us-consumption-forecasting

Forecasting U.S. Consumption Using Real Disposable Personal Income

Final project for ECON 409 (Public Policy Analysis)

Abstract

This project estimates the marginal propensity to consume (MPC) in the U.S. economy by modeling Real Personal Consumption Expenditures as a function of Real Disposable Personal Income, using FRED data accessed via API. Benchmark, ARIMA, ETS, and linear regression models were compared on in-sample and out-of-sample forecast accuracy, with the regression specification estimating an MPC of 0.75 (R² = 0.87) and providing the most theoretically grounded medium- and long-horizon forecasts. While a random walk with drift produced stronger short-term predictions, the regression model was ultimately selected for its ability to recover an economically meaningful income-consumption relationship rather than purely mechanical forecast accuracy.

View the report

📄 Read the full report

Tools used

• R / RStudio
• fredr (FRED API access)
• fable, fabletools, forecast (ARIMA, ETS, benchmark time-series models)
• Linear regression
• R Markdown (knitted to HTML)

Data

Data is pulled live from the FRED API (series PCEC96 and DSPIC96) rather than committed to the repo. To reproduce, set an environment variable FRED_API_KEY with your own FRED API key (free to request) before knitting the .Rmd file.

Repo contents

• us-consumption-forecasting.html — knitted final report (viewable in-browser)
• us-consumption-forecasting.Rmd — R Markdown source code (add this if you’re including your source)