Session 5: An Empirical Study on One-Step Ahead Stock Price Prediction Using Deep Learning Model and High Dimensional Method
Presentation Type
Oral
Student
Yes
Track
Finance/Insurance Application
Abstract
With the rapid development of machine learning and deep learning techniques, the integration between these techniques and quantitative investment is gaining prominence in both academia and industry. Traditional econometric models often face challenges in predicting financial time series due to various characteristics such as low signal-to-noise ratio, nonlinearity, and non-smoothness. In this study, we employ a novel approach by combining several machine learning and deep learning time series models, and expand our analysis to include more than one hundred factors. These factors encompass a wide range of variables, including technical indicators, macroeconomic indicators, capital flows, and more. Leveraging data acquisition and feature extraction techniques rooted in finance, our study focuses on forecasting the CSI 300 index across various time intervals. Our findings reveal the superior predictive performance of deep learning models when compared to traditional econometric models. Notably, long and short-term memory networks consistently outperform recurrent neural networks. While both models exhibit less sensitivity to multiple factors, they excel when applied to high-frequency data. Furthermore, our research extends beyond stock price prediction, as it also encompasses the forecasting of stock returns. We also introduce fractional differentiation as a key analytical tool to enhance our analysis.
Start Date
2-6-2024 11:00 AM
End Date
2-6-2024 12:00 PM
Session 5: An Empirical Study on One-Step Ahead Stock Price Prediction Using Deep Learning Model and High Dimensional Method
Dakota Room 250 A/C
With the rapid development of machine learning and deep learning techniques, the integration between these techniques and quantitative investment is gaining prominence in both academia and industry. Traditional econometric models often face challenges in predicting financial time series due to various characteristics such as low signal-to-noise ratio, nonlinearity, and non-smoothness. In this study, we employ a novel approach by combining several machine learning and deep learning time series models, and expand our analysis to include more than one hundred factors. These factors encompass a wide range of variables, including technical indicators, macroeconomic indicators, capital flows, and more. Leveraging data acquisition and feature extraction techniques rooted in finance, our study focuses on forecasting the CSI 300 index across various time intervals. Our findings reveal the superior predictive performance of deep learning models when compared to traditional econometric models. Notably, long and short-term memory networks consistently outperform recurrent neural networks. While both models exhibit less sensitivity to multiple factors, they excel when applied to high-frequency data. Furthermore, our research extends beyond stock price prediction, as it also encompasses the forecasting of stock returns. We also introduce fractional differentiation as a key analytical tool to enhance our analysis.