Colorado Housing Costs: County-Level Analysis

Across all 64 Colorado counties, housing costs have risen faster than household incomes over the past decade. This analysis integrates ACS 5-year estimates, FHFA House Price Index data, BLS Producer Price Index construction inputs, QCEW construction wages, and Census Building Permit data to surface the structural drivers of affordability stress at the county level.

Margin of Error (MOE) Caution: ACS 5-year estimates carry margins of error that can be substantial for small-population counties (fewer than 5,000 residents). County-level comparisons should be interpreted with caution. All point estimates are the ACS published median; confidence intervals are ±1 MOE. Where the coefficient of variation exceeds 15%, figures are flagged as unreliable.

Interactive County Maps

Maps are generated monthly from the Python pipeline and embedded below. If a map file is not yet available (first run), a placeholder is shown. Use the pipeline script to regenerate: python scripts/build_co_housing_costs_insight.py --refresh.

Current Housing Cost Indicators (2020–2024 ACS)

10-Year and 15-Year Windowed Change

Using non-overlapping ACS 5-year cohorts — 2020–2024 (current), 2010–2014 (10-year baseline), and 2005–2009 (15-year baseline) — we calculate the proportional change in median gross rent between endpoints.

FHFA House Price Index Appreciation

The FHFA All-Transactions House Price Index (county-level) measures home price appreciation using repeat-sales methodology, independent of ACS surveys.

Construction & Development Cost Pressure

Rising housing costs are not driven solely by demand. Supply-side constraints — including escalating construction input costs, labor wage growth, and permit bottlenecks — play a critical structural role.

BLS Producer Price Index: Construction Inputs

The chart below tracks key BLS PPI series for construction-related inputs (lumber, concrete, steel, labor) from 2010 to the most recent available month. Rapid PPI increases translate directly into higher per-unit development costs.

QCEW Construction Wages by County

Average weekly wages for construction workers (NAICS 23) from the Quarterly Census of Employment and Wages (QCEW) reflect the labor cost component of new development.

Building Permits Per Capita

The Census Building Permits Survey (BPS) tracks authorized new residential units at the county and place level. Permits per capita is a proxy for supply responsiveness — counties with low permit rates relative to population growth face the sharpest affordability stress.

What Best Predicts Home-Price Growth? A driver-ranking model across all 64 Colorado counties ?

Across Colorado's 64 counties, home prices have grown at very different rates over the past decade. Which county-level factors best explain the gap between counties that boomed and counties that stayed flat? We test five candidates — income growth, current vacancy rate, share of renters paying ≥30% of income on housing, building permits per capita, and average construction wages — and let a statistical model (ElasticNetCV) rank them by predictive weight. The model is built to handle messy real-world data: when two predictors move together (income growth and construction wages, for example), it splits the credit cleanly instead of arbitrarily picking one; and it re-checks itself on rotating subsets of counties so the ranking generalizes rather than memorizing patterns in any one set.

Want the technical detail or wondering what "ElasticNetCV" actually means? Expand the plain-English explainer below ↓ or hover the ? icon next to the heading.

🧠 What is an ElasticNetCV model? (plain English)

The question we're asking: across Colorado's 64 counties, which housing inputs best predict how much home prices have grown over the past decade?

We feed the model five candidate drivers per county — income growth, permit issuance, construction wages, vacancy rate, and rent burden — and let it figure out which ones actually matter for explaining the 10-year FHFA House Price Index change. The model produces a ranked list of standardized coefficients: a number for each driver where bigger means stronger predictive weight, and the sign tells you direction (positive = HPI grew faster when that driver was higher).

• "ElasticNet" = a penalty rule that shrinks weak predictors toward zero. So if vacancy rate doesn't really help explain HPI growth, the model will set its weight to nearly zero rather than pretending it matters. This is how it avoids overfitting and handles correlated inputs (income growth and construction wages tend to move together — the elastic-net penalty splits the credit cleanly).
• "CV" = cross-validation. Instead of fitting once and trusting the answer, the model re-fits itself on rotating subsets of counties (5-fold here) and keeps the version that generalizes best. This guards against the model just memorizing patterns specific to the 52–58 counties with complete data.
• "α via LOOCV MSE" = the strength of the shrinkage penalty is tuned by leaving one county out at a time, re-fitting, and picking the penalty value that minimizes total prediction error. Sounds fancy; in practice it just means "let the data pick the dial."

How to read the ranking table below: the top row is the strongest predictor, the magnitude is on a unit-free scale (so you can compare apples to apples), and the sign matters — positive coefficient = "more of this driver, more HPI growth," negative = "more of this driver, less HPI growth." For example, vacancy rate having a negative coefficient means tighter markets (lower vacancy) saw stronger appreciation, which matches how housing markets actually behave.

What this is NOT: a forecast. The model describes relationships across counties in a specific 10-year window — it doesn't predict next quarter's HPI. Use the ranking to understand which structural factors associate with appreciation pressure, not as a market-timing tool.

The ranked feature importance table is exported to assets/co-housing-costs/snapshots/drivers_ranking.csv and reproduced below (regenerated monthly).

Methodology Notes

ACS Window Definitions

We use non-overlapping ACS 5-year survey cohorts to compute change over time:

• 2024 cohort: 2020–2024 (latest published 5-year estimates)
• 2014 cohort: 2010–2014 (10-year baseline)
• 2009 cohort: 2005–2009 (15-year baseline)
• Change formula: (endpoint estimate / baseline estimate) − 1

Non-overlapping windows avoid double-counting survey years and provide cleaner before/after comparisons. Overlapping 5-year windows (e.g., 2014–2018 vs. 2019–2023) share up to 4 survey years and can obscure true point-in-time changes.

FHFA HPI

The Federal Housing Finance Agency All-Transactions House Price Index is a repeat-sales index covering all owner-occupied properties with conforming mortgages. County-level data is available for Colorado counties meeting minimum transaction thresholds. Not all 64 counties have FHFA data; counties below the threshold are excluded from HPI-based analysis.

BLS PPI Series Used

• WPUFD4111: Construction materials (nonresidential building)
• PCU236200236200: Nonresidential building construction
• PCU331110331110: Iron and steel mills
• PCU331315331315: Aluminium sheet/plate/foil
• PCU327310327310: Ready-mix concrete

QCEW Construction Wages

BLS Quarterly Census of Employment and Wages (QCEW) provides county-level employment and wage data by NAICS code. We use NAICS 23 (Construction) for the most recent four-quarter average annual wage. Counties with fewer than three establishments may have suppressed data per BLS disclosure avoidance policies.

Census Building Permits Survey

The Census BPS reports monthly permit authorizations by county and place. We aggregate to annual totals and compute a 5-year rolling average permits per capita using ACS population estimates for the denominator.