Factor models with many assets: Strong factors, weak factors, and the two-pass procedure

This paper re-examines the problem of estimating risk premia in unconditional linear factor pricing models. Typically, data used empirical literature are characterized by weakness some factors, strong cross-sectional dependence errors, and (moderately) high dimensionality. Using an asymptotic framework where number assets/portfolios grows with time span while exposures weak factors local-to-zero, we show that conventional two-pass estimation procedure delivers inconsistent estimates premia. We propose a new based on sample-splitting instrumental variables regression. The proposed estimator is robust to included presence unaccounted error dependence. prove consistency estimator, establish asymptotically valid inferences using Wald statistics, verify performance simulations, revisit studies.