Research on Stock Price Prediction and Quantitative Strategies
based on Deep Learning
Reporter:SHUI YUE
上海立信会计金融学院
Shanghai Lixin University of Accounting and Finance
Quantitative Investment Strategy based
on Lightgbm-BiLSTM
03
Abstract
01
A Comparative Study of LSTM, GRU and
BiLSTM on Stock Closing Price Prediction
02
Research Prospect
04
PART I
Abstract
PART I
01
上海立信会计金融学院
Shanghai Lixin University of Accounting and Finance
Brief Introduction
The first part of the experiments in this paper uses the SPDB and IBM data to build stock prediction models with
LSTM, GRU, and BiLSTM, respectively. By comparing the prediction results of these three deep learning models, it
is found that for both datasets the BiLSTM model outperforms the other models and has better prediction accuracy.
The second part uses the complete-market stock data of A-share and uses the Lightgbm model to filter the 50 price-
volume factors and select the 10 factors with the highest importance. After that, the BiLSTM model is used to
combine the factors and establish a quantitative investment strategy. Finally, the strategy is empirically tested and
back-tested, and it is found that the strategy outperforms the market benchmark index, which illustrates the practical
application value of the BiLSTM model in stock price prediction and quantitative investment.
Abstract
PART II
02
上海立信会计金融学院
Shanghai Lixin University of Accounting and Finance
PART II
A comparative study of
LSTM, GRU and BiLSTM on
stock closing price prediction