Customer profile

  • This project was delivered for an online trading platform (under NDA).

Business Goals

  • Develop consistently profitable, self-learning artificial trading agent acting at under a milliseconds timescale.
  • Mask the algorithm activity from bot detection software.

Challenge

  • Automate sub-second trading.
  • Maximize the signal-to-noise ratio.
  • Design and implement exchange simulation software. Take into account exchange fees, message delivery, and processing delays.
  • Automatically adapt trading strategy to a highly stochastic and non-stationary financial market environment.
  • Handle occasional trend changes due to global events like epidemics, earthquakes, wars.
  • Manage surplus of demand created by the agent itself.

Solution

  • Exchange simulator, emulating market and limit orders as well as global trend changes, and taking into account message delays and market fees was developed.
  • In order to regularise the dynamics of the profit and loss (P&L) function, the training set was constructed by sampling periods of high price activity.
  • In order to limit the state space and drop low-correlated features, dimensionality reduction techniques have been applied.
  • A number of state-of-the-art Deep Reinforcement algorithms (DQN, PPO, DDPG, QFTN, etc.) were trained and evaluated on LOB data with a return maximization objective. The best candidates were chosen and deployed as active self-learning trading agents.

Impact

  • The cumulative mean net profit over a month of backtesting ranged between 18% and 61%.
Roadmap
8 months
Exploratory Data Analysis
AI Solutions Architect, Data Scientist
Trading Strategies Design
AI Solutions Architect
Artificial Trader Design
AI Solutions Architect
Exchange Simulator Design
AI Solutions Architect
Feature Engineering
Data Scientist
Targets Engineering
Data Scientist
Metrics Engineering
Data Scientist
Baseline Supervised Model Development
Data Scientist
Model Performance Evaluation
Data Scientist
DQN RL Model Development
Data Scientist
PPO RL Model Development
Data Scientist
DDPG RL Model DEvelopment
Data Scientist
QFTN RL Model Development
Data Scientist
Standalone AI Service Design
AI Solutions Architect, MLOps
Standalone AI Service Coding
MLOps
Exchange Integration
MLOps
Release

Sign up to receive the project description

    Roadmap
    8 months
    Exploratory Data Analysis
    AI Solutions Architect, Data Scientist
    Trading Strategies Design
    AI Solutions Architect
    Artificial Trader Design
    AI Solutions Architect
    Exchange Simulator Design
    AI Solutions Architect
    Feature Engineering
    Data Scientist
    Targets Engineering
    Data Scientist
    Metrics Engineering
    Data Scientist
    Baseline Supervised Model Development
    Data Scientist
    Model Performance Evaluation
    Data Scientist
    DQN RL Model Development
    Data Scientist
    PPO RL Model Development
    Data Scientist
    DDPG RL Model DEvelopment
    Data Scientist
    QFTN RL Model Development
    Data Scientist
    Standalone AI Service Design
    AI Solutions Architect, MLOps
    Standalone AI Service Coding
    MLOps
    Exchange Integration
    MLOps
    Release

    Industry

    Keywords

    • High-Frequency Trading
    • Deep Reinforcement Learning
    • Machine Learning Deployment