Nonparametric adaptive estimation in linear regression models with the unknown error distribution
林金官教授
东南大学数学系
时间: 11月29日上午9:00-10:00
地点: 博纳楼5层第七会议室
Abstract：For normally distributed errors, the maximum likelihood estimate (MLE) is exactly the least squares estimate (LSE) in linear regression models. In the absence of Gaussianity, the MLE is more e®ective than the LSE. However, in practice, the error distribution is generally unknown, and the MLE is infeasible. In this talk, we propose an nonparametric adaptive method to estimate parameters of a linear regression model with unknown error distribution. We show that the resulting estimator is asymptotically as efficient as the oracle MLE, which assumes the error distribution were known. It offers a unified approach to nonparametric inference, including hypothesis testing and diagnostic analysis. Finally, we present some simulations and real data analysis, which show the proposed method is comparable to the traditional LSE and MLE.
Modelling spatial time series by graphical models
李元教授 
广州大学经济与统计学院
时间: 11月29日上午10:00-11:00
地点: 博纳楼5层第七会议室
Abstract：We propose the spatial temporal autoregressive models using a graphical approach. Our model extends the STARIMA model in the sense that ours does not require prior knowledge about the graph or weight matrices and   can be applied to multivariate cases. Compared with VAR models, our models are parsimonious in parameters and the structure of the covariance matrix in the model is largely simplified. Based on the concept of Granger causality, we first define the spatial temporal chain graph for spatial time series. With the chain graph, the spatial temporal autoregressive model is constructed. Model building procedures are established by selection of graphs and Bayesian method. Simulation results and an application to the study of air pollution in the Pearl River Delta of China are reported.