Additional Online Supplement for “Inference for Approximate Factor Models: Random Missing and Cross Validation”

Liangjun Su,Ke Miao,Sainan Jin

2019-01-01

Abstract:Proof of Lemma A.1. From the principal component analysis (PCA), we have the identity ( NT q̃2 )−1 X̃X̃ ′F̃ = F̃ D̃. Pre-multiplying both sides by T−1F̃ ′ and using the normalization T−1F̃ ′F̃ = IR yield T−1F̃ ′ ( NT q̃2 )−1 X̃X̃ ′F̃ = D̃. The lemma follows providedD =plimD̃, which we show below. Noting that X̃ = (F 0Λ0′ + ε) ◦G, we have D̃ = T−1F̃ ′ ( NT q̃ )−1 X̃X̃ ′F̃ = T−1F̃ ′ ( NT q̃ )−1 [ (F 0Λ0′ + ε) ◦G ] [ (F 0Λ0′ + ε) ◦G ]′ F̃ = T−1F̃ ′ ( NT q̃ )−1 [ (F 0Λ0′) ◦G ] [ (F 0Λ0′) ◦G ]′ F̃ + T−1F̃ ′ ( NT q̃ )−1 (ε ◦G) (ε ◦G)′F̃ +T−1F̃ ′ ( NT q̃ )−1 [ (F 0Λ0′) ◦G ] (ε ◦G)′F̃ + T−1F̃ ′ ( NT q̃ )−1 (ε ◦G)′ [ (F 0Λ0′) ◦G ]′ F̃