Value at risk intradiário, modelos de volatilidade condicional e distribuições de probabilidade: evidências para o Ibovespa

• Ano de defesa: 2018
• Autor(a) principal: Oliveira, Denise Correia de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Brasil; UFRN; PROGRAMA DE PÓS-GRADUAÇÃO EM ADMINISTRAÇÃO
• Palavras-chave em Português: Dados intradiários; Modelos de volatilidade; Value at risk; CNPQ::CIENCIAS SOCIAIS APLICADAS::ADMINISTRACAO
• Link de acesso: https://repositorio.ufrn.br/jspui/handle/123456789/25939

This research aims to update and expand the investigation found in Lemgruber and Moreira (2004) and Cappa and Valls Pereira(2010) referring on the use of high frequency data in the estimation of daily and intraday volatility of the IBOVESPA and its subsequent application to the pricing on 1the value at risk (VaR). The models of the ARCH family of short and long memory are estimated from four distributions (Normal, Student t, Student t-asymmetric and GED) and are used in conjunction with deterministic methods of intraday seasonal filtering and by day of the week according to the method proposed by Taylor and Xu (1997) for predicting volatility and intraday VaR. Moreover, the Daily returns of the Ibovespa were used as well as the returns composed continuously of 5, 10 and 60 minutes for the intraday series. The data window comprises the period from January 2012 to September 2015.The results suggest the presence of long memory in the intraday returns according to the estimates of conditional volatility models with emphasis on the models (FIGARCH and FIAPARCH). Regarding the intraday seasonality filtering, it was verified that there was superior performance in the predictive quality of VaR for some of the models estimated with intraday seasonal filter; however, these improvements were marginal. In relation to the frequency of data, the results indicate through Kupiec's backtesting that the predictability power for the VaR with daily series is more stable and has better quality than with the use of high frequency data.