Reinforcement learning–based bond portfolio optimization project leveraging Deep Q-Networks (DQN) for trading strategy development and interactive Power BI dashboards for portfolio analytics and agent behavior visualization.
The project simulates Buy–Sell–Hold decisions within a bond portfolio environment and evaluates portfolio evolution using cumulative rewards and trading behavior analysis.
This project focuses on optimizing bond portfolio decisions using Reinforcement Learning.
A Deep Q-Network (DQN) agent was trained to interact with a portfolio environment and learn optimal trading actions:
- Buy
- Sell
- Hold
The objective was to maximize cumulative rewards while learning effective portfolio management strategies dynamically from environment feedback.
The Power BI dashboard enables exploration of:
- Portfolio performance evolution
- Reward progression
- Trading decisions
- Agent behavior distribution
- Reinforcement learning outcomes
Bond market observations and portfolio state information were prepared for RL training.
Inputs included:
- Portfolio state representation
- Trading actions
- Reward signals
- Environment observations
Preprocessing steps:
- State generation
- Reward formulation
- Episode creation
- Environment simulation
A Deep Q-Network (DQN) agent was implemented for portfolio optimization.
Pipeline:
Bond Market Environment
↓
State Observation
↓
DQN Agent
↓
Buy / Sell / Hold Action
↓
Reward Calculation
↓
Portfolio Update
The agent continuously improved its policy using reward feedback from the environment.
Interactive dashboards were developed to visualize reinforcement learning outcomes.
Dashboard Components:
✅ Portfolio Performance Trend
✅ Reward Evolution During Training
✅ Agent Decision Distribution
✅ Trading Actions Across Episodes
✅ KPI Monitoring:
- Final Cumulative Reward
- Total Reward
- Average Reward
- Total Decisions
- Python
- Deep Q-Network (DQN)
- Reinforcement Learning
- NumPy
- Pandas
- Power BI
- DAX
- Portfolio Analytics
- Jupyter Notebook
- Git
- GitHub
Bond-Portfolio-Optimization-RL/
│
├── dashboard/
│ └── Bond_RL_Dashboard.pbix
│
├── data/
│ └── bond_rl_powerbi.csv
│
├── notebooks/
│ └── bond_portfolio_rl.ipynb
│
├── screenshots/
│ └── Bond Market Portfolio Optimization Dashboard Overview.png
│
├── README.md
└── .gitignore
The DQN agent demonstrated learning behavior through sequential interactions and achieved positive cumulative reward progression.
Power BI dashboards enabled interactive analysis of:
- Portfolio evolution
- Reward behavior
- Trading actions
- Decision distributions
- Reinforcement learning performance
- Reinforcement Learning portfolio optimization
- Deep Q-Network implementation
- Financial analytics dashboard
- Trading strategy analysis
- Power BI integration
- Interactive RL visualization
- PPO / A2C implementation
- Multi-asset portfolio optimization
- Risk-adjusted reward formulation
- Real-time market integration
- Deployment as analytics platform
Tej More
MS Data Science — Rochester Institute of Technology
AI • Reinforcement Learning • Finance • Healthcare AI