北航经管学院“经济统计论坛”系列讲座
（2026年第6期，总第43期）
讲座题目：Scalable inference for GARCH models
讲座嘉宾：Matias Quiroz副教授
讲座时间：2026年9月10日（周四），10:00-10:40
讲座地点：新主楼 A836
讲座嘉宾
Matias Quiroz is a Senior Lecturer in the School of Mathematical and Physical Sciences at the University of Technology Sydney. His research lies at the intersection of Bayesian statistics, computational statistics and machine learning, with a particular interest in scalable statistical inference. His research has been published in venues including the Journal of the American Statistical Association, Journal of Computational and Graphical Statistics, Journal of Machine Learning Research, ICML and AISTATS. He serves as an Associate Editor of Computational Statistics & Data Analysis and Econometrics and Statistics.
讲座概要
GARCH models are widely used for modelling time-varying volatility, but likelihood-based inference can become computationally expensive for large datasets because the conditional variance is defined recursively. This recursive structure also makes standard data subsampling ineffective. I will present a stabilised weighted subsampling approach that favours earlier observations, reducing the average depth of the recursion while controlling the resulting estimator variability. The method provides unbiased estimators of the log-likelihood and its gradient and can be embedded in different inference algorithms. Applications to high-frequency financial returns demonstrate substantial computational speed-ups for both variational inference and Markov chain Monte Carlo while maintaining accurate inference.
