What Actually Drives US Housing Prices?

Everyone has a theory about housing prices. Interest rates. Income. Foreign buyers. Greedy landlords. The Fed. But when you put each theory through rigorous statistical testing against 35 years of real data, most of them fall apart.

We ran 15 separate investigations, each testing a different economic variable against US housing prices from 1990 to 2024. Every dataset comes directly from FRED and the FHFA. Every correlation was checked for stationarity. Every causal claim was tested with Granger causality. No synthetic data. No cherry-picking.

The results are not what you'd expect.

We tested 12 variables. Only 5 showed statistically significant correlations with housing price growth. The top two — building permits and construction costs — are supply-side indicators, not the demand-side factors that dominate public debate.

Income, the factor most people blame, ranks dead last among significant predictors. Consumer sentiment — the Shiller "animal spirits" theory — shows no meaningful correlation at all.

Most variables move with housing prices — they're contemporaneous, useless for prediction. But M2 money supply is different. Last year's M2 growth predicts this year's housing prices with r = 0.61 and Granger-causes housing at p = 0.0007.

The chart above overlays housing price growth (gold) against M2 growth shifted forward one year (teal). The alignment is striking. When the Fed expanded M2 by 17.5% in 2020 and 14.9% in 2021, the housing surge of 2021–2022 followed like clockwork.

This is the only genuine leading indicator we found across all 12 investigations.

One of the most common claims is that easy credit drives housing prices up. Our Granger causality tests tell the opposite story:

Housing prices Granger-cause consumer credit (p = 0.006), not the reverse. People borrow because their house is worth more — the wealth effect — not the other way around. Similarly, housing drives consumer sentiment (p = 0.02), and banks tighten mortgage standards after prices drop (p = 0.0003).

The popular narrative has the causal arrows reversed.

We combined the top predictors into a multi-factor regression. After checking for multicollinearity (VIF) and running backward elimination, three variables survived:

What dropped out: Unemployment, lending standards, income, population, consumer sentiment, and the stock market all become insignificant when these three variables are present. Unemployment is absorbed by permits (both measure economic activity). Lending standards are absorbed by M2 (both measure monetary conditions).

The model is even stronger post-2008, explaining 90% of variance. In the QE era, monetary policy became the dominant force in housing.

The national model explains 79% — but anyone who's watched housing in Nevada vs North Dakota knows the national picture hides enormous variation. So we went deeper: all 50 states + DC, 1990–2024, using FHFA state-level house price indices (42 states with complete data used in the volatility model).

The variation is staggering. Nevada's housing volatility is 4.5× North Dakota's. Florida crashed 44% in 2008 while Iowa barely blinked.

Four distinct clusters emerged:

All eight states started at 100 in 1990. By 2024, California was at 278, Florida at 390 — but Nevada crashed from 360 to 160 in the 2008 crisis before climbing back. Ohio barely moved. Texas climbed steadily without the roller coaster.

So what makes some states amplify the national M2 signal while others dampen it?

We tested 11 candidate variables across all states. After VIF screening and backward elimination, four factors explain 84% of cross-state housing volatility:

The scatter plot above shows the key relationship: states where it's hard to build (low supply elasticity) have much more volatile housing. California (0.90) and Florida (1.00) sit at the inelastic extreme. North Dakota (5.09), Kansas (5.05), and Iowa (4.14) can build freely — so demand gets absorbed into new construction instead of price spikes.

The pre-2008 construction boom intensity tells an equally sharp story. States that went on a building frenzy in 2004–2007 (Nevada, Arizona, Florida) experienced the most violent crashes. The construction boom was both a symptom of speculation and a cause of the bust — overbuilding created the surplus that crashed prices.

Housing prices operate on two levels.

Nationally, three forces account for 79% of price movements: demand activity (permits), input costs (construction PPI), and monetary policy (M2, lagged one year). M2 is the only genuine leading indicator — the Fed's printing press today shows up in housing prices next year.

Locally, four factors explain 84% of why some states experience extreme boom-bust cycles while others stay calm: unemployment volatility, construction speculation, zoning regulation, and geographic land constraints.

Together, these findings explain why the same M2 surge makes Nevada swing 50 percentage points while North Dakota barely moves — and why Texas grows without the roller coaster.

What about real returns? Our CPI investigation (Inv 08) found that housing appreciates just 0.9% per year in real terms. The impressive-sounding 3.5% annual gain is mostly inflation. Housing is a decent inflation hedge over decades, but it's not the wealth engine people imagine.

A common narrative claims private equity firms are buying up all the housing. The data tells a different story.

The US has 148.7 million housing units. Of these, 87.8 million are owner-occupied (59%), 45.9 million are renter-occupied (31%), and 15.0 million are vacant (10%). The homeownership rate stands at 65.7% — near its historical average.

The rate peaked at 69.2% in 2004 during the housing bubble, crashed to 63.5% by 2016, and has slowly recovered. The current 65.7% is right where it was in the mid-1990s — before the bubble inflated.

Mom-and-pop landlords (owning 1–5 properties) control 80.6% of all rental homes. Mid-size investors (6–99 properties) hold 15.3%. Large investors (100–999) hold 0.9%. Institutional investors with 1,000+ properties — the private equity bogeymen — own just 0.7% of rental homes, roughly 300,000 units out of 45.9 million.

That's about 0.2% of all US housing.

The national number hides real local impact. Institutional buyers concentrate in Sun Belt metros with newer, affordable single-family homes. In Atlanta, 25% of single-family purchases in recent years went to institutional buyers. Jacksonville (21%), Charlotte (19%), and Phoenix (18%) show similar patterns.

So while PE firms own a tiny fraction nationally, they can meaningfully affect specific neighborhoods and price tiers — particularly starter homes in fast-growing Southern cities.

Sources: US Census Bureau (Housing Vacancies and Homeownership, Q4 2025), FRED (RHORUSQ156N, ETOTALUSQ176N), Redfin investor data, GAO Report GAO-24-106293. Metro-level institutional purchase shares from Parcl Labs and Redfin.

Data: 23+ datasets from FRED, FHFA, and MIT Urban Economics Lab (Saiz 2010 supply elasticity). Ownership data from FRED (RHORUSQ156N, ETOTALUSQ176N), US Census Bureau, Redfin, and GAO Report GAO-24-106293. Each download has a SHA-256 checksum receipt. No synthetic data. Period: 1990–2024 (ownership: 1965–2025).

Statistical pipeline: Each investigation followed a 4-step protocol: (1) Fetch real data with receipts, (2) Validate completeness and bounds, (3) Run gate tests (ADF stationarity, cointegration, VIF), (4) Analyze only if tests pass. Granger causality was run only on stationary series.

National model: OLS on 33 annual observations. VIF < 2.3, Jarque-Bera p = 0.89, Durbin-Watson = 1.55.

State model: Cross-sectional OLS on 42 states. VIF < 1.2, Shapiro-Wilk p = 0.24. Saiz elasticity aggregated from 269 MSAs to state level via population-weighted averaging.

Limitations: 33 national observations is small. Durbin-Watson 1.55 suggests mild autocorrelation. State model is cross-sectional (n=42), limiting predictor count. Supply elasticity data is from 2010 and may not reflect current regulation. Construction boom variable is backward-looking (pre-2008).