Multivariate Unemployment Rate Forecasting with Long Short-Term Memory (LSTM)

This repository contains the project code and data for our Topics in Economics - Deep Learning (ECON-420) course at McGill University. As part of our final project, we are to explore a topic sourcing data and applying a deep learning method. The topic of choice is Unemployment Rate predictions using LSTM.

Data Sources

All data was sourced from Federal Reserve Economic Data (FRED). The datasets range from January 1st, 1972 to July 1st, 2023, the data used to train the model can be found within this repository.

Raw Data

The raw FRED data can be found in the links below:

Historic Unemployment Data:

• Unemployment Rate (UNRATE.csv)

Economic Indicators:

• Real Gross Domestic Product (GDP) (GDPC1.csv)
• Consumer Price Index (CPI) (CPIAUCSL.csv)
• Federal Funds Effective Rate (FEDFUNDS.csv)
• Industrial Production - Manufacturing (IPMAN.csv)
• Labor Force Participation (CIVPART.csv)
• Velocity of M2 Money Stock (M2V.csv)
• Personal Current Transfer Reciepts: Government social benefits to person: Social Security (SOCSEC.csv)

Tl;dr:

In short, here are the general findings/visualizations of the project.

Visualization of Datasets:

Correlation Matrix:

Loss of LSTM:

Forecasting:

Model Evaluation - Actual vs Predictions:

Comparison to Vector AutoRegressive (VAR) Model:

Comparison to Moving Average: