ACD模型自加权M-估计的渐近性质及其应用OA

The self-weighted M-estimation(SM-estimation)for parameters of autoregressive conditional duration(ACD)model is proposed,and the strong consistency and asymptotic normality of this estimation are shown by allowing the variance of model errors to be infinite.The performances of SM-estimation,(quasi-)maximum likelihood estimation and least absolute deviation estimation are compared through numerical simulations,which demonstrate that the SM-estimation has the most robust performance when data exhibit heavy-tailed characteristics.Finally,an application of SM-estimation to the price duration modeling of Haier Smart Home stock is demonstrated,and the empirical results indicate that the SM-estimation outperforms other estimation methods.