With the development of Internet and computer technology, information resources grow explosively, the stock price is increasingly unstable and frequently fluctuating. In this case, in order to make timely response to new information and make correct trading decisions to obtain capital gains, it is unrealistic to only rely on human to collect all information and make short-term trading decisions. In addition, the irrational emotions of investors can not be utterly avoided. Quantitative trading uses mathematical model and computer technology to screen the events that can bring excess returns from historical data to formulate investment strategies without subjective judgment. This method greatly reduces the fluctuation of investors’ sentiment and avoids them from making irrationalinvestment decisions. This investment method is widely used at home and abroad.
The rapid development of computer technology has enabled thousands of stock tradings to happen within a fraction of a second. In this fast-moving market, it is crucial to automate trading with the help of machine learning and deep learning algorithms to free operators from repetitive work and to maximize profit. Buying low and selling high sounds easy but can be difficult to achieve in a highly volatile market. This paper proposes a method that ensembles ResNet, Multi-layer Perceptron and XGBoost to make profitable short-term trading decisions in real-time.
According to Selvin et al (2017) , three deep learningbased methods, including Long Short Term Memory, Recurrent Neural Network, and Sliding Window Convolutional Neural Networks were compared. The models were only trained on the stock price data of one company Infosys but were tested on three companies. During the testing phase, records of 90 minutes of the stock price were provided and the models were to predict the stock price for the next 10 minutes.
Machine learning models need less training data but are more sensitive to feature selection and noises in the dataset. Deep learning algorithms need larger datasets to train but can capture more complex patterns than machine learning algorithms.
This paper proposes a method that ensembles ResNet, Multi-layer Perceptron and XGBoost to make profitable short-term trading decisions in real-time.
- little profit
- avoid trading mistakes
Algorithms: ResNet, Multi-layer Perceptron and XGBoost
- Operating system : Windows.
- Coding Language : Python.
System : Pentium IV 2.4 GHz or intel
Hard Disk : 40 GB.
Floppy Drive : 1.44 Mb.
Mouse : Optical Mouse.
Ram : 512 Mb.
In this paper, we proposed a model that ensembles three deep learning and machine learning models to make short-term stock trading decisions based on features about the current situation. Our method is able to respond in real-time to the changes of the stock market with trading decisions that generate at least 25% more profit than other models. Besides, the hyper-parameters of our model have been generated automatically with the grid search method and our model doesn’t rely on any feature selection, making it robust and easy to deploy than models with a more complex pipeline.