Avaliação do value at risk do índice Bovespa usando os modelos garch, tarch e riskmetrics tm para se estimar a volatilidade

• Main Author: Farias Filho, Antonio Coelho Bezerra de
• Other Authors: Sicsú, Abraham Laredo
• Language: Portuguese
• Published: 2010
• Subjects: Risco de mercado; Volatilidade; Modelo GARCH; Modelo TARCH; Value at risk; RiskMetrics; Administração de empresas; Risco (Economia); Mercado financeiro - Modelos econométricos
• Online Access: http://hdl.handle.net/10438/4852

Made available in DSpace on 2010-04-20T20:14:59Z (GMT). No. of bitstreams: 0 Previous issue date: 1998-02-13T00:00:00Z === Apresenta o método value at risk (VaR) para se mensurar o risco de mercado, sob diferentes abordagens. Analisa a série histórica do índice Bovespa no período de 1995 a 1996 por meio de testes econométricos de normalidade, autocorrelação dos retornos e raiz unitária. Comparo valor obtido a partir dos diferentes modelos de estimação de volatilidade propostos e verifica qual dos modelos foi o mais adequado para o caso estudado. === The purpose of this dissertation is to compare the performance of three methods of volatility estimating used for value at risk models: an exponentially weighted moving average (RiskMetrics TM), GARCH (Generalized Autoregressive Conditional Heteroscedasticity) and TARCH (Threshold model). Concerning the latter, we decided to test it, given that GARCH models cannot properly capture the leverage etTect (negative shocks have a larger impact on volatility than positive shocks). The sample covers the daily São Paulo Stock Exchange index from 2 January 1995 to 30 December 1996. The test results indicated that the alternative models did not outperform RiskMetrics™ under the particular market conditions observed in the time period studied. Despite the fact that TARCH model can cope with negative or positive skewness, this model did not provide better results than RiskMetrics™. It seems to be reasonable not to attempt to make any general statement that one method is undoubtedly superior to another, given that test results may depend on the data period employed.