Enhancing High Frequency Technical Indicators Forecasting Using Shrinking Deep Neural Networks

Abstract

Recent years have witnessed the successful combination of finance innovations and AI techniques in various finance applications including quantitative trading. Despite great research efforts devoted to leveraging deep learning methods for building better quantitative strategies, existing studies still face serious challenges, such as how to establish effective high frequency predictor variables, how to solve in-sample overfitting in a high-dimensional setting and how to balance the risk and return. In this paper, we propose a hybrid deep learning based high frequency technical indicators investment strategy approach enhanced by elastic net model, which called SDNN, to address the above challenges. Our main contributions are summarized as follows: i) We establish several high frequency technical indicators and investigate the statistically and trading significant in-sample and out-of-sample predictive power for each indicators. ii) we suggest a elastic net model to shrinking the dimensional of predictive factors in order to improve the out-of-sample performance in high-dimensional setting. iii) we integrate deep learning method with a Sharpe-optimised framework to achieve a risk-return balanced investment strategy. The experiments on Chinese stock market demonstrate the Sharpe-optimised SDNN, improved traditional linear method by more than 75% percent annualized return and outperformed other machine learning methods as well.