Working Paper: CEPR ID: DP6043
Authors: Catherine Doz; Domenico Giannone; Lucrezia Reichlin
Abstract: This paper shows consistency of a two step estimator of the parameters of a dynamic approximate factor model when the panel of time series is large (n large). In the first step, the parameters are first estimated from an OLS on principal components. In the second step, the factors are estimated via the Kalman smoother. This projection allows to consider dynamics in the factors and heteroskedasticity in the idiosyncratic variance. The analysis provides theoretical backing for the estimator considered in Giannone, Reichlin, and Sala (2004) and Giannone, Reichlin, and Small (2005).
Keywords: factor models; kalman filter; large cross-sections; principal components
JEL Codes: C32; C33; C51
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| two-step estimator (C51) | consistency of the two-step estimator (C51) |
| OLS on principal components (C38) | captures the bulk of cross-sectional comovements driven by common factors (F62) |
| Kalman smoother (C32) | estimates common factors allowing for dynamics and heteroskedasticity (C51) |
| two-step estimator (C51) | achieves consistency even with misspecification (C62) |
| Kalman filter (C53) | treats unbalanced panels and reconstructs common shocks (C22) |
| proposed parametric approach (C59) | enhances estimation efficiency (C51) |
| proposed parametric approach (C59) | facilitates evaluation of uncertainty in factor estimates (C51) |