摘要
分别以筛选的4种技术指标和6个宏观经济指标作为国债期货指数预测变量,利用随机森林算法构建4种机器学习预测模型;依据价格波动集聚性设计跟踪交易规则,通过比较4种模型的预测精度和跟踪交易收益率,检验宏观经济指标、技术指标和随机森林算法对国债期货指数的预测能力。研究结果发现:用主成分精选技术指标构建的预测模型,对国债期货指数的跟踪交易收益率虽然明显优于市场收益率,但不如遵循单个技术指标经验交易规则的跟踪交易收益率;用主成分精选技术指标和宏观经济指标构建的模型能够取得很好的预测精度和跟踪交易收益率,这表明宏观经济指标与技术指标都对国债期货价格具有预测意义,可以利用随机森林机器学习算法构建有效的国债期货量化投资模型。
Four technical indicators and six macroeconomic indicators are selected as the predictor variables of the Treasury bond futures price.Four Treasury bond futures price prediction machine learning models are constructed by using the random forest algorithm.Designing the track trading rules in accordance with price fluctuation aggregation,the predict ability of macroeconomic indicators,technical indicators and random forest algorithm has been tested by comparing prediction accuracy and track trading return of the four RF models.The empirical results show that,the track trading return of the RF model constructed with the principal component selection technical indicators is much better than market return,but obviously worse than the track trading return of a single technical indicators empirical trading rules.The RF model constructed with the principal component selection technical indicators and macroeconomic indicators can achieve good prediction accuracy and track trading return.It means that both macroeconomic indicators and technical indicators have ability to predict the Treasury bond futures price,and random forest machine learning algorithm can be used to construct Treasury bond futures quantitative investment model effectively.
作者
陈标金
王锋
CHEN Biao-jin;WANG Feng(College of Economics and Management,South China Agricultural University,Guangzhou 510642,China;Jicheng Futures Markets Research Institute,South China Agricultural University,Guangzhou 510642,China)
出处
《统计与信息论坛》
CSSCI
北大核心
2019年第6期29-35,共7页
Journal of Statistics and Information
基金
国家社会科学基金一般项目《基于投机视角的农产品期货定价机制研究》(13BJL065)
广东省教育改革项目《金融科技嵌入“金融工程”课程教育的教学方法研究》(2018JR009)
关键词
国债期货
价格预测
随机森林
机器学习
treasury bond futures
price prediction
random forest
machine learning
