Essays in econometrics and random matrix theory

Abstract

This dissertation develops new econometric procedures for the analysis of high-dimensional datasets commonly encountered in finance, macroeconomics or industrial organization. First, I show that traditional approaches to the estimation of latent factors in financial data underestimate the number of risk factors. They are also biased towards a single market factor, the importance of which is overestimated in samples. In Chapter 3, I derive a new consistent procedure for the estimation of the number of latent factors by examining the effect of the idiosyncratic noise in a factor model. Furthermore, I show that the estimation of factor loadings by Principal Components Analysis is inconsistent for weak factors and suggest alternative Instrumental Variables procedures. Chapter 4 uses the theoretical results of the earlier chapters to estimate the stochastic dimension of the US economy and shows that global risk factors may obfuscate the relationship between inflation and unemployment. Chapter 5 (co-authored with Jerry Hausman) suggests a new procedure for the estimation of discrete choice models with random coe±cients and shows that ignoring individual taste heterogeneity can lead to misleading policy counterfactuals.