Deep Dynamic Factor Models
This is an incremental improvement for economists and analysts needing more accurate time-series predictions.
The paper tackles the problem of encoding hundreds of macroeconomic and financial time-series into a few latent states for nowcasting and forecasting, and shows that the Deep Dynamic Factor Model (D^2FM) improves performance over a state-of-the-art dynamic factor model in real-time US data and Monte Carlo experiments.
A novel deep neural network framework -- that we refer to as Deep Dynamic Factor Model (D$^2$FM) --, is able to encode the information available, from hundreds of macroeconomic and financial time-series into a handful of unobserved latent states. While similar in spirit to traditional dynamic factor models (DFMs), differently from those, this new class of models allows for nonlinearities between factors and observables due to the autoencoder neural network structure. However, by design, the latent states of the model can still be interpreted as in a standard factor model. Both in a fully real-time out-of-sample nowcasting and forecasting exercise with US data and in a Monte Carlo experiment, the D$^2$FM improves over the performances of a state-of-the-art DFM.