Stock Price Prediction Project

Author: Grafton Cook

Contact: grafton.cook@gmail.com

Table of Contents

• Stock Price Prediction Project
  • Table of Contents
    • Project Overview
      • Time Series Components
      • Traditional Analysis
      • Deep Learning
      • Hybrid Model
  • Directory Structure
  • Setup & Installation
    • Clone this repository
    • Install dependencies
    • Jupyter Notebooks
  • Usage & Examples
  • Results & Analysis
  • Next Steps
  • License & Acknowledgments

Project Overview

This repository contains a multi-part time series forecasting project focused on predicting stock prices for a financial institution. You will see both classical and deep learning modeling approaches, as well as a final hybrid model that combines the best of both worlds. Specifically, the project is divided into four sub-projects:

Time Series Components

• Decompose and examine the time series (trend, seasonality, stationarity).
• Perform ACF, PACF, and ADF tests.

Traditional Analysis

• Apply classical forecasting models (Moving Average, Exponential Smoothing, AR, ARIMA).
• Compare results using RMSE and data visualizations.

Deep Learning

• Prepare data for deep learning.
• Implement RNN/LSTM models for time series forecasting.
• Compare performance against classical methods.

Hybrid Model

• Combine classical and deep learning approaches.
• Evaluate performance on real-world stock data.

Directory Structure

stock-price-prediction/
├── 01-time-series-components/
│   ├── data/
│   ├── notebooks/
│   │   ├── EDA.ipynb
│   │   └── stationarity_tests.ipynb
│   └── scripts/
├── 02-traditional-analysis/
│   ├── data/
│   ├── notebooks/
│   │   ├── traditional_analysis.ipynb
│   └── scripts/
├── 03-deep-learning/ **(TBD)**
│   ├── data/
│   ├── notebooks/
│   └── scripts/
├── 04-hybrid-model/ **(TBD)**
│   ├── data/
│   ├── notebooks/
│   └── scripts/
├── requirements.txt
└── README.md

part-x/: Each sub-project directory with its own notebooks, data, and scripts. environment.yml or requirements.txt: Conda or pip environment details.

Setup & Installation

Clone this repository

git clone https://github.com/tacotuesday/time-series-stock-forecasting.git
cd time-series-stock-forecasting

Install dependencies

Using Conda:

conda create --name stock-forecasting-env --file requirements.txt
conda activate stock-forecasting-env

Or using Pip:

pip install -r requirements.txt

Jupyter Notebooks

Ensure Jupyter is installed and launch notebooks:

jupyter notebook

Navigate to the relevant sub-project under part-x/notebooks.

Usage & Examples

• Data: Sample stock price data is located in each data/ folder (or instructions to download from a public source).
• Model Training: Run the notebooks in chronological order to see how the time series is analyzed and modeled.
• Hyperparameter Tuning: Some notebooks contain sections for adjusting hyperparameters (e.g., ARIMA p/d/q, LSTM architecture).

Results & Analysis

• Comparisons: We compare RMSE, MSE, and/or MAE across different models.
• Visualizations: Time series plots, predicted vs. actual, residual analysis.
• Insights: The best-performing model (in this dataset) is typically the hybrid approach, though performance may vary depending on the data.

Next Steps

• Enhance the hybrid model by experimenting with other neural network architectures (e.g., Transformers).
• Add real-time inference or streaming pipelines for updated price data.
• Extend the code to additional financial instruments (crypto, bonds, etc.).

License & Acknowledgments

License: MIT License.

Acknowledgments: This project is part of a Manning LiveProject. Datasets from Alpha Vantage.