P.A.R.T

PART_IDENT - " Identification Partielle de Modèles Economiques Structurels"

ANR -2011-BSH1-C04-01
 

Summary:

From the very start of structural modeling, identification meant point identification of the structural parameters. There is a single "true" value of parameters which is compatible with the reduced form. Dispersed in the literature though, there are examples of weaker concepts of identification. Specifically, all values of the structural parameters in intervals or regions could be compatible with the reduced form. It is a richer setting that the traditional one for identification since it allows us to relax the strict requirements leading to point identification and study how certain restrictions limit the degree of partial identification. There are two aspects in an application that can give rise to set or partial identification. On the one hand, information on some variables might be missing. On the other hand, structural equations might not generate enough moment restrictions or might imply inequality restrictions only.

This project is intended to produce research that belongs to the growing recent literature on partial identification. This literature has initiated some new perspectives for estimating structural models. Furthermore, this change of paradigm moving from a point to a set-identified setting is conceptually innocuous from the point of view of an applied economist because the quantities of concern are the confidence regions for structural parameters. The construction of those confidence regions does not differ across the two settings provided that this construction encompasses partial and exact identification. The specificity of this project is the blend between specialized research in certain fields of applications like micro-econometrics, time series and industrial organization and a general theoretical setting seeking to identify the conceptual difficulties that each specialized research encounters. Bundling these various researches into one project allows us to exploit scale economies in the development of theoretical tools. It also allows us to investigate the portability of theoretical results to applied fields and checks the tractability of the theoretical approaches for applied economists. It is why the project is divided into three tasks.

The first task aims at developing investigations in Financial Econometrics and turn around the estimation of the Euler equation for asset prices or for the marginal utility of consumption.

The second task of the project deals with the econometrics of industrial organization and specifically models of firm entry which is the prototype model for partial identification.

Finally, in the last task of the project, our object of interest is the dynamics of earnings in France as they can be observed using very long individual panel data series.

Our project is to construct counterfactuals and analyze how attrition can be cast in a partial identification framework. More abstract research is needed and will be conducted. First, the previous developments that we aim at developing in these three tasks beg the question of the conceptual understanding of what are the characteristics of models which lead to partial identification. Second, it seems crucial to us to characterize models in which convexity assumptions hold since this case leads to much simpler estimation methods. It is also worth investigating what these boundedness and convexity assumptions can bring to the analysis of the moment inequality set-up which is the main model used by theoretical econometricians and statisticians to develop estimation and inference methods. The main issue there is the uniformity issue with respect to the partial or point-identified character of the underlying model. The question we hope to be able to deal with is what bounded and convexity assumptions or more generally any additional structure that the economic model could bring upon the case of interest, bring to this debate.

Project date: 01/09/2011 -  31/08/2015

 


Contact in TSE: Christian BONTEMPS & Thierry MAGNAC