Econometrics of High Dimensional Data: Pitfalls and Opportunities

Rich data sets with a whopping number of covariates require new approaches to variable selection, estimation and inference. We present a partial survey on modern econometric shrinkage methods which are capable to select the appropriate model specifications for within-sample inference and forecasting when the number of covariates and specifications is prohibitively large or if the number of regressors exceeds the number of observations. Empirical applications will be presented for IV-based causal inference, estimation of peer effects, portfolio analysis and labor economics.