A1554
Title: Forecasting with dynamic factor models estimated by partial least squares
Authors: Samuel Rauhala - University of Turku (Finland) [presenting]
Abstract: Dynamic factor models (DFMs) have found great success in nowcasting and short-term macroeconomic forecasting. The factor loadings are typically estimated cross-sectionally, for example, with principal components. This ignores whether or not the factors have predictive capabilities. Two alternative approaches are suggested, using partial least squares, which takes the time series structure better into account. The first one is close akin to a large vector autoregression, and the second is more akin to a conventional DFM. Monte Carlo exercises are conducted, and it is found that in finite samples, this method outperforms the typical methods, such as the two-step estimator and quasi-maximum-likelihood.
