Forecasting U.S. Unemployment Claims using Prophet

Overview

This project pulls weekly U.S. initial unemployment insurance claims from the U.S. Department of Labor Open Data API, performs basic exploratory analysis, and builds a baseline 8-week forecast using Prophet.

Why this matters:

Initial jobless claims are a key barometer of U.S. labor market strength. Forecasting them provides early insight into economic turning points that guide policy, investment, and business decisions.

Data

Source: Employment and Training Administration (ETA), Unemployment Insurance National Weekly Claims.
Range: 10 years of historical weekly data (2015–2025).

Key fields used:

• rptdate: week ending date
• c5: initial claims, seasonally adjusted
• c8: 4 week moving average of seasonally adjusted initial claims

Methods

• Processing: Pulled via custom Python script; filtered, cleaned, and checked for missingness
• Exploratory Analysis:
  • Summary statistics and visualizations of claims trends.
  • Rolling averages and volatility features (e.g., 4-week standard deviation).
• Modeling Approach:
  • Prophet model applied on log-transformed claims for stability.
  • Captures long-term trend and yearly seasonality.
  • Forecast horizon: 8 weeks ahead.
• Visualization:
  • Historical vs. forecasted claims with confidence intervals.
  • Decomposition plots of trend and seasonality components.

Results

Chart Elements:

• Black dots: Actual observed data (historical jobless claims in log scale)
• Blue line: Prophet's fitted trend (model's prediction/estimate)
• Light blue shaded band: Uncertainty interval (typically 80% confidence interval by default)

Forecasted values: continuation of the blue line to the right of the last black dot (see below)

• Initial claims are projected near ~225K. The bands (≈±25%) reflect inherent weekly volatility and Prophet’s 80% confidence interval.
• Uncertainty widens with horizon as errors compound and is amplified by post-COVID exposure to extreme labor market shocks.

Repository structure

.
├── notebooks/
│ └── forecasting-us-jobless-claims.ipynb # analysis
├── src/
│ └── jobless_claims_data_pull.py # data pull script
├── data/
│ └── ui_national_weekly_claims_metadata.csv # meta data
│ └── ui_national_weekly_claims_10y.csv # claims data
├── assets/
│ └── [images].png
├── README.md
└── LICENSE

Limitations and Next Steps

Future extensions:

• Evaluation metrics to be added