D. C. Gamboa Pinilla, A. M. Alonso, D. Peña
In this paper we work with a measure known as generalized cross correlation (GCC) in order to cluster time series by using dependencies [3]. In particular, we study the conditional heteroskedastic case analyzing the ARCH(p) model. Based on the structure of the ARCH(p) models, we work with cross correlations between the squared residuals of ARMA(p, q) models. Some Monte Carlo experiments are developed in order to analyze the improvement obtained when we are working with squared residuals instead their levels, and we found that we are able to recover the
original clustering structures in all cases studied. Finally, we show this methodology in a set of real data.
Palabras clave: Unsupervised learning , Correlation matrix , Correlation coefficient , ARCH(p) models
Programado
EET-1 ST-2 Estadística Espacial y Espacio Temporal. Series Temporales
4 de septiembre de 2019 14:45
I2L5. Edificio Georgina Blanes