City of Boston · MIT
Field experiment · 2024
In the first nationwide field experiment of flood-risk disclosure in the U.S. housing market, nearly a million registered Redfin users were randomly shown (or not shown) the flood risk of every home they viewed. Among those already shopping in high-risk areas, treated buyers placed offers on homes with 57% less flood exposure. The market repriced. Republican and Democratic counties responded the same.
Weekly difference between treated and control users, in the flood-risk score of homes they searched for. The treatment cohort had been searching homes in high-risk areas before the experiment.
Source: Fairweather, Kahn, Metcalfe & Sandoval-Olascoaga (NBER w33119, May 2026 revision) · CATE on flood score of properties searched, by week, for registered users with high-flood-risk baseline search behavior · 95% confidence intervals shown · Pre-experiment balance held across specifications
~964K
Registered Redfin users in the experiment, randomized over 12 weeks
−57%
Flood score of offers placed by treated high-risk buyers
−$8,900
Per-home drop on high-flood-risk non-waterfront homes when every Redfin viewer saw the score (1.9% of list price)
Why it matters
Flood risk in the United States is rising. Housing prices in exposed neighborhoods do not reliably reflect that risk. Buyers often don't know what they're buying, lenders don't always price it, and home prices don't track the science of where the water is going.
The standard policy lever is mandated disclosure, which requires sellers to tell buyers about risk. This experiment tested a weaker but more deployable cousin. Just show people the number, at the moment they're choosing. If buyers act on it, disclosure has bite. If they don't, the policy needs reinforcement elsewhere.
How we did it
From October 12, 2020 through January 3, 2021, Redfin ran an unannounced randomization on its platform. Treatment users saw a First Street Foundation Flood Factor score (1–10) on every property page they viewed; control users saw the standard listing. The primary analysis sample is the roughly 964,000 registered users with pre-experiment search histories. Sellers, agents, and users were unaware of the experiment. The information was just there.
We tracked the same individuals across every step of the funnel (search, tour, offer, close) and linked the platform behavior to MLS transactions. Identification rests on three estimators: an average-treatment difference-in-differences, a conditional-average-treatment-effect interacted with each user's baseline search risk-category, and a dynamic event-study. Standard errors clustered at the user level, with Romano-Wolf multiple-testing corrections (1,000 bootstrap resamples for search, 3,000 for offers). The hedonic price analysis uses random variation in the share of treated viewers per listing as an exogenous exposure measure, a Crépon-style design applied to homes rather than labor markets.
What we found
Search. Treated registered users who had been browsing the riskiest properties shifted toward homes with 12.4% less flood risk by week 9 (p<0.01). Effects grew over time. Pre-experiment balance held across all specifications.
Bids and closes. Those same users placed offers on homes with 57% less flood-score exposure than control buyers (p<0.01). Closings followed a qualitatively similar pattern. Treated users were not less likely to bid. They were bidding on different homes.
Trade-offs. Treated high-risk buyers were 35 percentage points less likely to place an offer on waterfront properties. They didn't adjust price, bedrooms, or square footage. They traded the water view for safer ground.
The market repriced. Scaling random viewer-exposure intensity from 0 to 100% within Redfin, high-flood-risk non-waterfront homes lost ~$8,900 (1.9%) in sale price. Low-risk homes gained about $4,000. Linearly extrapolated to full-market saturation, that implies a ~$112,250 (24%) price reduction on high-risk properties. Investors responded too. Moving from 0 to 100% treated raised the probability that a low-risk property was bought by an institutional investor by 2.5 percentage points over a 9.76% baseline.
No partisan split. "We cannot reject the null that Biden and Trump's counties respond to the flood score in the same way for both searches and offers." A hypothesis that the polarization literature predicted would split. It did not.
In the press
The New York Times
"While banks are becoming more careful about extending credit in areas that might flood, and homeowners are responding to flood risk disclosures on property search websites, the government-sponsored enterprises do not formally take such risks into account when underwriting loans."
Mortgage Regulators Are Shrugging Off Climate Risk. It Could Cost Taxpayers Billions.·December 2024
NPR Marketplace: Why Zillow removed climate risk information from listings
December 2025
HousingWire: Redfin defends climate risk transparency after Zillow removes data
December 2025
Bloomberg: Real Estate Listings With Flood Scores Shift Home-Shopper Habits
September 2022
Planetizen: Access to Flood Risk Data Shifts Homebuyer Behavior
September 2022
The Weather Channel: Home Buyers Reject Listings With High Flood Risk, According To Redfin
November 2022
Yale Environment 360: When Told About Flood Risk, Homebuyers Shun Vulnerable Homes, Study Finds
September 2022
Cite this paper
Fairweather, D., Kahn, M. E., Metcalfe, R. D., & Sandoval-Olascoaga, S. (2024). Expecting Climate Change: A Nationwide Field Experiment in the Housing Market (NBER Working Paper No. 33119). National Bureau of Economic Research.