Census Building Permits and Housing Starts: The Federal Leading Indicator Behind the US Housing Market

The Census Bureau publishes three linked monthly surveys that together form the federal government's primary leading indicator for US housing activity: the Building Permits Survey, the New Residential Construction release covering housing starts, and the New Residential Sales report. These three releases — each a joint product of the Census Bureau and the Department of Housing and Urban Development — trace the lifecycle of a new home from the moment a permit is issued through the signing of a sales contract, and they do it faster than any other federal economic series.

Three Linked Monthly Releases

Understanding US housing requires understanding how the three Census/HUD housing releases relate to each other in time. They are not independent snapshots — they are sequential measurements of the same pipeline.

The Building Permits Survey (BPS) captures the first formal step in residential construction: the issuance of a permit by a local jurisdiction authorizing a builder to break ground. Permits are a leading indicator because they precede actual construction by weeks to months. A builder who pulls a permit in January may start construction in February or March; in practice, the median lag from permit issuance to construction start is roughly one month for single-family homes and somewhat longer for large multifamily projects.

The New Residential Construction release — commonly called the housing starts report — measures how many units actually began construction in a given month, defined as the moment excavation begins. Starts lag permits by approximately one month in the aggregate, though some permitted projects are abandoned before breaking ground, and some construction begins in jurisdictions that do not require permits (those units are estimated separately). Starts then lead completions by roughly six months for single-family homes and up to twelve months or longer for large multifamily buildings, which require more time to build.

The New Residential Sales report tracks signed sales contracts for new single-family homes. A new home can be sold at any stage — before construction begins, while under construction, or after completion — so the sales release is partially coincident with starts and partially leading completions. It is a distinct survey from the National Association of Realtors existing-home sales report, which covers previously built homes and is a lagging rather than leading indicator (existing-home sales are counted at closing, after the contract is signed and the transaction is complete).

All three releases are published approximately three weeks after the reference month ends. Together they give economists, the Federal Reserve, and housing market analysts the most timely available read on where residential construction and home sales are heading.

The Building Permits Survey

The Building Permits Survey covers approximately 20,000 permit-issuing jurisdictions across the United States — counties, cities, towns, townships, and boroughs that require building permits for new residential construction. This universe captures roughly 96% of new residential construction activity. The remaining 4% occurs in areas that do not require permits, primarily rural counties and unincorporated territories; the Census Bureau estimates construction activity in those areas through a separate survey component called the Survey of Construction (SOC).

Each month, local building departments report to the Census Bureau the number of building permits issued, the number of units authorized under those permits, and the number of buildings authorized, broken down by structure type: single-family (one unit), two-unit, three- and four-unit, and five-or-more-unit buildings. The five-or-more category captures the bulk of multifamily apartment construction; a single permit for a 300-unit apartment complex counts as one permit but 300 units.

The headline figure from the BPS is the total units authorized expressed as a seasonally adjusted annual rate (SAAR). Because construction activity varies strongly with weather and season — fewer permits are pulled in January in Minnesota than in July — the raw monthly count is adjusted to remove predictable seasonal patterns and then annualized by multiplying by 12. A SAAR of 1.5 million means the current month's pace, if sustained for a full year, would produce 1.5 million newly authorized units.

The BPS provides geographic breakdown to the metropolitan statistical area (MSA) level and to the county and place (city/town) level. HUD's State of the Cities Data Systems (SOCDS) extends the place-level BPS data back to 1980, making it possible to track permit activity in individual cities and CDPs (census-designated places) over four decades. This long historical record is particularly valuable for analyzing how local zoning and development policy changes have affected permitting trends in specific jurisdictions.

Housing Starts: What Gets Measured and How

Housing starts are measured by the Census Bureau's Survey of Construction (SOC), which covers approximately 900 geographic sample areas selected to represent the full national distribution of construction activity. Field agents visit sample areas to identify new construction that has begun since the previous visit, regardless of whether a permit was pulled. A housing start is defined as the moment excavation begins for the footings or foundation of a building.

The SOC distinguishes two primary structural categories that behave very differently as economic indicators. Single-family homes (structures with one housing unit, including attached townhomes built on separate lots) respond quickly to changes in mortgage rates, consumer confidence, and builder sentiment. Their permit-to-start lag is short and their start-to-completion cycle averages roughly seven months. Multifamily buildings, defined as structures with five or more units, are driven primarily by apartment demand, rental vacancy rates, and construction financing rather than by retail mortgage rates. Their permit-to-start lag can be substantially longer due to the complexity of construction financing and municipal approvals, and their start-to-completion cycle for large buildings often exceeds twelve to eighteen months.

The relationship between permits and starts is not one-to-one. Some issued permits expire before construction begins — a builder who pulls a permit but cannot secure financing or faces cost overruns may let the permit lapse. Conversely, some construction activity begins in non-permit areas and is counted in starts but absent from the BPS. At the national level, the aggregate ratio of starts to permits fluctuates around 0.95 to 1.0, but the ratio can diverge meaningfully at the state or MSA level depending on local construction economics and permitting efficiency.

Historical Context: From the 2005 Peak to the Pandemic Surge

Total housing starts peaked at approximately 2.07 million SAAR in January 2006, near the height of the housing bubble. That figure reflected a decade of credit expansion, loose mortgage underwriting, and speculative construction in Sun Belt markets — particularly in Florida, Nevada, Arizona, and the Inland Empire of Southern California — that dramatically overbuilt relative to sustainable household formation demand.

The collapse was correspondingly severe. By April 2009, total housing starts had fallen to approximately 554,000 SAAR — a decline of 73% from the peak in just over three years. Single-family starts were hardest hit, falling from roughly 1.7 million at the peak to under 360,000 at the trough. The recovery was slow and incomplete; a decade after the peak, total starts in 2016 remained below 1.2 million, well short of the level needed to keep pace with household formation.

The 2020–2021 period produced a surge driven by pandemic-era dynamics. Remote work enabled millions of households to relocate from high-density urban cores to lower-density suburban markets, accelerating demand for single-family homes precisely as the Federal Reserve held the federal funds rate at the zero lower bound and the 30-year fixed mortgage rate fell below 3% for the first time in modern US history. Total housing starts reached approximately 1.8 million SAAR by early 2021, a fifteen-year high. Single-family starts peaked at roughly 1.2 million SAAR in early 2021 before supply chain disruptions — particularly in lumber, windows, and electrical components — began constraining builder capacity.

The 2022–2023 pullback was sharp for single-family construction and represented the most direct transmission of Federal Reserve monetary policy into the real economy in modern experience. As the 30-year fixed mortgage rate climbed from below 3% in early 2022 to above 7% by late 2022 and into 2023, single-family starts fell from their 2021 peak of roughly 1.2 million SAAR to under 800,000 SAAR by the end of 2022. Monthly permit data telegraphed the collapse two months before starts data confirmed it, illustrating the leading indicator value of the BPS.

A notable geographic shift accompanied the pandemic-era surge: Sun Belt markets dominated new construction volume. Texas, Florida, North Carolina, Georgia, and Arizona accounted for a disproportionate share of national starts as households relocated from higher-cost coastal metros. The Houston, Dallas, Phoenix, and Jacksonville MSAs consistently ranked among the highest-volume permitting metros throughout the 2020–2024 period, a pattern traceable in the HUD SOCDS place-level permit data.

Single-Family vs. Multifamily: The 2022–2023 Bifurcation

One of the most analytically striking features of the 2022–2023 housing slowdown was the divergence between single-family and multifamily construction. Single-family starts collapsed as mortgage rate increases made homeownership dramatically less affordable for the marginal buyer. Multifamily starts, by contrast, held up and in some months increased during the same period.

The explanation lies in the different demand drivers for the two segments. Single-family home buyers take out retail mortgages; when the 30-year fixed rate rises from 3% to 7%, the monthly payment on a $400,000 mortgage rises from roughly $1,686 to $2,661 — a 58% increase — which prices a meaningful share of potential buyers out of the market. Apartment renters, by contrast, do not take out mortgages to occupy a unit; their rental costs are influenced by construction financing rates but not directly tied to the 30-year fixed rate. As homeownership became less affordable, demand for rental apartments strengthened in many markets, sustaining multifamily starts even as single-family construction fell.

This bifurcation also exposes the structural underproduction of what housing economists call the “missing middle”: townhomes, duplexes, triplexes, and small apartment buildings with two to four units. The Census BPS data shows this segment — two-unit and three-to-four-unit structures — has been chronically underbuilt relative to its historical share for decades. Exclusionary zoning in most US municipalities prohibits two-to-four-unit structures on lots zoned for single-family use, eliminating the middle of the density spectrum and forcing the market to choose between detached single-family homes and large apartment buildings. The 2022–2023 bifurcation made the missing middle particularly visible: markets that needed modestly denser attainable housing had almost no production of that type.

Geographic Data: SOCDS and MSA-Level Starts

Two geographic data systems provide the granular location detail that the national headline figures obscure. HUD's State of the Cities Data Systems (SOCDS) publishes building permit data by place — individual cities, towns, and census-designated places — back to 1980. The SOCDS database allows analysts to track permit trends in specific cities over four decades, comparing construction activity across boom and bust cycles, identifying jurisdictions where permitting has accelerated or collapsed, and examining how state-level housing policy changes have filtered down to municipal permitting behavior.

The Census Bureau publishes MSA-level housing starts as part of the New Residential Construction release, covering the 25 largest metropolitan areas separately and reporting the remainder by Census region (Northeast, Midwest, South, West). The regional breakdown shows that the South has consistently accounted for roughly 55 to 60% of total US housing starts since 2000, reflecting both population migration patterns and the lower land costs, less restrictive zoning, and more favorable construction economics that characterize Sun Belt markets relative to coastal metros.

The concentration of starts in Sun Belt states is striking in the BPS data. In recent years, Texas alone has typically accounted for 15 to 18% of all US housing permits — more than the entire Northeast Census region. Florida contributes another 10 to 12%. North Carolina, Georgia, and Arizona each contribute 4 to 6%. Together, these five states can account for nearly half of all US residential permits in strong building years, even though they represent a much smaller share of the existing housing stock.

New Residential Sales and the New-Home Market

The New Residential Sales report measures signed sales contracts for new single-family homes and is published jointly by the Census Bureau and HUD in the third or fourth week of the month following the reference month. The headline figures include the monthly count of new home sales (expressed as SAAR), the median sale price, the average sale price, and the months supply of new homes for sale.

Months supply is calculated as the inventory of homes for sale divided by the current monthly sales rate. A months supply of 5 to 6 is generally considered equilibrium for the new-home market; lower values indicate a seller's market with limited buyer choice, and higher values indicate oversupply that will pressure builder pricing and production decisions. During the 2005–2008 bubble, months supply of new homes climbed to 10 to 12 months as unsold inventory piled up. During the 2020–2021 surge, months supply fell below 4, its lowest level in decades.

The new-home sales report differs meaningfully from the NAR's existing-home sales report in what it measures and when it leads. New-home sales are counted when the sales contract is signed, which can be months before closing; existing-home sales are counted at closing. This makes new-home sales a genuinely leading indicator of future construction and closing activity, while existing-home sales lag the underlying demand signal by 30 to 60 days. In a rising-rate environment, new-home sales data reveals the demand shock to builders faster than any other transaction-based measure.

The median new-home sale price and the mix of homes sold also provide insight into builder strategy. During the 2022–2023 rate shock, many large builders shifted their product mix toward smaller homes and entry-level price points and began offering mortgage rate buydowns — paying points to reduce the buyer's effective rate — as a sales incentive. These dynamics are visible in median price trends in the New Residential Sales data and in builder earnings releases, but the Census release is the only federal source capturing the full builder-funded portion of the market.

Housing Starts, Permits, and Fed Policy

Housing starts and building permits occupy a prominent place in Federal Reserve monetary policy deliberations because residential construction is one of the most interest-rate-sensitive sectors of the US economy. The transmission mechanism from Fed rate decisions to housing activity is faster and more direct than in most other sectors: the 30-year fixed mortgage rate tracks the 10-year Treasury yield with a spread of roughly 150 to 200 basis points under normal conditions, and mortgage rates move within days of Fed communications that shift market expectations.

The Federal Open Market Committee's 2022–2023 tightening cycle targeted housing inflation explicitly. Fed Chair Jerome Powell identified the “lock-in effect” by name in public remarks: millions of existing homeowners held mortgages at 2.5 to 3.5% and faced little financial incentive to sell their homes and take on a new mortgage at 7%, which suppressed listing inventory for existing homes and placed unusual importance on new construction as the primary source of available housing supply. Builders with the financial capacity to buy down buyer mortgage rates became dominant in markets where affordability was most stressed.

The Federal Reserve Board's Beige Book — its qualitative economic summary compiled from regional Fed banks — tracks builder and real estate contact reports on construction activity. The Beige Book, combined with the quantitative BPS and housing starts data, forms the primary housing market evidence base used in FOMC deliberations. When building permits turn down sharply before starts, as occurred in late 2022, the Fed staff can project that starts and completions will slow six to twelve months forward, which factors into rate path projections.

Lumber and Construction Materials

Single-family construction economics are closely linked to lumber prices, which affect the cost of the framing package — the structural wood framework of a house — that is typically the largest variable cost component in SFH construction. The CME random length lumber futures contract (LBS) is the benchmark for framing lumber traded in North America, denominated in dollars per thousand board feet (MBF).

The 2021 lumber price spike was extreme by historical standards. Random length lumber futures reached approximately $1,700 per MBF in May 2021, compared to a pre-pandemic baseline of roughly $350 to $450 per MBF. The spike reflected simultaneous demand surges from US new construction, Canadian and US DIY home improvement activity, and mill production cuts made at the onset of the pandemic that took months to reverse as mills recalled workers and restarted capacity. At $1,700 per MBF, the lumber cost premium on a typical 2,000-square-foot single-family home relative to pre-pandemic prices was estimated at $35,000 to $50,000, a substantial addition to build costs that some builders passed through to buyers and others absorbed as reduced margin.

Lumber prices subsequently collapsed — returning to the $350 to $500 range by late 2022 — but framing lumber remained just one constraint among several. Supply chains for windows, engineered wood products (OSB and LVL beams), HVAC equipment, electrical panels, and garage doors experienced independent disruptions during 2021–2022 that extended construction cycle times and raised builder costs beyond the lumber component alone. The BPS and SOC data do not directly track material costs, but the divergence between issued permits and actual starts during that period — permits running ahead of starts as builders waited for components — is visible in the data and reflects supply chain constraints on construction capacity.

Accessing the Data: API, FRED, and SOCDS

The Census Bureau Building Permits Survey is accessible via the Census API at api.census.gov/data/timeseries/bps. The API supports queries by state, MSA, county, and place, with time-series parameters for year and month. Structure type is controlled by the unittype parameter: 0 returns all types combined, 1 returns single-family only, and 5 returns five-or-more-unit structures. No authentication is required for basic queries, though a free API key obtained from the Census registration page removes rate limits.

FRED (the St. Louis Fed's economic data portal) publishes the key derived series in seasonally adjusted annual rate form, updated monthly:

• PERMIT — total privately owned housing units authorized by building permits, SAAR.
• HOUST — total housing starts, SAAR.
• HOUST1F — single-family housing starts, SAAR.
• HOUST5F — housing starts in buildings with five or more units, SAAR.
• HSN1F — new single-family houses sold, SAAR (from the New Residential Sales report).
• MSPNHSUS — median sales price of new houses sold in the United States.

All FRED series are freely accessible via the FRED API at api.stlouisfed.org/fred/series/observations with a free API key. The FRED series are the fastest path to a clean time series for visualization or econometric modeling; the Census API is the appropriate source when geographic granularity below the regional level is required.

HUD's SOCDS place-level permit data is accessible at the HUD User portal (huduser.gov/portal/datasets/socds.html). The interface allows downloading building permit data for a selected place across the full history from 1980 to the present, broken down by structure type. Bulk downloads of all places in a state are available as CSV files. This is the appropriate source for any analysis of permitting trends in specific cities, towns, or CDPs — questions such as whether a particular city has increased permitting following a zoning reform, or how a coastal city's permit volume compares to its peer cities over forty years.

Python: State-Level Permit Trends from the Census API

The following script calls the Census Building Permits API to download monthly permit counts by state for the past five years, computes a twelve-month rolling sum for each state, and ranks states by year-over-year growth in rolling permit volume. The rolling sum is used rather than a raw monthly comparison because month-to-month permit counts are volatile; the twelve-month sum smooths seasonal variation and permit backlog fluctuations, providing a cleaner read on underlying construction trend.

import requests
import pandas as pd
from datetime import date

# Census Building Permits API (BPS)
# Endpoint: api.census.gov/data/timeseries/bps
# Pull monthly permit counts by state for the last 5 years.
# "units" = total units authorized, "bldgs" = buildings authorized.
# "unittype" = 1 (single-family), 2 (2-unit), 3 (3-4 unit), 5 (5+ unit)
# Use unittype=0 for all types combined.

API_KEY = "YOUR_CENSUS_API_KEY"  # register at api.census.gov/data/key_signup.html

END_YEAR = date.today().year
START_YEAR = END_YEAR - 5

# Build list of (year, month) combos we want
records = []

for year in range(START_YEAR, END_YEAR + 1):
    end_month = 12 if year < END_YEAR else date.today().month
    for month in range(1, end_month + 1):
        params = {
            "get": "state_name,units,bldgs",
            "for": "state:*",
            "time": str(year) + "-" + str(month).zfill(2),
            "unittype": "0",          # all structure types combined
            "key": API_KEY,
        }
        resp = requests.get(
            "https://api.census.gov/data/timeseries/bps",
            params=params,
            timeout=60,
        )
        if resp.status_code != 200:
            continue
        data = resp.json()
        cols = data[0]
        for row in data[1:]:
            rec = dict(zip(cols, row))
            rec["year"] = year
            rec["month"] = month
            records.append(rec)

df = pd.DataFrame(records)

# Coerce numeric columns
df["units"] = pd.to_numeric(df["units"], errors="coerce").fillna(0)
df["bldgs"] = pd.to_numeric(df["bldgs"], errors="coerce").fillna(0)
df["period"] = pd.to_datetime(
    df["year"].astype(str) + "-" + df["month"].astype(str).str.zfill(2) + "-01"
)

df = df.sort_values(["state_name", "period"])

# Compute 12-month rolling sum of units for each state
df["rolling_12m"] = (
    df.groupby("state_name")["units"]
    .transform(lambda s: s.rolling(12, min_periods=12).sum())
)

# Compare most-recent 12-month rolling total to the one 12 months prior
latest_period = df["period"].max()
cutoff = latest_period - pd.DateOffset(months=12)

latest = df[df["period"] == latest_period][["state_name", "rolling_12m"]].rename(
    columns={"rolling_12m": "recent_12m"}
)
prior = df[df["period"] == cutoff][["state_name", "rolling_12m"]].rename(
    columns={"rolling_12m": "prior_12m"}
)

ranked = latest.merge(prior, on="state_name", how="inner").dropna()
ranked["growth_pct"] = (
    (ranked["recent_12m"] - ranked["prior_12m"]) / ranked["prior_12m"] * 100
).round(1)
ranked = ranked.sort_values("growth_pct", ascending=False)

print("State permit growth (12-month rolling total, year-over-year):")
print(ranked[["state_name", "recent_12m", "prior_12m", "growth_pct"]].to_string(index=False))

The script returns a ranked table of states with their most-recent twelve-month permit total, the total from twelve months prior, and the year-over-year growth rate. States with sharp increases in the rolling total are either experiencing genuine demand surges or processing a backlog of permits issued after a period of administrative slowdown; the SOCDS place-level data or local news sourcing can disambiguate those cases. States with persistent declining rolling totals — particularly those where the decline began before mortgage rates rose, as would indicate supply-side rather than demand-side constraint — are candidates for deeper zoning and regulatory analysis.

To switch from all-types to single-family only, change the unittypeparameter from “0” to “1”. To pull county-level data instead of state-level, change the for parameter to “county:*” and add an “in”parameter specifying the state FIPS code. The API will return all counties in the specified state for the requested month.

Limitations and Analytical Cautions

The 4% of construction activity occurring in non-permit areas is estimated rather than directly measured. The Census Bureau's Survey of Construction field agents identify non-permitted starts through area sampling, but the estimates carry higher uncertainty than the permit-based counts. In states with large rural non-permit areas — particularly some Mountain West and Great Plains states — the BPS understates construction activity relative to total starts figures from the SOC.

Permit data reflects authorization, not intent or outcome. A developer who pulls permits for 200 units may build all 200, may build half, or may build none if the project falls through. In markets with long permit-to-start lags — some California jurisdictions have permit-to-start gaps of six to twelve months due to environmental review, design approval, and utility connection backlogs — a spike in permits may not translate to starts for a year or more. Analysts working in California, New York, or other high-regulation permitting environments should be cautious about treating permit trends as coincident with construction activity.

Seasonal adjustment introduces its own uncertainty. The Census Bureau applies X-13ARIMA-SEATS seasonal adjustment to the BPS and SOC data; the seasonally adjusted annual rate is the figure most commonly quoted in economic reporting, but it can be revised substantially when seasonal patterns shift. The 2020 pandemic disrupted seasonal construction patterns in ways that the existing seasonal adjustment factors did not anticipate, producing unusually large revisions in the permits and starts data during 2020–2021.

The New Residential Sales report has the widest margins of error of the three monthly releases. Monthly new-home sales counts are based on a relatively small sample of builders and are subject to substantial revision in subsequent months. The Census Bureau itself cautions that the 90% confidence interval for a single month's new-home sales figure can be plus or minus 15 to 20% of the point estimate; month-to-month changes should be interpreted with corresponding caution, and three-month moving averages are a more reliable signal than any individual monthly print.

Building permit activity feeds directly into FHFA House Price Index trends: markets with sustained high permit volumes tend to experience slower home price appreciation as supply responds to demand, while permit-constrained markets see sharper appreciation. See FHFA House Price Index: The Federal Repeat-Sales Benchmark for US Home Prices.

Construction employment is tracked at the sector level by the BLS Current Employment Statistics program, which publishes monthly job counts for residential and nonresidential specialty trade contractors and general building contractors. The CES construction payroll series closely tracks housing starts with a one- to two-month lag. See BLS Current Employment Statistics: The Monthly Federal Payroll Benchmark.

The Census American Community Survey complements the BPS by measuring housing stock characteristics, vacancy rates, and housing cost burden at the census tract level — the demand-side context for interpreting where new construction is and is not responding to need. See Census ACS: The American Community Survey and the Federal Demographic Dataset Behind Every Policy Decision.