基于机器学习的权益型开放式基金风格漂移预警研究  

Study of Warning of Style Drift for Equity Open-Ended Funds Based on Machine Learning

作　　者：曹子裕 陈琪琪 彭云宸 Cao Ziyu;Chen Qiqi;Peng Yunchen

机构地区：[1]云南红塔银行总行 [2]云南省农村信用社联合社 [3]中国移动云南分公司

出　　处：《投资研究》2023年第11期91-105,共15页Review of Investment Studies

摘　　要：本文针对权益型开放式基金设计了一个风格漂移的预警系统,希望能一定程度上解决基金风格漂移有危害,而现有监管方式有时滞等问题。本系统分为事后和事前两部分。在事后诊断部分,本文使用SDS值对基金的风格漂移程度进行了衡量;在事前预警部分,我们从基金收益风险特征、基金公司和经理特征以及股市特征三个层面爬取了共15个特征变量,并通过KNN等机器学习模型构建了一个预测准确率约为73%的预警系统。We design a warning system for style drift of equity open-ended funds and hope this system could partly solve problems such as harms bringing by style drift and time-lag of existing supervision method.This system includes two parts:diagnosis and warning.In diagnosis part,we use SDS value to measure the level of style drift.In warning part,we select 15 variables from 3 dimensions:fund return and risk characteristics;company and manager features;market conditions and expect these characteristics could predict the level of style drift(i.e.SDS).Then a warning system with a prediction accuracy of 73%is constructed through machine learning models such as KNN,random forests,SVM,logit regression and Xgboost.

关 键 词：风格漂移 权益型开放式基金 机器学习 预警 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] F832.51[自动化与计算机技术—控制科学与工程]