Value at Risk prediction model using long term volatility - DOI: http://dx.doi.org/10.16930/2237-7662/rccc.v15n45p23-33

• Autor(a) principal: Maia, Vinicius Mothé
• Data de Publicação: 2016
• Outros Autores: Monteiro, Igor Swinerd, Pinto, Antonio Carlos Figueiredo, Klotzle, Marcelo Cabus
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Revista Catarinense da Ciência Contábil (Online)
• Texto Completo: https://revista.crcsc.org.br/index.php/CRCSC/article/view/2206
• topic: Value at Risk; Volatility; GARCH; ARLS.; Value at Risk; Volatilidade; GARCH; ARLS.

Having in mind the importance of Value at Risk (VaR) as a risk measure for financial institutions and rating agencies, this study evaluated whether the ARLS model is more accurate in the calculation of the long term VaR than the traditional models, considering it is more appropriate for predicting the long-term volatility. Due to the fact that VaR s being used for market players as a measure of risk for the portfolio management, its proper measurement is important. Based on daily data from the stock markets and exchange of BRICS (Brazil, Russia, India, China and South Africa) future volatilities for 15 days, 1 month and 3 months ahead were calculated. Then the traditional measures of VaR accuracy were calculated. The results suggest the superiority of ARLS model for predicting the exchange rate volatility, being able to  predict precisely the number of violations in 33% of cases, while traditional models did not perform well. Regarding the stock market, GARCH and ARLS models showed similar performance, with higher accuracy of the GARCH model considering the quadratic average loss function. These results have shown that the choice of ARLS model in VaR calculation to currency portfolios is better due to higher achieved accuracy, thus helping market participants to better manage the risk of their portfolios. In relation to the stock market, considering the similar performance of GARCH and ARLS models, the GARCH model is more suitable because of its greater simplicity and easy computational implementation.