# US Total Factor Productivity Growth (5-year rolling average)

**Retrieved:** 2026-08-19
**Source URL:** https://www.frbsf.org/wp-content/uploads/quarterly_tfp.xlsx
**Primary citation:** Fernald, John G. (2014). *A Quarterly, Utilization-Adjusted Series on Total Factor Productivity.* FRBSF Working Paper 2012-19 (updated 2026-03-24).
**Producer:** Federal Reserve Bank of San Francisco, John Fernald and Shane Boyle
**License:** Freely available; © FRBSF, no explicit reuse license. (An earlier version of this file said "Public.")

## Columns in our CSV

- `year` — calendar year
- `tfp_growth_5yr_avg_pct` — 5-year centered rolling average of the `dtfp_util` column (utilization-adjusted TFP growth, annualised, percent)

## Companion file: `us_tfp_growth_annual.csv` (unsmoothed)

The same build script also writes `us_tfp_growth_annual.csv` (`year`,
`dtfp_util_pct`) — the raw annual `dtfp_util` values with no rolling
average. This is the **inference series for the spectral-verdict pipeline**
(`scripts/spectral_verdict.py`): the 5-year moving average barely attenuates
the 54–55-year band (|H| ≈ 0.99 at that frequency) but reddens the spectrum
and corrupts an AR(1) noise fit, so any statistical test must run on the
unsmoothed series. The rolled CSV remains the display series on the chart.
Both files are written from the same download in the same run, so they can
never drift apart. The 2026-08-19 refresh also picked up upstream Fernald
revisions confined to 2012–2025 (max |Δ| = 0.14 pp on the rolled values);
coverage is unchanged at 1948–2025.

## What was filtered and transformed

Source file is `quarterly_tfp.xlsx`, sheet `annual`. We took the `dtfp_util` column (annual utilization-adjusted TFP growth in percent), then computed a centered 5-year rolling average (window of 5, clipping at the edges so the boundary years average a smaller window rather than dropping out). Coverage: 1948–2025. Reproducible: `scripts/build_us_tfp_growth.py`. Verification: `scripts/verify_tfp.py` confirms the local CSV is a perfect match (78/78 rows within 0.0001) against `dtfp_util` centered-5.

An earlier version of this CSV was silently derived from the `dtfp` column (raw TFP growth, no utilization adjustment), which contradicted the documentation calling the series "utilization-adjusted." The current CSV is rebuilt from `dtfp_util` to match the documented intent.

## Why this series pairs with Kondratiev

Kondratiev's long-wave framework predicts 50–60 year cycles of technological paradigm expansion and exhaustion. TFP growth is the most direct measurable output of a technological paradigm: high during diffusion of a general-purpose technology, low during maturity and exhaustion. The high-TFP post-WWII period is exactly Kondratiev's expansion phase; the slowdown starting in the early 1970s matches the top of the post-war K-wave.

## Caveats

- Coverage starts 1948. For Kondratiev's earlier waves (pre-1948) you'd need Bergeaud–Cette–Lecat or similar long-run TFP estimates.
- 5-year centered window dampens the most recent year and the earliest year by clipping. A one-sided rolling window would differ.
- TFP measurement itself is contested — utilisation adjustment is the Fernald approach, not a universal standard.