Temos o prazer de convidar a todos para mais um seminário do Programa de Pós-Graduação em Estatística da UFRGS (PPGEst-UFRGS).
Título: Forecasting Volatility via Dynamic Copula-Based Regressions.
Resumo: Realized volatility series are generally found to exhibit very slowly decaying autocorrelation functions, suggesting the use of long-memory models. The linear HAR regression model is well known for its ability to parsimoniously model the long-memory behavior of realized volatility. Besides, there is vast empirical evidence of its good forecasting performance. We propose an alternative approach to the linear HAR model, considering a dynamic vine copula based regression capable of capturing long-memory and nonlinearity in the time series dependence of realized volatility, and it also allows for the possibility of changes in the dependence structure over time. More specifically, we derive a regression from a dynamic D-vine decomposition of the joint density of the dependent and explanatory variables of the HAR model and compute point forecasts of volatility for the one-day horizon from this regression model. We evaluate the forecasting ability of the dynamic HAR-D-vine model in an empirical study. Our data sets consist of realized volatilities from the S&P 500, NASDAQ 100, DOW JONES, FTSE 100, CAC and HANG SENG stock indexes. Regarding the main results, we find that both static and dynamic D-vine copula-based HAR regressions and the linear HAR model have a statistically equivalent forecast performance, while they outperform the bivariate copula model.
Data: 21/10/2025
Horário: 13:30
O seminário acontecerá na modalidade Online e poderá ser assistido pelo link: https://mconf.ufrgs.br/webconf/ppgest