A Systematic Investigation of the effect of wildfire events and risks on property values

Wildfires frequency and severity have been increasing in the western United States over the past few decades. This rising threat is caused by the accumulated fuel load, climate change, and the rapid expansion of housing in the wildland-urban interface (WUI). Since most mitigation and suppression costs are borne by taxpayers, policy analysts seek both market (e.g., protection and suppression cost) and non-market cost estimates of wildfires. As one tool, the hedonic pricing method is commonly used to investigate wildfire effects on property values. There are a variety of hedonic studies investigating wildfire, with mixed and/or inconsistent results. Model estimates are further complicated by a variety of data availability issues as well as varied econometric modeling decisions made by analysts. This analysis applies spatial econometrics modeling strategies in a hedonic pricing model framework to examine the joint effect of both past fire occurrence and current risk on property values in Santa Fe County, New Mexico. The objective of this analysis is twofold. First, I systematically investigate wildfire effects on property values via the hedonic model using a variety of modeling approaches, including varying or alternative measures for property values, wildfire event and risk, and econometric modeling techniques. Secondly, using hedonic results as primary estimates, I then investigate how the effect of wildfire varies with data availability and econometric modeling techniques through internal meta-analysis. The systematic investigation can be grouped and classified as measures for property values, wildfire occurrence and risk (which capture data availability issues), or econometric modeling techniques (which capture subjective modeling decisions of the analyst). The systematic investigation includes: two dependent variables (estimated sale price and assessed property value); two measures for wildfire events (the nearest fire measure and the aggregate fire measure with 4 buffer zones), each with two time frames (7 year and 15 year); three risk measures (GIS-based composite hazard and risk assessment, WUI risk assessment and individual-level house risk assessment); two commonly-used hedonic functional forms (semi-log and double-log); and four spatial dependency approaches (independent, spatial lag, spatial error, general spatial model), with three weight matrix. Overall, variations in data and econometric specification produce 2,000 regression results for hedonic model. Summarizing the direction of wildfire estimates, I find that past wildfire events/occurrences have a negative effect on property value. Specifically, the marginal implicit price (MIP) for a one kilometer increase in distance from the nearest fire $3,461 (in 2013 dollars), implying an increase in assessed value of 1.1%. The MIP for one additional burn near the house is $14,375 (in 2013 dollars), implying a decrease in assessed value of 4.6%. Secondly, the effects of risk on property value vary by risk measure, risk level, and geographic area. For composite risk and WUI risk, wildfire risk increases property values below a certain risk level and the relationship tends to be negative or insignificant once risk reaches that threshold; for house level risk it reduces property value. The effects of wildfire risk also differ across Non-WUI and WUI. In the Non-WUI area, the positive effects of amenity dominate, and thus wildfire risks tend to increase property value. However in the WUI the negative effects of wildfire risk offset, or even exceed the positive effects of amenities, resulting in a non-significant or negative relationship. Further, meta-analysis reveals the following results. First, models that use assessed value data give higher R2 than models that use estimated sales price data. The assessed value models also lead to more significant estimates and larger MIP estimates. Secondly, ignoring spatial autocorrelation either leads to overestimate of MIP or has no significant effect on MIP estimates. Third, the measurements of wildfire risk significantly influence effects of past wildfire events on property value. This reveals the importance of joint estimation of both wildfire events and wildfire risks. Ignoring the effects of wildfire risk in hedonic models might result in inappropriate estimates. Overall, this analysis systematically investigates the effect of past fire occurrence and current risk on housing prices, using a variety of data measures and modeling techniques. Different from previous studies, which only present “the best fit model”, this analysis conducts 2,000 hedonic regressions on wildfire effects, and then examines how judgements and choices made by researchers affect wildfire effects on property values. This approach synthesizes results of hedonic models in a concise and structured way, but also improves the robustness and reliability of our results in ways that are useful for informing policy recommendations