Forecasting and Disentangling Time Series with Dynamic Factor Models
This page is outdated since meta-residuals can now be used. Please check the corresponding publication.
Introduction
"...coherent assumptions on what is still invisible may increase our understanding of the visible."
Jean Baptiste Perrin in his Nobel Prize lecture
That’s exactly what dynamic factor models (DFM) achieve by extracting a few unobserved, latent factors from a large number of time series. Those factors can be used for forecasting or for disentangling the dynamic of the model. Here is an example for a marketing scenario, where the impact of advertising is decomposed into
(1. factor) increased consumer interest in seeking information about a product and
(2. factor) conversion of consumer interest into sales.
Introduction
"...coherent assumptions on what is still invisible may increase our understanding of the visible."
Jean Baptiste Perrin in his Nobel Prize lecture
That’s exactly what dynamic factor models (DFM) achieve by extracting a few unobserved, latent factors from a large number of time series. Those factors can be used for forecasting or for disentangling the dynamic of the model. Here is an example for a marketing scenario, where the impact of advertising is decomposed into
(1. factor) increased consumer interest in seeking information about a product and
(2. factor) conversion of consumer interest into sales.
Advantages
If the factor structure is existing in the data (which can’t be imposed), forecasts can be improved compared to classical (vector) auto-regressive forecasts. Some more advantages of DFM are:
If the factor structure is existing in the data (which can’t be imposed), forecasts can be improved compared to classical (vector) auto-regressive forecasts. Some more advantages of DFM are:
- A small number of factors can explain dynamic co-movements of many series.
- Sometimes our (visible) observations are indicating a few (invisible) driver, thus making DFM a tool for consistent theory.
- Can be used for real-time monitoring and nowcasting.
- Handle (arbitrarily) large data sets .
- The accelerated digitization of business will lead to some of the companies using Structural Vector Autoregressive (SVAR) Analysis, where, among others, the effect of a sudden change of one variable to another is examined. A condition is to include all relevant information, otherwise the results (impulse responses) are inaccurate. DFM can fulfill this condition since they can use the information of the entire system.
Application Spectrum
DFM were used in a variety of areas such as process monitoring, energy market, telecommunications market, marketing and business research, environmental science, hydrology, medical science, psychology, macroeconomics, astrophysics, genomics and for analyzing emotion and affect.
Here is a short video where we show an industry 4.0 scenario. Machine sensor data is forecasted with DFM and given to another machine for real-time process monitoring. Here are the slides shown with publications from different areas. Here are some theory slides.
DFM were used in a variety of areas such as process monitoring, energy market, telecommunications market, marketing and business research, environmental science, hydrology, medical science, psychology, macroeconomics, astrophysics, genomics and for analyzing emotion and affect.
Here is a short video where we show an industry 4.0 scenario. Machine sensor data is forecasted with DFM and given to another machine for real-time process monitoring. Here are the slides shown with publications from different areas. Here are some theory slides.
Contact
If you are interested in using DFM for your business, please contact us under [email protected].
We are a Berlin based company where strict data protection laws apply. It is possible to transfer data through a Swiss cloud storage provider with end-to-end encryption.
If you are interested in using DFM for your business, please contact us under [email protected].
We are a Berlin based company where strict data protection laws apply. It is possible to transfer data through a Swiss cloud storage provider with end-to-end encryption.
