FRED-QD

Quarterly U.S. macroeconomic panel maintained by the Federal Reserve Bank of St. Louis. Sister dataset to FRED-MD, designed for longer-horizon forecasting and for factor models requiring quarterly frequency (GDP, productivity, balance-sheet aggregates). Loaded via macroforecast.load_fred_qd() when path.1_data_task.fixed_axes.dataset == "fred_qd".

Citation & authoritative source

  • Original paper: Michael W. McCracken and Serena Ng, “FRED-QD: A Quarterly Database for Macroeconomic Research,” Federal Reserve Bank of St. Louis Review 103(1): 1–44, 2021 (Q1). NBER working paper: NBER WP 26872 (March 2020). St. Louis Fed WP: 2020-005.

  • Official landing page: St. Louis Fed — FRED-MD & FRED-QD (documentation, appendix, vintages — same page as FRED-MD).

  • Variable appendix (current): FRED-QD_updated_appendix.pdf — the authoritative list of every series, T-code, source FRED ID, and variable description. The package does not redistribute the appendix.

What macroforecast downloads

  • Current vintage: https://www.stlouisfed.org/-/media/project/frbstl/stlouisfed/research/fred-qd/quarterly/current.csv (referenced in macroforecast/raw/datasets/fred_qd.py). Replaced by the maintainers at each quarter’s advance release.

  • Historical vintage: https://www.stlouisfed.org/-/media/project/frbstl/stlouisfed/research/fred-qd/quarterly/{vintage}.csv with vintage as YYYY-Q (e.g., 2020-06). Accessed when the recipe pins information_set_type == "real_time_vintage" plus leaf_config.data_vintage.

Same two-row header format as FRED-MD: row 1 holds the T-code per series, rows 2+ are quarterly observations starting 1959:Q1.

Structure — 14 variable categories

FRED-QD is benchmarked to the 200-series quarterly panel of Stock and Watson (2012a), extended with 48 additional series (net: 248 series), with emphasis on non-household balance-sheet aggregates and credit data that the original Stock–Watson panel did not cover. The current panel is grouped into 14 categories; the group membership is maintained in the appendix and evolves over time.

Representative group coverage (exact membership: see appendix):

  1. NIPA (National Income and Product Accounts) — real GDP (GDPC1), personal consumption expenditures (PCECC96), gross private investment, net exports, government spending.

  2. Industrial Production — IP total and sectoral indices (INDPRO, sector-level from FRED-MD consolidated).

  3. Employment and Unemployment — quarterly averages of payrolls, unemployment rate, participation rate, hours.

  4. Housing — starts, permits, existing sales (quarterly averages plus quarterly-only series like owner-occupier equity).

  5. Inventories, Orders, and Sales — manufacturing inventories, new orders, wholesale / retail sales.

  6. Prices — CPI, PCE price, core variants, commodity price indices.

  7. Earnings and Productivity — nonfarm-business labor productivity (OPHNFB), unit labor costs, average hourly earnings.

  8. Interest Rates — federal funds, Treasury yields, credit spreads.

  9. Money and Credit — monetary aggregates, bank reserves, total credit to the nonfinancial sector.

  10. Household Balance Sheets — household net worth, financial obligations ratios.

  11. Exchange Rates — trade-weighted dollar, bilateral major currencies.

  12. Stock Markets — S&P 500, dividend yield, realised volatility proxies.

  13. Non-Household Balance Sheets (new in FRED-QD) — corporate and non-corporate business liabilities, bank-sector balance sheets. This is the primary extension over the Stock–Watson 2012a panel.

  14. Other — miscellaneous series the paper treats separately from the 13 numbered groups.

Exact category counts vary across vintages — St. Louis Fed tracks it in the appendix.

Transformation codes (T-codes)

FRED-QD extends the FRED-MD T-code table with one additional entry (Case 8), reflecting a quarterly-specific transform pattern:

T-code

Transform

1

No transformation

2

First difference $\Delta x_t$

3

Second difference $\Delta^2 x_t$

4

Natural logarithm $\log x_t$

5

First difference of logs $\Delta \log x_t$

6

Second difference of logs $\Delta^2 \log x_t$

7

First difference of percent change $\Delta (x_t / x_{t-1} - 1)$

8

GARCH-type volatility transform (for specific market-volatility series only)

Case 8 is used sparingly — the paper notes that unit-root-test-driven T-code selection sometimes differs from codes used by prior literature, and the appendix flags these cases explicitly.

Flow into macroforecast is via the same tcode_policy axis as FRED-MD (1.1 does not override it).

Changes from the 2020 working paper / 2021 Fed Review publication to current

The working paper (NBER WP 26872, March 2020) and the 2021 Fed Review publication documented 248 series. Since then, the St. Louis Fed has maintained the panel; typical maintenance patterns are:

  • Series discontinuation when underlying FRED IDs retire (e.g., when a BEA table or Fed H.8 breakdown consolidates). Such discontinuations are flagged in the appendix’s history log; the composite series remains but its upstream mapping may move to a replacement FRED ID.

  • T-code revision for individual series when a new unit-root test outcome contradicts the paper-era code. The authors’ stated protocol is to revise T-codes based on ongoing unit-root test re-evaluation; revisions are documented in the appendix.

  • Additions when a new Fed release adds a directly useful aggregate — rare but non-zero. The 14-category structure accommodates additions without a group rename.

  • Balance-sheet refinements — the “Non-Household Balance Sheets” group was the working paper’s main contribution over Stock–Watson (2012a); refinements to this group (e.g., updating a “banks vs non-banks” split) occur as Fed source data gets restructured.

As with FRED-MD, the authoritative change log lives in the appendix PDF; macroforecast does not mirror it. Bit-identical replication of a published study requires pinning the exact vintage via leaf_config.data_vintage.

Loader behaviour

Mirror of FRED-MD:

  • Caches download at ~/.cache/macroforecast/raw/.

  • Parsing via parse_fred_csv (same shared helper).

  • No data redistribution. Network access or local_source path required on first load.

  • Schema: date index (quarter-end) + numeric columns with FRED series IDs as column names.

Known limitations in macroforecast v1.0

  • Same as FRED-MD: no per-variable metadata surface, no auto T-code validation across vintages, data_vintage required for real-time_vintage mode.

  • Quarterly / monthly alignment: when a study mixes FRED-QD with FRED-MD, aligning them requires the alignment_rule axis (1.5). v1.0 implements month-to-quarter aggregation (operational) but mixed-frequency evaluation is limited to single-frequency panels at each horizon.

See also