Please use this identifier to cite or link to this item: http://repositorio.unicamp.br/jspui/handle/REPOSIP/326552
Type: Artigo
Title: Copula-based Prediction Of Economic Movements
Copula-based prediction of economic movements
Author: García, J. E.González-López, V. A.Hirsh, I. D.
Abstract: In this paper we model the discretized returns of two paired time series BM&FBOVESPA Dividend Index and BM&FBOVESPA Public Utilities Index using multivariate Markov models. The discretization corresponds to three categories, high losses, high profits and the complementary periods of the series. In technical terms, the maximal memory that can be considered for a Markov model, can be derived from the size of the alphabet and dataset. The number of parameters needed to specify a discrete multivariate Markov chain grows exponentially with the order and dimension of the chain. In this case the size of the database is not large enough for a consistent estimation of the model. We apply a strategy to estimate a multivariate process with an order greater than the order achieved using standard procedures. The new strategy consist on obtaining a partition of the state space which is constructed from a combination, of the partitions corresponding to the two marginal processes and the partition corresponding to the multivariate Markov chain. In order to estimate the transition probabilities, all the partitions are linked using a copula. In our application this strategy provides a significant improvement in the movement predictions.
In this paper we model the discretized returns of two paired time series BM&FBOVESPA Dividend Index and BM&FBOVESPA Public Utilities Index using multivariate Markov models. The discretization corresponds to three categories, high losses, high profits and the complementary periods of the series. In technical terms, the maximal memory that can be considered for a Markov model, can be derived from the size of the alphabet and dataset. The number of parameters needed to specify a discrete multivariate Markov chain grows exponentially with the order and dimension of the chain. In this case the size of the database is not large enough for a consistent estimation of the model. We apply a strategy to estimate a multivariate process with an order greater than the order achieved using standard procedures. The new strategy consist on obtaining a partition of the state space which is constructed from a combination, of the partitions corresponding to the two marginal processes and the partition corresponding to the multivariate Markov chain. In order to estimate the transition probabilities, all the partitions are linked using a copula. In our application this strategy provides a significant improvement in the movement predictions.
Subject: Partition Markov Models
Inference Methods
Copula For Discrete Data
Cópulas (Estatística matemática)Markov, Processos deAnálise de séries temporais
Country: Estados Unidos
Editor: AIP Publishing
Citation: Proceedings Of The International Conference On Numerical Analysis And Applied Mathematics 2015 (icnaam-2015). Amer Inst Physics, v. 1738, p. , 2016.
Rights: fechado
Identifier DOI: 10.1063/1.4951928
Address: http://aip.scitation.org/doi/10.1063/1.4951928
Date Issue: 2016
Appears in Collections:Unicamp - Artigos e Outros Documentos

Files in This Item:
File SizeFormat 
000380803300172.pdf163.79 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.