Bayesian
methodology was
adopted. This paper was an illustration of the important question of
model
choice and model selection. A not well suited procedure could lead to
biased
results.
The proposed reading group is based on three lists of papers: a first list concerns Bayesian theoretical groundings and the second list concerns applications from the economic literature. The last list contains a single paper which is devoted to the implementation of the procedures in R.
I am
thus going to give you these papers, with
some comments each time. You can access the paper by clicking on it.
This reading group is designed for PhD students, but all
reaserachers in terested in econometrics are welcome.
Bayesian
theory
I
suppose that you are not familiar with Bayesian
inference. It is always difficult to give a first and concise approach
to the field. For those who have already followed my lecture on the
Econometrics of poverty and inequality measurement, a short
introduction to the topic was given, but it has certainly to be
completed. We can proceed in two different ways. You can try to read my
text book, especially the first chapter. It should be available at the
library. I can also give you a pdf file, but not directly on the web:
Bauwens L., M. Lubrano and J.F. Richard (1999) Bayesian inference in dynamic econometric models. Oxford University Press. Chapter 1: this chapter is an introduction of Bayesian theory and its differences with the classical approach.
Or, we can decide to read directly the founding paper of the field of model selection. This paper was published in a sociological journal, so it should be understood by readers having a soft varnish in statistics and econometrics. I think I prefer this way. Then I could try to treat more specific problems when needed.Adrian E. Raftery (1995) Bayesian Model Selection in Social Research. Sociological Methodology, Vol. , 111-163. This is one of the founding papers on Bayesian model choice.
We can try to play with the data that are used in this paper (this is a data file that can be read directly in R). I have collected various testing procedures using these data in the following R file. I hope that you all know R, because there are very nice packages for Bayesian inference and model selection and plenty of other econometric questions:
Then, we have to make a choice.We could go directly to recent economic papers where these techniques of model selection were used. I have collected four of them:
Xavier Sala-i-Martin, Gernot Doppelhofer and Ronald I. Miller (2004) Determinants of Long-Term Growth: A Bayesian Averaging
of Classical Estimates (BACE)
Approach. The American Economic Review, Vol. 94, No. 4 (Sep., 2004), pp. 813-835
Eicher, T.S., C. Papageorgiou and O.
Roehn. 2007. “Unraveling the
Fortunes of the Fortunate: An Iterative
Bayesian
Model Averaging (IBMA) Approach,” Journal of Macroeconomics 29(3),
494-514.
Lesage J.P. and M.M. Fischer (2008) Spatial growth regressions: Model specification, estimation and interpretation. Spatial Economic Analysis, 3(3), 275-304.
You can notice that one author is recurrent. Or we can
continue with statistical papers, which appear as more advanced like:
Kass, R.E. and A.E. Raftery. (1995).
“Bayes Factors,” Journal of the American Statistical Association, 90(430),
773–795.
The last paper is in fact mandatory, because it clearly
treat the question that we are occupied in: the case when there are
more variables than observations. It was published in a Bio journal:
Yeung, K.Y., R.E. Bumgarner and A.E.
Raftery. 2005. “Bayesian Model Averaging:
Development of an Improved
Multi-class, Gene Selection and Classification Tool for Microarray
Data,'' Bioinformatics, 21(10),
2394-2402.
Adrian E. Raftery, Ian S. Painter and Christopher T. Volinsky (2005) BMA: An R package for Bayesian Model
Averaging. R News, 5(2): 2-8.