HOUST: US Housing Starts Monthly SAAR Data from FRED (1959–2026)

The US Housing Starts series counts new privately-owned housing units on which construction began, monthly since 1959 at a seasonally adjusted annual rate, from the Census Bureau and HUD. It is the most rate-sensitive real activity series in US macro — the classic transmission channel of monetary policy — and among the most violent: from the all-time peak of 2.49 million (January 1972) to the 2009 trough of 478,000, a fall of four-fifths. Mid-2026 stands at 1.18 million, below the 1.3–1.5 million range most estimates place long-run household formation.

Dataset: US Housing Starts (1959–2026) · Updated 2026-06-01

Source: US Census Bureau and HUD · FRED series HOUST

Macro Takeaway

Dataset Overview

Dataset Variables

The CSV and Excel files contain the following columns.

A value of 1,177 means construction was starting at a pace of 1.18 million units per year.

Download the Complete Dataset

The full dataset is available in CSV and Excel formats.

FRED Direct CSV Access

The underlying data is available from FRED under series code HOUST:

https://fred.stlouisfed.org/graph/fredgraph.csv?id=HOUST

Direct CSV Access — Eco3min Structured Dataset

https://eco3min.fr/dataset/housing-starts.csv

This URL returns the complete dataset in CSV format. It can be used directly in pandas, R, curl, or any data tool.

Using the Dataset in Python

import pandas as pd

url = "https://eco3min.fr/dataset/housing-starts.csv"
df = pd.read_csv(url, parse_dates=["date"])

print(f"Latest pace: {df['housing_starts'].iloc[-1]/1000:.2f} million units/year")
print(df.tail())

Using the Dataset in R

library(readr)

url <- "https://eco3min.fr/dataset/housing-starts.csv"
df <- read_csv(url)

tail(df)
summary(df$housing_starts)

Both examples load the dataset directly from the URL — no download or API key required.

Methodology

A start is counted when excavation begins for the footings or foundation, from the Census Bureau’s Survey of Construction. The headline covers all privately-owned units — single-family homes and units in multifamily buildings — expressed at a seasonally adjusted annual rate.

Starts sit in the middle of the construction pipeline: permits lead them by one to two months, completions follow by six to twelve depending on structure type.

This dataset is updated daily (Mon–Sat, 08:00 UTC) via automated pull from the FRED API; new observations appear with each source release.

Data Quality & Provider Notes

• Release latency. ~3 weeks after the reference month, jointly with permits and completions.
• Revisions policy. Two subsequent monthly revisions plus annual benchmarking; initial prints regularly move 2–4%.
• Sampling error. The monthly change carries a confidence interval of roughly ±10% — the Census explicitly notes many monthly moves are not statistically significant.
• Weather sensitivity. Winter distortions are large even after seasonal adjustment; harsh-weather months routinely produce misleading prints.
• Known gaps. None; continuous since 1959.

Common Pitfalls When Using Housing Starts

1. Trading the monthly print. With a ±10% confidence band, single-month starts moves are mostly noise; the 3-month average is the minimum readable unit.
2. Reading the total without the single/multi split. Since 2022 the two segments have moved in opposite directions; aggregate stability can hide a multifamily bust offset by single-family resilience.
3. Confusing starts with supply. Units started are not units delivered — completions lag by up to a year, and cancellations occur; the housing-supply impulse arrives with a long delay.
4. Anchoring on the 1972 record. The 2.49M peak served a population two-thirds smaller with different household formation; per-capita normalization changes cross-era rankings entirely.

Historical Regimes

1959–1979 — The boom-bust decades. Starts repeatedly spiked above 2 million (record: 2.49M, January 1972) and crashed below 1.2 million — demographic demand from boomer household formation colliding with stop-go monetary policy.

1980–1982 — The Volcker crush. Double-digit mortgage rates cut starts to ~840k — the sharpest policy-induced housing contraction until 2008, and the era’s textbook case of rate transmission.

1983–2005 — The long moderation. Two decades mostly between 1.3 and 1.8 million, with the 1991 dip — housing as steady expansion fuel, culminating in the 2.27M cycle peak of January 2006.

2006–2009 — The great collapse. Starts fell 79% to 478k (April 2009) — the deepest housing depression in the modern record and the real-economy core of the GFC.

2010–2019 — The decade of underbuilding. The slowest housing recovery on record: below 1 million until March 2013, ending 2019 near 1.4M — the cumulative deficit that reshaped 2020s housing economics.

2020–2026 — Boom, rate shock, plateau. A pandemic surge to 1.81M (April 2022), then the fastest mortgage-rate repricing in four decades pulled starts to ~1.18M by mid-2026 — with multifamily and single-family on opposing paths beneath the flat total.

Related Macroeconomic Datasets

Starts are the mid-pipeline construction gauge; the series below cover what leads them, what follows, and what finances them.

Macroeconomic Dataset Hub

This dataset is part of the Eco3min macro-financial data repository.

Sources

• US Census Bureau and US Department of Housing and Urban Development — New Residential Construction
• Federal Reserve Bank of St. Louis — FRED series HOUST

Dataset Reference