Long memory in high frequency time series using wavelets and conditional volatility models

• Ano de defesa: 2021
• Autor(a) principal: Pinto, Mateus Gonzalez de Freitas
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Asset returns; Dados de alta frequência; Dados intradiários; FIGARCH; High frequency data; Intraday data; Long memory; Memória longa; Ondaletas; Retornos; Volatilidade; Volatility; Wavelets
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/45/45133/tde-06052021-100559/

The goal of this dissertation is to describe a methodology for modelling the volatility of high frequency financial data, considering its features and stylized facts. In order to account for the long-range dependence in conditional mean and conditional variance, ARFIMA and FI(E)GARCH models are used respectively, when observed. To account for the non-normality, skeweness and kurtosis, features observed in the the distribution of the log-returns in high frequency, the Skewed Student t and the Generalized Error Distribution (GED) are adopted for the innovation term of the aforementioned models. Wavelet shrinkage is used in a non-parametric identification and separation of the intraday jumps from the time series data. The application of this procedure is presented using real high frequency asset returns from the Brazilian Exchange and OTC, as well as exchange rates from cryptocurrencies traded in Crypto Exchanges.