Consumer Spending in the U.S. Around Russia's Invasion of Ukraine

DPSS Capstone — University of Chicago Harris School, Data & Policy Summer Scholar program. July 2022 · Revised July 2026

Daily card-transaction data from the SafeGraph Spend panel (seven retail categories, February 1 – March 31, 2022) are used to ask whether Russia's February 24, 2022 invasion of Ukraine — and the sanctions that followed — visibly changed U.S. consumer spending.

Author: Pablo Zavala

📄 Report: PDF · HTML (self-contained; all code embedded behind fold-out buttons)

Key findings

1. No break at the invasion. Aggregate panel spending rose a modest +6.1% after February 24, led by categories with mundane seasonal explanations (clothing +25.1%, traveler accommodation +17.6%). Grocery — the stockpiling candidate — was flat (−0.6%).
2. The apparent "reaction" on February 21 is a calendar artifact. Presidents' Day is a bank holiday: card processing paused (−4.9 s.d. residual on Feb 21) and the weekend backlog cleared on Tuesday (+4.9 s.d. on Feb 22). What looks like a geopolitical consumption response is a payment-processing displacement — a useful caution for event studies on high-frequency payments data.
3. The war's clearest imprint is prices at the pump. Gasoline-station spending per transaction climbed steadily (+10.6%) as fuel prices hit record nominal levels — yet county-level shifts in general-merchandise spending are uncorrelated with gasoline-spending shifts (β ≈ 0 with state fixed effects and HC1 errors, winsorized robustness), i.e., no evidence of early crowd-out.

Analysis

Section	What it does
National trends	Daily panel spending by category with the invasion and Presidents' Day marked; documents the Monday processing sawtooth
Deseasonalization	Day-of-week fixed effects estimated within category; standardized residuals isolate genuine surprises (and expose the holiday whipsaw)
Per-transaction spending	Category series indexed to pre-invasion mean = 100; isolates the intensive margin where the fuel-price shock shows up
NYC geography	3,198 establishments spatially joined to the five boroughs; deciles of spending change computed within NYC
Substitution test	County-level Δgeneral-merchandise on Δgas-station spending: OLS, + state FE, + winsorization; HC1 robust SEs (modelsummary table)

Repository structure

capstone.Rmd                  Source of record — R Markdown report
capstone.R                    Generated from capstone.Rmd via knitr::purl(); do not edit
data/
  spending_timeseries_total.csv   Daily total spending by category
  spending_timeseries_per.csv     Daily per-transaction spending by category
  spending_shift_in_baskets.csv   County-level spending shifts (general merch vs. gas)
  spending_maps.csv               Establishment-level spending changes with coordinates
  get_shapefile.sh                Downloads the county shapefile (not committed)
output/
  capstone-report.pdf             Rendered report (PDF)
  capstone-report.html            Rendered report (self-contained HTML)
  figs/                           Figure PNGs exported at render time

Reproducing

1. Install R (≥ 4.3) and packages:

   install.packages(c("tidyverse", "sf", "scales", "kableExtra",
                      "modelsummary", "sandwich", "ragg", "rmarkdown"))

   or create a conda environment with everything prebuilt:

   conda create -p ./env -c conda-forge r-base r-tidyverse r-sf r-rmarkdown \
     r-modelsummary r-kableextra r-sandwich r-ragg pandoc tectonic

2. Download the 2016 TIGER/Line county shapefile (~124 MB unzipped, not committed):

   bash data/get_shapefile.sh

3. Render from the repository root:

   rmarkdown::render("capstone.Rmd", output_file = "output/capstone-report.html")                    # HTML
   rmarkdown::render("capstone.Rmd", output_format = "pdf_document",
                     output_file = "capstone-report.pdf")                                            # PDF (Tectonic)

Data notes

• data/*.csv are course-provided extracts derived from SafeGraph Spend data, included for coursework reproducibility; review SafeGraph's terms before redistributing.
• The establishment- and county-level change measures (delta_STD) are provided by the course extract; their exact construction is not fully documented, so the report interprets them ordinally (deciles, signs) rather than as precise magnitudes.
• County boundaries: U.S. Census Bureau TIGER/Line 2016.

Acknowledgments

The spending-data extracts were prepared and provided by the Data & Policy Summer Scholar (DPSS) program at the University of Chicago Harris School of Public Policy. All analysis, interpretation, and any errors are the author's own.

License

Code in this repository (the R Markdown source, the generated script, and shell scripts) is released under the MIT License. The files in data/ are course-provided extracts derived from SafeGraph Spend and are not covered by that license; they remain subject to SafeGraph's terms and are included solely to reproduce this coursework.