Business Problem
Investment analysis often relies on disconnected spreadsheets, manual model comparisons, and portfolio decisions that are difficult to evaluate consistently over time.
Case Study
Stock Forecasting and Investment Decision Support System
A web-based investment decision-support platform combining machine-learning forecasting with portfolio optimisation.
This project supports investment analysis and decision-making. It does not guarantee investment returns or financial outcomes.
Forecast Model Comparison
Walk-forward backtestElasticNet
RMSE 0.042
LightGBM
RMSE 0.038
LSTM
RMSE 0.045
Portfolio Allocation (MPT)
Visual placeholder representing model comparison and portfolio allocation views
Project Objective
Build a decision-support platform that compares forecasting models, evaluates performance rigorously, and supports portfolio allocation using established financial theory.
Data and Tools Used
- Historical stock price and return data
- Python analytics and machine-learning libraries
- TensorFlow and LightGBM for model development
- Modern Portfolio Theory for allocation logic
- React for the web interface
Approach
- Built and compared ElasticNet, LightGBM, and LSTM forecasting models
- Created an ensemble forecasting framework
- Used Modern Portfolio Theory for portfolio optimisation
- Applied Ledoit-Wolf covariance estimation
- Evaluated models using RMSE, MAE, MASE, and directional accuracy
- Used walk-forward backtesting to reduce look-ahead bias
Solution
- A structured platform for comparing model performance across multiple metrics
- An ensemble forecasting approach combining selected models
- Portfolio optimisation workflow based on MPT principles
- Visual outputs to support analysis and review
Key Insights
- Different models performed better under different evaluation criteria
- Walk-forward backtesting helped reduce overly optimistic model assessment
- Combining forecasting with portfolio theory created a more complete decision-support workflow
- Clear model comparison made it easier to review trade-offs between approaches
Technology Stack
PythonPandasNumPyScikit-LearnTensorFlowLightGBMReact
Ready to Make Better Use of Your Business Data?
Tell me about your current reporting challenges, and I will help you identify a practical next step.
No technical knowledge is required. Explain the problem in simple words, and we can identify the right solution together.