一、主题:Time-varying Model Averaging via Adaptive LASSO
二、主讲人:孙玉莹,中国科学院数学与系统科学研究院助理研究员。2016年获得中国科学院大学管理学博士学位,获得中国科学院数学与系统科学研究院“重要科研进展奖(2017,2019)”、系统所关肇直青年研究奖,中国管理科学与工程学会优秀博士学位论文奖(2019)。长期从事经济预测理论与方法研究,在国内外重要期刊发表论文10余篇,包括Journal of Econometrics, Energy Economics, Quantitative Finance, China Economic Reviews,被Journal of Management Science and Engineering 邀请为Special Issue Guest Editor之一。在经济政策分析领域也做出了多项有影响的研究工作,作为主笔撰写政策研究报告数十篇,其中多篇得到了国家领导人的重要批示,多篇得到中办、国办采用;参与研究开发的客户风险预警系统在国家开发银行与银监会的发挥重要作用;参与开发的经济监测、预测、预警及政策仿真系统在国家发展和改革委员会、商务部和国家外汇管理局,支持了政府高层的科学决策,也对相关领域的研究与发展,产生了积极的作用。
三、时间:2020年11月6日(周五),上午10:00-11:30
四、地点:腾讯会议【508 739 216】
五、主持人: 彭俞超副教授,suncitygroup太阳新城学术交流部主任
点评人: 吴锴助理教授
六、内容简介
Modelling and forecasting economic time series with model instability and model uncertainty is a long-standing problem. Little attention has been paid to models with time-varying combination weights in large dataset, which may be more realistic in economics. This paper proposes a new time-varying model averaging method via an adaptive LASSO to determine optimal time-varying combination weights to candidate models, thus avoiding over-fitting and yielding sparseness from a set of various potential predictive variables, simultaneously. For any fixed time point t, the asymptotic optimality and the asymptotic convergence rate of the selected weights are derived. Furthermore, the asymptotic consistency and normality of the proposed time-varying model averaging estimator are obtained. Simulation studies and empirical applications to inflation rate forecasting highlight the merits of the proposed method relative to other competing methods in the literature.