# Deep Dynamic Factor Models

arXiv: https://arxiv.org/abs/2007.11887 · HTML: https://mlsanity.com/paper/2007.11887/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 23 Jul 2020.

## Built on by (most cited first, 6 of 6)

- [A Neural Phillips Curve and a Deep Output Gap](https://mlsanity.com/paper/2202.04146/) (2022)
- [Deep Learning-based Approaches for State Space Models: A Selective Review](https://mlsanity.com/paper/2412.11211/) (2024)
- [Non-linear dimension reduction in factor-augmented vector autoregressions](https://mlsanity.com/paper/2309.04821/) (2023)
- [Predicting Inflation with Recurrent Neural Networks](https://mlsanity.com/paper/2104.03757/) (2021)
- [Time-varying Factor Augmented Vector Autoregression with Grouped Sparse Autoencoder](https://mlsanity.com/paper/2503.04386/) (2025)
- [Conditionally Identifiable Latent Representation for Multivariate Time Series with Structural Dynamics](https://mlsanity.com/paper/2603.22886/) (2026)

## Abstract (arXiv)

A novel deep neural network framework -- that we refer to as Deep Dynamic Factor Model (D2FM) --, 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 D2FM improves over the performances of a state-of-the-art DFM.