摘要
本文在无指导学习的研究框架下,运用分位数回归模型结合变点检验,对中国证券市场的异常交易行为进行甄别研究。通过分析持股比例变动与股价收益率间协同演化关系的异常,为甄别异常交易行为设立判别标准并客观的界定阈值提供了一种新的方法。基于这一方法监管者可以构建分期、分级、分类的实时监管体系,提高监管效率。
Based on unsupervised learning theory we use quantile regression model and change point test in research of detecting outlier transaction behavior. We propose a new method which can specify discriminating standard and threshold in order to detect outlier transaction behavior by analyzing outlier in coherently evolutionary relationship between changes of shareholders＇ owner-ratio and stock price return. In use of this method regulators can build a real-time regulation system according to period, level and class for enhancing regulation efficiency.
出处
《数理统计与管理》
CSSCI
北大核心
2009年第4期671-677,共7页
Journal of Applied Statistics and Management
关键词
分位数回归
异常甄别
无指导学习
价格操纵
变点
quantile regression, outlier detection, unsupervised learning, price manipulation, change point
