Back to Projects
FinTech
Time Series
ARIMA
LSTM

Time Series Portfolio Optimization

Forecast-driven portfolio allocation with Modern Portfolio Theory

Project Overview

A machine learning and data science solution for Guide Me in Finance (GMF) Investments. This project applies ARIMA and LSTM time series models to predict asset price movements, then optimizes asset allocation with Modern Portfolio Theory (MPT). The strategy is validated through rigorous backtesting against historical benchmarks.

Business Context

GMF Investments required a data-driven approach to improve portfolio returns while controlling downside risk. By integrating statistical and deep learning models into the investment pipeline, the solution enables more accurate forecasts, risk-adjusted allocation, and client-ready portfolio insights.

Key Features

Hybrid Forecasting: Combines ARIMA and LSTM models for robust market trend predictions

Automated Workflow: Modular architecture supports extension, automation, and reproducibility

Risk-Aware Optimization: Uses MPT and Efficient Frontier analysis for return vs. volatility trade-offs

Backtesting & Metrics: Evaluates portfolio strategies with MAE, RMSE, Sharpe Ratio, and cumulative returns

Technical Implementation

Historical financial data sourcing and preprocessing with yfinance and custom pipelines

ARIMA/SARIMA parameter tuning using auto_arima; LSTM deep learning model optimization (layers, epochs, batch size)

Feature engineering for returns, volatility, and Bayesian change point detection

Portfolio optimization with PyPortfolioOpt and validation via historical backtests

Technologies Used

Python 3.8+
TensorFlow
Keras
pmdarima
PyPortfolioOpt
Pandas
NumPy
Matplotlib
Streamlit
yfinance
Docker
Jupyter
GitHub Actions

Results & Impact

Delivered a reproducible ML-driven portfolio optimization framework that improved Sharpe Ratio and cumulative returns versus baseline benchmarks, providing GMF with a scalable, client-ready investment tool.

Interested in Learning More?

Explore the complete implementation, documentation, and code on GitHub.