Can anyone still afford a home

Two numbers claim to measure the same thing. The first is the price of the median home divided by the median household income in the same metro. It is the number most stories lead with, and it moves slowly, because incomes and prices both compound. Across the 375 metropolitan areas in both the price index and the census release, weighted by owner households, the ratio stood at 4.8x in 2005 and 5.0x in 2024, a change of +3.4% over the whole span. The peak of the last cycle was 5.0x in 2006.

The second number is what the same home costs to carry: the monthly payment on a thirty year mortgage with 10% down at the rate of each quarter, twelve of them divided by the median income. That is the number a household lives with. It stood at 30.8% in 2005, fell for a decade as rates fell, and in 2024 it is 34.9% (interval 34.8% to 35.0%), against a government line of 30%.

Set the two side by side and the story of the last twenty years splits in half. The price story took two decades. The payment story happened between the first rate rise and the eighteen months that followed it. In 2024 the payment share sits +1.1 pts against the 2006 peak (interval +1.0 pts to +1.3 pts), and 283 of the 375 metros (75%) now sit above their own 2006 payment share. 161 metros (42%) sit above the 30% line for the median household. Affordable is a decision about which of these two numbers you let define the word.

Take the owner weighted median home as it was priced in 2019Q1 and ask what it costs to carry each quarter since. The payment was $1,316 a month at the start of the window. By 2026Q2 it was $2,704, a change of +$1,388. The chart splits that change exactly into two parts: what the price did, holding the rate at its starting level, and what the rate did, on the home as priced. The price part is the slow one. It built through the pandemic years and stands at +$839 in the latest quarter.

The rate part is the fast one. It was close to nothing while the thirty year rate sat near its lows, then it took over the chart in a few quarters. In the latest quarter the rate part is +$549, 40% of the whole change; it peaked at +$730 in 2023Q4.

The quarter the rate part overtook the price part was 2023Q3. That is the date of the payment story. The second half of this chapter asks what the shock did to the people who already owned. 18.4 million mortgages were originated in 2019 to 2021 at rates under 4.00%, and the lock in intensity is those loans per hundred owner households in each metro, from 10.7 at the tenth percentile to 36.6 at the ninetieth. Turnover, purchase originations per hundred owner households, fell from 5.28 in the base years to 3.71 after the shock. Regressed on the intensity with metro and year fixed effects, each point of lock in moved turnover by -0.056 (interval -0.077 to -0.034); the same regression on the two years before the shock, the placebo, gives +0.001 (interval -0.002 to +0.004). Between the tenth and ninetieth percentile metro that is -1.45 purchases per hundred owners.

The census asks every renter what share of income the rent takes and publishes the answer in bands. The two lines that matter are the ones the government draws: cost burdened at 30% of income and severely burdened at 50%. Across the 392 metros in 2024, 52.5% of renter households with a computed share pay 30% or more (interval 51.5% to 53.5%). In 193 of those metros (49%) the median renter sits under the line. A story that reports the median rent against the median renter income can report either of those facts and miss the other.

The severe band is the one a median hides entirely. 26.5% of renter households pay half or more of their income in rent (interval 25.7% to 27.2%), against the 20% line this chapter stated before it looked. Nationally the severe share went from 25.1% in 2005 to 26.1% in 2024, a change of +0.9 pts. By household income the picture is a cliff: among renters in the lowest income group 93% are severely burdened in 2024; in the highest group it is 1%. Owners with a mortgage sit beside them for scale: 30.9% pay 30% or more of income in owner costs.

Did the national severe series drift, or did it move and stay moved? A changepoint search with a penalty fixed before the data finds 3 breaks, in 2009, 2015, 2021. The years are on the chart; the reader can decide what each one was. What the search rules out is a smooth line.

Every metro is cut into 5 year windows from 2001 to 2025, 5 windows and 1,925 cells over 385 metros. For each cell the chart plots the building rate, permits per hundred households averaged over the window and shifted back one year so that permits lead prices, against the annual price growth of the window. The building rate runs from 0.34 at the tenth percentile cell to 2.55 at the ninetieth, around a mean of 1.28. The spec asked for permits per new household; that ratio has no value in a metro that lost households, which 68% of cells avoid, so it is published as a descriptive beside the rate (median 0.56) and the decision is logged.

With metro and window fixed effects, each extra permit per hundred households is associated with -8.0% price growth over the window (interval -14.3% to -1.7%), with domestic migration per hundred residents over the window and the two years before it in the regression as the observed demand shock. Between the tenth and the ninetieth percentile building rate that is -17.7% over a window.

Take the demand control out and the estimate is -0.4% (interval -4.6% to +3.7%). The two are published side by side because the gap between them is the size of the objection: builders build where people are arriving, and a regression that cannot see the arrivals credits the building with what the demand did. The simulator, where the elasticity is planted at -0.025, puts the controlled estimate's mean bias at +0.0026 and the uncontrolled at +0.0058; under strong demand the uncontrolled bias is +0.0078 against +0.0031 with the control. Both estimates are pulled toward zero when builders also answer the price growth they are regressed on, and the recovery table on the data page quantifies that by condition, which is why the headline is read as a lower bound in size.

Rank the 382 metropolitan areas by the price of the median home in 2024 and you get the ranking every list publishes. The three most expensive are San Jose-Sunnyvale-Santa Clara; San Francisco-Oakland-Fremont; Santa Cruz-Watsonville; the three cheapest are Charleston; Weirton-Steubenville; Johnstown. 142 micropolitan areas are in the warehouse and the explorer but not in this ranking, because the one year census release covers only the largest of them and a ranking with holes in it is not a ranking.

Now rank the same metros by the payment on that home as a share of what the median household there earns, and carry the census margins of error through 10,000 Monte Carlo draws so that every rank has an interval. The three least affordable are Kahului-Wailuku; Los Angeles-Long Beach-Anaheim; Santa Cruz-Watsonville; the three most affordable are Charleston; Enid; Johnstown. The least affordable metro, Kahului-Wailuku, HI, asks 75.5% of the median income (interval 69.7% to 81.7%); the most affordable, Johnstown, PA, asks 16.1%. The rank correlation between the two rankings is 0.94, the mean rank change is 29.1 places and the largest is 171, for Lexington Park, MD, from 72 by price to 243 by payment share. Among the 20 largest metros, 17 move beyond their rank interval. A third measure, the share of local households whose income clears the payment on the median home at the 30% line, runs from 11% to 75% across the metros with a median of 53%.

The loans say who got a mortgage. From the purchase subset of the mortgage disclosure extract, 3,533,474 originations across 5,029 lenders in 2019 and 2,722,577 across 4,341 in 2025, the median purchase borrower earned 1.23x the median household income of the metro in 2019 and 1.39x in 2025, a move of +0.15 (interval +0.12 to +0.19, resampling lenders). The share of buyers earning more than twice the local median went from 20.9% to 26.7%, +5.8 pts. The ratio moved most in Los Angeles-Long Beach-Anaheim, +0.44, and 17 of the 20 largest metros moved beyond their intervals.

Age is the other half of the question. Buyers under thirty five were 38.3% of purchase borrowers with a reported age in 2019 and 39.4% in 2025, a move of +1.2 pts (interval +0.1 pts to +2.3 pts). The median property value to borrower income went from 3.38x to 3.38x, and the rate spread over the benchmark sat at 0.19% for the median loan, from -0.69% to 1.06% across the deciles. Denials are not in this chapter: fair lending is a different build with a different standard of evidence, and the race and ethnicity fields were not loaded.

In 2020 and 2021 people moved. The shock this chapter measures is domestic migration per hundred residents over those two years less the same measure over 2018 and 2019, so that a metro which gained people every year is not credited with a pandemic it did not have. Across 385 metros the shock runs from -1.11 at the tenth percentile to +1.21 at the ninetieth. Against it the chart plots price growth from 2019Q4 to 2021Q4, which averaged 22.2%. The slope is +1.48% per person gained per hundred residents (interval +0.40% to +2.55%), which between the tenth and the ninetieth percentile metro is +3.4% of price growth. The regression a reader would run first, growth on the pandemic migration level alone with no control, gives +2.41% (interval +1.87% to +2.95%); it is on the shocks page because the simulator shows what the control removes.

The objection is that the metros that gained people were already on a different path. The placebo asks the same question of the years before the pandemic: growth from 2016Q4 to 2018Q4 on the same shock, with the migration of those years as its control. It is -0.78% (interval -1.35% to -0.20%). The message rule reads the two intervals together and has a branch for every outcome, including the one where the placebo moves.

For the 3 metros with the largest shocks, Panama City-Panama City Beach; Amherst Town-Northampton; Fairbanks-College, a synthetic control is built from the metros whose shock was small, with every donor searched and then treated in turn as a placebo. Panama City-Panama City Beach, FL ran +9.4% above its synthetic twin at the peak and -2.0% in the latest quarter, with a permutation p value of 0.829 over 192 donors; Amherst Town-Northampton, MA ran +1.5% at the peak and +0.9% now (0.725); Fairbanks-College, AK ran +0.5% and -6.0% (0.254). The gap after the shock window is the reversal, or its absence, and the chart shows it rather than naming it.

Everything above is about places. This chapter is about a household. At the 30% line, 10% down and the latest quarterly rate of 6.72%, the median household can afford the median home in 221 of 382 metros (58%); across metros the affordable price sits a median of +$21,570 from the median home, and a median of 53% of local households clear the payment on it. A renter at the renter median income saving 10% of income needs 6.4 years to reach the down payment in the typical metro and 16.8 years in Kahului-Wailuku, HI. The chart draws the rent or buy question for New York-Newark-Jersey City, NY-NJ: the cumulative cost of owning net of the equity built, less the cumulative cost of renting with the down payment invested instead, year by year to a 30 year horizon. Where the curve crosses zero is the breakeven; here it is beyond 30 years.

The paths behind the curve are not a forecast. Appreciation and rent growth are drawn in blocks of three years from the metro's own history, 400 paths per metro, and the band is the spread of those paths. Across the metros with enough history, 382 of them, the owner weighted median breakeven is 10.0 years (interval 1.0 years to beyond 30 years), and in 4% of paths owning does not catch up inside the horizon at all. One point of rate moves the median breakeven by +4.5 years. The quickest breakevens among the largest metros are Miami-Fort Lauderdale-West Palm Beach (3 years), Riverside-San Bernardino-Ontario (3 years) and Tampa-St. Petersburg-Clearwater (3 years); the slowest are Boston-Cambridge-Newton (21 years), Chicago-Naperville-Elgin (20 years) and San Francisco-Oakland-Fremont (18 years). The planner page runs the same arithmetic for any metro, income, rate, down payment and horizon, through the live API, and logs the check so that the next quarter of prices and rates can grade it.