dfms
Efficient estimation of Dynamic Factor Models using the Expectation Maximization (EM) algorithm or Two-Step (2S) estimation, supporting datasets with missing data and mixed-frequency nowcasting applications. Factors follow a stationary VAR process of order p. Estimation options include: running the Kalman Filter and Smoother once with PCA initial values (2S) as in Doz, Giannone and Reichlin (2011) <doi:10.1016/j.jeconom.2011.02.012>; iterated Kalman Filtering and Smoothing until EM convergence as in Doz, Giannone and Reichlin (2012) <doi:10.1162/REST_a_00225>; or the adapted EM algorithm of Banbura and Modugno (2014) <doi:10.1002/jae.2306>, allowing arbitrary missing-data patterns and monthly-quarterly mixed-frequency datasets. The implementation uses the 'Armadillo' 'C++' library and the 'collapse' package for fast estimation. A comprehensive set of methods supports interpretation and visualization, forecasting, and decomposition of the 'news' content of macroeconomic data releases following Banbura and Modugno (2014). Information criteria to choose the number of factors are also provided, following Bai and Ng (2002) <doi:10.1111/1468-0262.00273>.
- A Quasi–Maximum Likelihood Approach for Large, Approximate Dynamic Factor Models
- A two-step estimator for large approximate dynamic factor models based on Kalman filtering
- Determining the Number of Factors in Approximate Factor Models
- MAXIMUM LIKELIHOOD ESTIMATION OF FACTOR MODELS ON DATASETS WITH ARBITRARY PATTERN OF MISSING DATA
This page was built for software: dfms
