Abstract: Normal copula underestimates extreme events due to its asymptotic independence property. In the first part of this talk, we show that dynamic normal copulas are able to catch both asymptotic independence and asymptotic dependence so as to predict extreme events accurately. Further we propose both parametric and nonparametric inference procedures for the involved correlation function. In the second part of this talk, we will provide a uniform test for testing predictability for a regression model with dependent AR(p) errors rather than independent errors. We propose empirical likelihood method for the case of small p. When p is large, empirical likelihood method is quite computationally intensive. So we further propose a jackknife empirical likelihood method to reduce computation. Simulation study shows the proposed methods are effective. |
