Development of Automated Trading Software Powered with Data Science
The Customer is a US startup in the sphere of financial services.
The Customer wanted to develop a trading system that was to recommend traders certain courses of action on US stock exchanges, such as NASDAQ and AMEX. Striving to make the recommendations as precise as possible, the Customer was looking for a professional data science team.
ScienceSoft’s team of 20 data scientists examined publicly available research papers that dwelled on best practices in financial trading. We took the described patterns, such as a wave, an ascending scallop, a pennant, a descending triangle, and turned them into algorithms, as the Customer’s trading system was to make well-considered decisions as the most successful traders would do.
The system was designed in a way that allowed a new pattern to be easily added. This flexibility made it possible to continuously improve the model and keep pace with the quickly changing stock market environment.
Our data scientists tuned the system so that it scanned the stock market data at different time intervals (for example, every minute and every five minutes). This allowed the system to better understand securities’ behavior as some securities showed various patterns on different interval scales.
As the trading system was able to look at the data from different perspectives and, correspondingly, recognize several patterns, our team introduced majority vote classifiers. This enabled prioritizing the identified patterns based on the weights assigned to them.
Initially, the system was designed to be fully automated, which means that it sold and bought securities itself, without any trader’s involvement. Later, our team also added manual control. This is how the system turned into a tool that provided traders with science-based recommendations on whether to buy or to sell securities, as well as enabled efficient investment portfolio management.
To ensure that the system’s predictions and recommendations are accurate and reliable, our team tested the developed algorithms on historical stock exchange transactions and validated them on real-time data.
An update as of 2019:
The active development of the project was in the early 2000s. As of 2019, deep learning would be applied.
The Customer got a proprietary multi-user system and started offering it on the market as a ready-to-use solution for the financial industry. With data science, predictive and prescriptive analytics at its core, the system translated various patterns into precise trading recommendations and allowed users to efficiently manage their investment portfolios.
Technologies and Tools
Prediction and recommendation engine: C++.
User application back end: Java.