MDFA: R-Code, Tutorials, and References

The MDFA (Multivariate Direct Filter Approach) is a novel prediction approach which generalizes classic linear (maximum likelihood, one-step ahead mean-square error) forecast and signal extraction approaches. It is based on closed-form (fast) optimization criteria with unique solutions based on simple principles. It can account for simple one-step ahead forecasting as well as for more complex real-time signal extraction (trend, cycle nowcast/forecast) problems. It fully integrates a fundamental Trilemma between Accuracy (predicting the future level), Timeliness (lead, advancement) and Smoothness (suppression of spurious noise) of a predictor. 

 

MDFA-tutorial denotes a R-package of examples and use cases introducing to and illustrating the MDFA: it is hosted on a github repository. The proper MDFA R-package is also hosted in a (separate) Github repository: the tutorials load that package to work through the examples.

 

The method is explained in a series of articles: working paper versions are available through github. 

 

Click on the buttons below to either access (a description of) the MDFA tutorial project or to obtain background information.