We Proposed a Paper About Mortgage Rates. We Turned In a Paper About Income.
Introduction and Motivation
Our original proposal treated the 30-year mortgage rate as the main driver of the story, with income and unemployment sitting in the background as controls. That model didn't survive contact with the real data. Once we built out the full panel, the mortgage rate variable ran into a problem we couldn't work around, so we rebuilt the model with income as the primary driver and added the COVID and recession years as their own direct controls instead of trying to route everything through rates. What follows is the version that held up.
Housing affordability has become a growing concern in the United States, particularly in the years following the COVID-19 pandemic. Rising home prices, combined with shifting labor market conditions, have made it increasingly difficult for individuals to enter the housing market. The share of first-time homebuyers has fallen to historically low levels, while the median age of first-time buyers has reached record highs.
This paper examines the determinants of housing prices across U.S. states using a panel dataset from 2000 to 2023, about 1,200 state-year observations across all 50 states. The central question: what actually drives variation in housing prices across states and over time? We focus on median household income, the unemployment rate, and two major economic shocks, the COVID-19 pandemic and NBER-dated recession years.
Why This Matters
Home prices have basically gone one direction for the last twenty-plus years, through a housing crash, a slow recovery, and a pandemic, and that pattern is really why we picked this topic in the first place. If mortgage rates were the real engine, you'd expect prices to actually fall when rates spike the way they did recently. They didn't, not for long. What this data says instead is that income is doing most of the work, and income doesn't move in two-year cycles the way rates do, it just grinds higher over time and drags home prices with it. That's a much sturdier long-term story than betting on where the Fed lands next, and it's the reason this isn't just a class assignment to me. It's the same logic behind why real estate sits in both mock portfolios on this site, more on that below.
Literature Review
Income has long been treated as a core driver of housing demand. Case and Shiller (2003) use panel data across states and find a strong positive relationship between income and housing prices, since higher income simply means more purchasing power. Mian and Sufi (2009) find that higher unemployment significantly reduces housing demand and price growth across regions, which lines up with standard theory, since less income stability means less appetite to buy. More recent work has looked at shocks directly. Gupta, Mittal, and Van Nieuwerburgh (2022) examine the COVID-19 period and find that housing prices increased despite the uncertainty, largely driven by low interest rates and a shift in what people wanted out of their homes. This paper builds on that literature by combining income and unemployment with direct shock indicators for COVID and recession in one panel model.
Data
The dependent variable is the Housing Price Index (HPI), a repeat-sales index from the Federal Housing Finance Agency. The main independent variable is real median household income (U.S. Census / FRED). Unemployment comes from the Bureau of Labor Statistics. Two dummy variables capture the shocks: COVID equals 1 for 2020 through 2022, and Recession equals 1 for NBER-dated recession years. All variables are measured at the state-year level, all 50 states, 2000 to 2023.
Model and Methods
HPI = β1 + β2(Income) + β3(Unemployment) + β4(COVID) + β5(Recession) + ε The model is estimated with OLS using heteroskedasticity-robust standard errors, added after the Breusch-Pagan test rejected homoskedasticity at the 1% level (F(4, 1195) = 66.29, p < 0.001). The Ramsey RESET test also flagged the linear specification (F(2, 1193) = 16.89, p < 0.001), suggesting a log-linear version could fit better, worth a note for anyone building on this. Variance Inflation Factors all came in under 1.4 (mean 1.19), so multicollinearity isn't a real concern here despite adding four variables to one model.
Results
Income is positive and significant at the 1% level in every specification, moving from 0.00463 with income alone (R² = 0.567) to 0.00390 once unemployment, COVID, and recession are all included (R² = 0.618). Unemployment is negative and significant throughout, and gets more negative as more controls are added, from -4.518 to -5.710. The COVID dummy has a large, positive, statistically significant coefficient (55.686), which points to a real structural shift in the housing market during the pandemic, not just noise. The recession dummy is also positive and significant (21.279), which runs against what standard theory predicts. Our best explanation is that housing prices simply move slower than the rest of the economy, and the recession years in this sample overlap heavily with the pandemic period, when prices were already being pushed up by low rates and a shift in housing demand.
Conclusion
This project set out to answer one question: what actually drives variation in housing prices across states and over time. The clearest answer in this data is income. It came back positive, significant, and remarkably stable across every version of the model we ran, more so than unemployment and more so than either shock variable. That's the real finding here, affordability is fundamentally a labor market and income story before it's a rate story, which is the opposite of how we framed this back in March.
COVID is the other half of the story. The size and significance of that coefficient points to a real structural break in how the housing market behaves, not just a blip. The recession result cuts against standard theory, and the most honest explanation is that prices lag the rest of the economy, and our recession years overlapped almost entirely with the pandemic period, so we're likely picking up two effects tangled into one variable.
Policies that support income growth and labor market stability are, in a very direct sense, housing affordability policy. That's a bigger claim than the one we set out to test, and it's the one I'd stand behind.
The model still has real limits. There are no supply-side variables here, no housing inventory, construction costs, or zoning data, and state-level aggregation can hide what's actually happening inside any single metro or county. Those are the honest gaps.
Final Thoughts
If AJ and I did this again, the first thing I'd add is supply. Income and unemployment tell you about demand, but nothing here says anything about how many homes actually got built, or how hard it is to build one in a given state, and that feels like half the story we didn't tell. I'd want a housing inventory measure and something on permitting or zoning restrictiveness by state, since two states can have identical income growth and completely different price paths if one of them makes it nearly impossible to build.
What I actually learned doing this is that the variable I was most attached to going in, mortgage rates, turned out to be the wrong lead actor. It's a good reminder that a strong proposal is just a hypothesis, and the data gets the final vote. Income carried this paper in a way I didn't expect back in March, and figuring out why was the most interesting part of the whole project.
What This Means for My Portfolios
This is also why real estate sits in both of the mock portfolios I run on this site. The Martins book holds Realty Income and the Vanguard Real Estate ETF (VNQ) as a 20% real estate sleeve, and the Clarks book runs Digital Realty and VNQ at 10%. Both of those positions are really a version of the same argument this paper makes. Income and structural shifts, not interest rates, are what actually move real estate values over time, so a diversified real estate sleeve is a way to own that engine directly instead of just watching home prices climb from the sidelines as a renter. It's a big part of why I'm bullish on real estate as an asset class generally, and it's the direction I want to keep pulling on with more variables down the line.
You can see both books, and the actual reasoning behind every position, on the portfolio page: https://www.renfrocapital.com/portfolio
References
Case, K. E., & Shiller, R. J. (2003). Is there a bubble in the housing market? Brookings Papers on Economic Activity, 2003(2), 299–362.
Mian, A., & Sufi, A. (2009). The consequences of mortgage credit expansion: Evidence from the U.S. mortgage default crisis. Quarterly Journal of Economics, 124(4), 1449–1496.
Gupta, A., Mittal, V., & Van Nieuwerburgh, S. (2022). Work from home and the office real estate apocalypse. National Bureau of Economic Research Working Paper.
Bureau of Labor Statistics. (2023). Local area unemployment statistics. https://www.bls.gov Federal Housing Finance Agency. (2023). House price index. https://www.fhfa.gov Federal Reserve Bank of St. Louis. (2023). FRED economic data. https://fred.stlouisfed.org U.S. Census Bureau. (2023). Median household income in the United States. https://www.census.gov