Fire in the Fine Print: Pricing Wildfire Risk Disclosure into California Home Values
An independent research project for my Economics senior seminar, later recognized with the Department of Economics' Brashear Research Prize: a difference-in-differences analysis of all 58 California counties testing whether the state's mandatory wildfire risk disclosure law changed how homes get priced in the areas most exposed to fire.
Does mandatory wildfire risk disclosure change how homes get priced?
California has become the epicenter of residential wildfire risk in the United States, and the stakes behind this question keep rising:
- The 2020 wildfire season was the largest in state history: over 4 million acres burned and an estimated $19.9 billion in insured losses
- The January 2025 Los Angeles fires destroyed more than 16,000 structures and caused upward of $135 billion in estimated damages
- Before 2021, prospective buyers had no formal way to know if a property sat in a state-designated Fire Hazard Severity Zone, while sellers often did
in estimated damages from the January 2025 Los Angeles fires alone, the kind of event that raises the real-world weight of a question that's usually confined to a textbook: do housing markets actually price in risk buyers can't see?
California's Assembly Bill 38, effective January 1, 2021, requires sellers of older homes to disclose when a property sits in a Fire Hazard Severity Zone: a classic information asymmetry problem. If buyers had been underpricing fire risk, disclosure should make prices fall once they can finally see what sellers already knew. My hypothesis: counties with more disclosure-triggering land should see larger relative home value declines after AB 38 took effect.
A 12-year, 58-county natural experiment
AB 38's 2021 effective date created a natural experiment: a before-and-after policy change I could exploit with a difference-in-differences design rather than a simple correlation.
- Built a panel of 690 county-year observations covering all 58 California counties, 2013 through 2024
- Merged five public datasets: Zillow's Home Value Index, CAL FIRE's Fire Hazard Severity Zone acreage, Census TIGER/Line land area, BLS unemployment rates, and Census median household income
- Constructed a continuous treatment variable, the share of each county's land classified as disclosure-triggering hazard territory, ranging from 0% in San Francisco and Imperial counties to 80% in Calaveras
- Estimated four model specifications: a continuous DiD, binary threshold splits at the median and 75th percentile, a quadratic dose-response test, and an event study to check that high- and low-hazard counties were trending similarly before the law
The Design
County and year fixed effects isolate the AB 38 effect from permanent county differences and California-wide trends, with standard errors clustered at the county level.
The Treatment Variable
A continuous measure of hazard exposure rather than an arbitrary in/out cutoff, so the model can pick up whether the effect scales with how much of a county is actually at risk.
The Controls
Unemployment, median income, and fire incident counts, so the estimate isn't just picking up a bad local economy or an actual fire rather than the disclosure law itself.
A modest, concentrated, and directionally consistent effect
Across every specification, the sign pointed the same direction: higher-hazard counties saw relatively lower home values after AB 38, exactly what the information asymmetry hypothesis predicts. Whether that effect cleared the bar for statistical significance depended on how tightly I defined "high hazard."
3.9%lower home values in California's most fire-exposed counties, the 15 counties above the 75th percentile of hazard exposure, relative to lower-hazard counties after the disclosure law took effect (p < 0.05).
The effect across every specification
Every specification points the same direction. The estimate sharpens from statistically insignificant to p < 0.05 once the sample is split at the 75th percentile of fire hazard exposure, evidence of a real threshold rather than a smooth, even effect.
Try the threshold yourself
At the median hazard split, the effect isn't statistically distinguishable from zero. Move to the 75th percentile, the 15 highest-hazard counties, and it sharpens into a real, detectable decline.
Pre-trends held, then the effect emerged
Year-by-year coefficients relative to 2020 (the reference year), from the event study specification. Pre-2021 coefficients hover without a systematic trend toward zero (joint test p = 0.15), then turn negative after AB 38 takes effect, reaching about −3.5% by 2023 and 2024.
Disclosure looks like it's doing its job, right where it matters most
The pattern across every specification suggests AB 38 is correcting a real information gap, and doing it most where the welfare stakes are highest: the counties buyers had the most reason to underestimate.
Where It's Working
The disclosure effect is concentrated among California's highest fire-hazard jurisdictions, not spread evenly, consistent with disclosure mattering most where buyers previously had the least visibility into risk.
Why the Estimate Is Conservative
AB 38's rollout overlapped with California's record 2020 fire season and pandemic-era remote-work migration into lower-density, higher-hazard areas, both of which push the estimated effect toward zero. The true property-level effect is likely larger than what county-level data can detect.
What Comes Next
Parcel-level transaction data matched to individual property hazard designations, or a regression discontinuity design at the hazard zone boundary, would identify the effect more cleanly than county aggregation allows.
Rigor means being honest about what your data can and can't tell you
This was the first time I'd carried a research question through the entire pipeline myself: pulling and merging five separate government and industry datasets, building a treatment variable from scratch, choosing a model, and defending it against the confounds that could undermine it. The hardest and most useful part wasn't running the regression, it was sitting with results that were directionally consistent but not always statistically clean, and reporting that honestly instead of overselling a headline number. AB 38's effective date landed right on top of California's worst fire season on record and a pandemic-era migration pattern that worked against my own hypothesis, and the paper had to say so plainly rather than bury it. That's the version of rigor I want to carry into policy work: state what the evidence shows, state what it doesn't, and let the honest answer be the interesting one.
"Grace goes well beyond what one might expect from an undergraduate study, carefully going through the necessary diagnostics to assess the plausibility of a causal interpretation of her results. The analysis is thorough and well-written."
— Faculty endorsement, Department of Economics