Total homes needed
Owner supply and demand
Recent construction

Household growth (owner-occupied)
Five-year growth
What’s shaping demand
  • No major demand signals available.

Recent construction

Vacancy rate
Market balance
Vacant, for sale units
Healthy market need
Additional units needed

Housing loss and replacement

Vacancy mix
Current Change
For sale only
For rent

Rented / sold, unoccupied
Seasonal / occasional
Other vacant
Market size and change
5-yr. Change Pct. Ch.
Owner market
Owner permits

Renter market
Renter permits

The total market is occupied units + for sale/rent + sold/rented but not occupied.

Occupied units by age
Owner built 2010+
Owner built 1980-2009
Owner built before 1980

Renter built 2010+
Renter built 1980-2009
Renter built before 1980
County breakdown
Forecasted owner household growth and the strongest modeled demand signals in each county.

Methodology

What is being projected?
We project the change in owner-occupied households over the next five years for each county.

What data goes into the model?
The model uses recent housing market and local economic signals, including home prices, sales activity, recent household change, employment change, and domestic and international migration.

How does the model work?
For each county, we create a five-year “window.” We look at what the county looked like in one year, then measure how much owner households changed five years later. The model learns from many of those past windows.

Why use a rolling five-year model?
It lets the model learn from multiple time periods instead of just one. That helps capture how different local conditions have shaped growth over time.

How was it tested?
We used leave-one-window-out validation. That means we trained the model on earlier five-year periods, held one later period out, and tested how well it predicted that unseen period. Then we repeated that across the available windows.

What does that validation tell us?
It shows how the model performs on real historical periods it did not train on. That gives us a better sense of how it may perform in a future forecast.

What is the final forecast based on?
After testing, we fit the selected model on all valid historical five-year windows, then apply it to the most recent year of county data to project the next five-year change.

Why do the forecasts differ across counties?
Because the inputs differ. Counties with stronger recent migration, price growth, market activity, or household momentum may project differently than counties with weaker or negative recent trends.

Are these projections just a continuation of the last trend?
No. The model considers recent trend, but it also weighs market conditions, migration, and employment. So the forecast is based on several signals together, not just a straight-line extension.

What are the “drivers” shown in the dashboard?
They are grouped model contributions that help explain which factors pushed a county’s projection up or down the most, such as migration, employment, market conditions, or recent trend.