Asymptotics of rank order statistics for ARCH residual empirical processes. (Q2574560)

Let \(\{\varepsilon _t\}\), \(\{\xi _t\}\) be independent sequences of i.i.d. \((0,1)\) random variables. Consider ARCH processes \(\{X_t\}\) and \(\{Y_t\}\) given by \(X_t=\sigma _t(\theta _X)\varepsilon _t\), \(\sigma _t^2(\theta _X) = \theta _X^0+\sum _{i=1}^{p_x} \theta _X^i X_{t-i}^2\) \((t=1,\dots ,m)\) and \(Y_t=\sigma _t(\theta _Y)\xi _t\), \(\sigma _t^2(\theta _Y) = \theta _Y^0+\sum _{i=1}^{p_y} \theta _Y^i Y_{t-i}^2\) \((t=1,\dots ,n)\), respectively. The vectors of unknown parameters are \(\theta _X=(\theta _X^0,\dots ,\theta _X^{p_x})'\) and \(\theta _Y =(\theta _Y^0,\dots ,\theta _Y^{p_y})'\). Their conditional least-squares estimators are denoted by \(\hat \theta _X\) and \(\hat \theta _Y\), respectively. Then the empirical squared residuals are \(\hat {\varepsilon }_t=X_t^2/\sigma _t^2(\hat {\theta }_X)\) and \(\hat {\xi }_t=Y_t^2/\sigma _t^2(\hat {\theta }_Y)\). Let \(F\) and \(G\) be the distribution functions of \(\varepsilon _t^2\) and \(\xi _t^2\), respectively. The authors investigate the two-sample problem of testing \(H_0:F(x)=G(x)\) for all \(x\). The asymptotic theory of a class of rank order statistics \(\{T_N\}\) is developed. The limiting distribution of \(\{T_N\}\) is different from that of ARMA case. Numerical illustrations show the reliability of confidence intervals, asymptotic relative efficiency, and ARCH affection. A measure of robustness for \(\{T_N\}\) by means of the influence function is introduced.