Artificial Neural Networks and Automatic Time Series Analysis, methodological approach, results and examples using health-related time series

Time series modeling by the use of automatic signal extraction methods has been widely studied and used in different contexts of economic analysis. The methodological innovation of ARIMA / SARIMA models estimation made significant contributions to the understanding of temporal dynamics of events, even when the time structure was apparently irregular and unpredictable. The popularity of these models was reflected in the development of applications that implemented algorithms that automaticaly extract temporal patterns of the series and provide a reasonably accurate adjustment by a mathematical model, making it also in a quick and consistent manner. One of the most common use of these programs is in the univariate analysis context, to achieve its filtering for its posterior use in a multivariate structure. However, there is significant untapped potential in the results provided by those applications. In this paper there's a description of the methodology with which the use of TRAMO SEATS and X13 ARIMA is implemented directly in a multivariate structure. Specifically, we have applied data analysis techniques related to artificial neural networks. UNder the neural networks philosophy, events are conceived as linked nodes which activate or not depending on the intensity of an imput signal. At that point come into play STRETCH or X13. To illustrate the methodology and the use of the model, series of health-related time are used, and a consistent model able to "react" to the dynamic interrelations of the variables considered is described. Standard panel data modeling is included in the example and compared with the new methodology.