Stock Market Forecasting Panel App

This repository contains a Panel-based web application for interactive stock market forecasting, developed as part of a master's thesis. The application enables users to compare multiple forecasting models—including ARIMA, XGBoost, LSTM, RNN, and Transformer-based architectures (Crossformers, PatchTST)—across various asset classes and stock markets.

🔍 Research Motivation

Stock market forecasting is inherently complex due to its nonlinear and volatile nature. Traditional statistical models like ARIMA and machine learning techniques such as XGBoost have shown effectiveness, but Transformer-based models have recently gained traction due to their superior ability to capture long-term dependencies and complex patterns.

This study provides a comparative analysis of these models on stock, commodity, and forex data from India, Germany, and the USA, evaluating their predictive performance based on:

• SMAPE (Symmetric Mean Absolute Percentage Error)
• RMSE (Root Mean Square Error)
• Directional Accuracy (Trend Prediction Effectiveness)

🌟 Features of the Panel App

✔ Interactive Model Selection – Compare ARIMA, XGBoost, LSTM, RNN, and Transformer-based models.
✔ Multi-Market Forecasting – Evaluate stock predictions across different financial markets.
✔ User-Friendly Visualization – View time-series forecasts, error metrics, and trend accuracy.
✔ Historical Stock Data Processing – Analyze market trends using real-world financial datasets.
✔ Flexible Forecasting Horizons – Short-term vs. long-term prediction comparisons.

📂 Project Structure

panel_app/
│── app.py                    # Main entry point for the Panel app
│── logging_config.py          # Configures logging for debugging
│── train_model.py             # Script for training stock forecasting models
│── components/                # Modular app components
│   ├── ForecastTable.py       # Displays forecast results
│   ├── MachineLearningFramework.py  # Implements different forecasting models
│   ├── metrics.py             # Defines evaluation metrics (SMAPE, RMSE, etc.)
│   ├── prediction_duration.py # Handles time horizon settings
│   ├── Selectors.py           # UI selectors for model configuration
│   ├── Sidebar.py             # Sidebar UI component

🛠 Installation

Ensure you have Python 3.8+ installed. Then, install the required dependencies:

pip install -r requirements.txt

🚀 Running the Application

To start the Panel-based forecasting tool, run:

python panel_app/app.py

📊 Model Training

To train the models before running predictions:

python panel_app/train_model.py

📜 Research Contributions

• Comprehensive Benchmarking: Provides a structured comparison of forecasting models across diverse market conditions.
• Transformer Model Evaluation: Highlights the strengths and weaknesses of Crossformers and PatchTST in financial time-series forecasting.
• User-Centric Financial Tool: Enables interactive model selection and visualization, making stock prediction more accessible.

🤝 Contributing

Contributions, suggestions, and feedback are welcome! Feel free to raise issues or submit pull requests.

📄 License

This project is licensed under the MIT License.