zbMATH — the first resource for mathematics

On residual empirical distribution functions in ARCH models with applications to testing and estimation. (English) Zbl 0921.62063
Let \(\{y_i\}\) be an ARCH(1) model, i.e. \[ y_i=\sigma_i \varepsilon_i,\quad\sigma^2_i= \alpha_0+ \alpha_1y^2_{i-1}, \quad i\in\mathbb{Z}, \] where the \(\{\varepsilon_i\}\) are i.i.d. random variables with unknown distribution function \(G(x)\), \(E\varepsilon^2_1=1\), \(\alpha_0>0\), \(0\leq\alpha_1 <1\); \(\alpha= (\alpha_0,\alpha_1)\) is an unknown parameter vector. The author considers the residuals \(\varepsilon_k(\theta)= y_k(\theta_0+ \theta_1y^2_{k-1})^{-1/2}\), \(k=1,\dots,n\), and the residual empirical distribution function \(G_n(x,\theta)= n^{-1}\sum^n_{k=1} I(\varepsilon_k (\theta)\leq x)\). Then it is proved that \[ \sup_{x\in R^1, |\tau|\leq \theta}\biggl| n^{1/2}\bigl[ G_n(x,\alpha+ n^{-1/2} \tau)-G_n(x,\alpha) \bigr]-2^{-1}xg(x){\mathbf E}[(\tau_0+ \tau_1 y_1^2)/(\alpha_0+ \alpha_1y^2_1) ]\biggr|=o_p(1). \] This interesting result is used for testing and robust rank estimation of unknown parameters.

62G30 Order statistics; empirical distribution functions
62G35 Nonparametric robustness
62M10 Time series, auto-correlation, regression, etc. in statistics (GARCH)
62G05 Nonparametric estimation