Title: Macroeconomic forecasting with fractional factor models
Authors: Tobias Hartl - University of Regensburg (Germany) [presenting]
Abstract: Instead of pre-differencing time series for the estimation of dynamic factor models, the use of models that incorporate common fractionally integrated unobserved components in levels is suggested. Three frameworks that allow for long-range dependence both in the common components and idiosyncratic errors are derived. In these models the factors either establish fractional cointegration relations, or they can be eliminated by taking non-integer differences. A two-stage estimator, that combines principal components and the Kalman filter, is proposed. The forecast performance is studied for a large macroeconomic dataset for the US, where we find that benefits from the fractional factor models can be substantial, as they outperform univariate autoregressions, principal components in integer differences, a combination of both and factor-augmented error-correction models.