From 2015 to 2024 the effective federal funds rate sat below its baseline Taylor-rule prescription in 33 of 40 quarters, by as much as 8.3 pp in Q1 2022, and the gap carries statistically significant predictive content for core PCE inflation one to two quarters ahead.

Paper · Data · Site

Note

Needs R 4.0 or later and a free FRED API key. Sample window 2015 Q1 to 2024 Q4, FRED vintage 2026-09-08 (pinned in R/config.R). The published figures predate that pin; see Data availability.

This is the replication package for the deep dive Discretion Over Rules: Federal Reserve Policy Departures and the Inflation Surge of 2021–2022. R pulls quarterly macroeconomic series from FRED, computes three Taylor-rule specifications and an ECB comparison, and tests for Granger causality and cross-correlation between the deviations and core PCE inflation. It exports Chart.js widgets and static PNG fallbacks.

Quick start

1. Clone: git clone https://github.com/nik110403/taylor-rule.git && cd taylor-rule
2. Get a free FRED API key at https://fredaccount.stlouisfed.org/apikeys, then cp .Renviron.example .Renviron and set FRED_API_KEY= to your key. R loads .Renviron from the project root.
3. Run everything from the repo root: Rscript run_all.R. Missing packages (fredr, dplyr, vars, tseries, lmtest, zoo, lubridate) install on first run through R/setup.R.
4. Check the result: the run ends with a checklist of every expected output file. PNG figures are in output/figures/, widgets in output/widgets/.

Exhibit map

Exhibit	Script	Output
Hook: policy gap and core PCE inflation	R/04_hook_figure.R, R/04_pngs.R	output/figures/fig0_hook.png, output/widgets/fig0_hook.html
Three Taylor-rule variants against the funds rate	R/03_figures.R, R/04_pngs.R	output/figures/fig1_taylor_vs_actual.png, output/widgets/fig1_taylor_vs_actual.html
Baseline deviation by quarter	R/03_figures.R, R/04_pngs.R	output/figures/fig2_deviation.png, output/widgets/fig2_deviation.html
Deviation against core PCE two quarters later	R/03_figures.R, R/04_pngs.R	output/figures/fig3_scatter.png, output/widgets/fig3_scatter.html
ECB deviation against euro area HICP	R/03_figures.R, R/04_pngs.R	output/figures/fig4_ecb.png, output/widgets/fig4_ecb.html
Cross-correlation across lags	R/03_figures.R, R/04_pngs.R	output/figures/fig5_ccf.png, output/widgets/fig5_ccf.html
VAR(4) assumption diagnostics	R/05_assumption_checks.R	output/assumption_report.md

The paper numbers its figures in reading order, so its Figure 1 is the hook and its Figure 2 is fig1_taylor_vs_actual, and so on.

Method

Three US variants, all floored at zero:

Rule	Formula	r*	Notes
Baseline	r* + π + 0.5(π − π*) + 0.5·y_gap	0.5%	Post-GFC Laubach-Williams
Alternative	same	1.0%	Robustness
Inertial	0.75·r_{t−1} + 0.25·baseline	0.5%	Partial adjustment

π is core PCE year-over-year %, π* is 2%, and y_gap = 100·(GDPC1 − GDPPOT)/GDPPOT.

The ECB comparison (fig4) uses r* = 1.0%, euro area HICP for π and no output-gap term, because euro area potential output is not available on FRED. It is prescribed against ECBDFR.

R/02_analysis.R runs ADF stationarity tests on inflation and the deviation series, bivariate Granger causality at lags 1 to 4 in both directions, VAR(4) Granger causality (vars::causality), and a cross-correlation of the baseline deviation against core PCE at lags 0 to 8 quarters. R/05_assumption_checks.R reports normality, serial-correlation and heteroskedasticity diagnostics for the VAR(4).

Data availability

All data come from FRED and are subject to its terms. They are public. They are not covered by the code licence.

Series ID	Description	Transformation
FEDFUNDS	Effective federal funds rate	levels
PCEPILFE	Core PCE deflator	YoY % (pc1)
PCEPI	Headline PCE deflator	YoY % (pc1)
GDPC1	Real GDP	levels
GDPPOT	CBO potential real GDP	levels
UNRATE	Unemployment rate	levels
NROU	Natural rate of unemployment (CBO)	levels
ECBDFR	ECB deposit facility rate	levels
CP0000EZ19M086NEST	Euro area HICP	YoY % (pc1)

Monthly series convert to quarterly by taking the last observation of each quarter. The raw pulls are not committed to this repository (data-raw/ is git-ignored); R/01_data.R writes them locally.

Vintage and revisions:

• FRED revises historical series, including GDPPOT (CBO potential output), with each CBO Budget and Economic Outlook. Results reproduce exactly only against the vintage retrieved for the published piece. Later pulls shift the output-gap series and therefore every Taylor prescription.
• The sample window is locked at 2015 Q1 to 2024 Q4 in R/config.R (START_DATE, END_DATE).
• R/config.R pins FRED_VINTAGE (2026-09-08), passed as realtime_start and realtime_end on every FRED pull, with the raw response archived under data-raw/fred/<FRED_VINTAGE>/. The published figures predate this pin and reflect whatever FRED reported at publication time. Re-running R/01_data.R reproduces the FRED state as of FRED_VINTAGE, which diverges from the published numbers as FRED revises GDPPOT and other series.
• The paper documents the vintage used and the direction of any revisions.

Repository layout

taylor-rule/
├── run_all.R              # runs the pipeline in order, then lists expected outputs
├── R/
│   ├── config.R               # sample window, π*, r* assumptions, FRED key
│   ├── setup.R                # package management
│   ├── 01_data.R              # FRED pull, quarterly conversion -> output/data/df_quarterly.rds
│   ├── 02_analysis.R          # Taylor variants, ECB rule, ADF/Granger/CCF -> df_analysis.rds, df_ccf.rds
│   ├── 03_figures.R           # Chart.js widgets, figures 1 to 5
│   ├── 04_hook_figure.R       # hook widget (fig0)
│   ├── 04_pngs.R              # static PNG export
│   └── 05_assumption_checks.R # VAR(4) diagnostics -> output/assumption_report.md
├── output/
│   ├── figures/           # PNG figures
│   ├── widgets/           # Chart.js fragments (*.html)
│   └── assumption_report.md
├── assets/                # banner and social preview (made by tools/banner)
├── .Renviron.example      # template for the FRED API key
├── CITATION.cff
├── LICENSE                # MIT (code)
└── LICENSE-docs           # CC BY 4.0 (text and figures)

The *.rds files are written to output/data/ at run time and are not committed. The widgets load Chart.js 4.4.1 from jsDelivr and follow the author's site embed contract (theme set by the is-light class on <body>). To reuse one, paste the HTML fragment into a page that serves Chart.js from the same CDN.

Citation

Archived on Zenodo: doi:10.5281/zenodo.23404665 (all versions). GitHub shows a "Cite this repository" button from CITATION.cff. BibTeX:

BibTeX

@misc{taylor_rule2026,
  author = {Khosravipour, Nik},
  title  = {Discretion Over Rules: Federal Reserve Policy Departures and the Inflation Surge of 2021--2022},
  year   = {2026},
  doi    = {10.5281/zenodo.23404665},
  url    = {https://github.com/nik110403/taylor-rule}
}

License

Code: MIT. Text and figures: CC BY 4.0. Third-party data (FRED, CBO) is not covered by either licence and stays subject to its source terms.