Title: Fast solution to the automatic identification of unobserved components models
Authors: Diego Pedregal - University of Castilla-La Mancha (Spain) [presenting]
Juan R Trapero - Universidad de Castilla-La Mancha (Spain)
Abstract: A fast algorithm and software for the automatic identification of Unobserved Components models are presented. Solutions of this sort is compulsory nowadays once the big data era has come and is staying among us. Crafted identification procedures are reserved for problems in pure scientific contexts where the number of time series are manageable, but are not practical for organizations that want to process a tsunami of information efficiently, online and in record time. Automatic identification tools are the usual way to deal with modeling problems in many contexts, typically Machine Learning and some statistical areas, but it has never been tried out in Unobserved Components models (UC). A new piece of software is introduced with is developed in R with the core written in C++ with the help of RcppArmadillo for the automatic identification and forecasting of UCs with some enhancing features. The package searches among many combinations of different specifications for each of the components according to a stepwise algorithm to reduce the universe of possible models and selects the one with the best metrics. The forecasting results suggest that UC models are powerful potential forecasting competitors to other well-known methods both in computation time and forecasting accuracy. Though there are several pieces of software available for UC modeling, this is the first implementation of an automatic algorithm for this class of models, to the authors' knowledge.