摘要
本文立足于我国金融市场的现状提出了基于LASSO方法和神经网络模型的量化交易智能系统。该系统首先使用LASSO方法从众多技术指标中选出极少数最有效的指标作为输入变量,然后通过神经网络方法来搜索最优的交易规则,并以沪深300股指期货为例进行回测检验。结果显示:第一,与AIC和BIC回归模型相比,LASSO选出的变量少、预测高、且稳健性强;第二,经过神经网络的优化,交易系统的收益率和风险控制能力都得到了显著提高;第三,即使在考虑交易成本的前提下,该系统也可以获取超额收益。
This paper proposes an intelligent system for quantitative trading based on LASSO method and artificial neural net- work model. The proposed system uses LASSO to select a few most effective variables from a large number of technical indi- cators. Then, the selected indicators will be treated as input variables of neural network model to search for the optimal trading rules, and CSI 300 Index will be used as an example for back-testing. The results show that： First, compared with AIC and BIC criteria of OLS regression models, LASSO method selected the least number of variables, while it shows comparable pre- diction accuracy and the strongest robustness; Second, since the trading system is optimized by neural network model, its profit- ability and risk control capabilities have been significantly improved; Finally, even considering the transaction costs, it can al- ways outperform the ＂buy ＆ hold＂ strategy and obtain excess returns.
出处
《投资研究》
北大核心
2014年第9期23-39,共17页
Review of Investment Studies
基金
国家自然科学基金资助项目(71101083
71271128
71331006)
上海市教育委员会科研创新项目(12ZZ072)
上海财经大学创新团队支持计划
关键词
LASSO
变量选择
神经网络
量化交易
LASSO
Variable Selection
Neural Network
Quantitative Trading
