Wildfires, Insurer Withdrawal, and the Expansion of FAIR Plans in California

Data Analytics Capstone Project

Author

Affiliation

Ann Brennan

SUNY Geneseo

1 Introduction

In recent years, wildfires have become more frequent and severe across the western United States, particularly in California. This has contributed to growing economic costs for households, insurers, and policymakers. These events have also begun to reshape the functioning of private insurance markets in California. As wildfire risk increases, insurers face rising claims and uncertainty, causing many to reassess their exposure in high-risk areas. This is visible in an increase of policy non-renewals. This means insurers choose not to continue coverage for existing customers. This shift raises concerns about access to insurance and the stability of housing markets in wildfire-prone areas.

Homeowners who are unable to obtain or maintain private coverage must turn to alternative options. The most prominent alternative is California’s FAIR (Fair Access to Insurance Requirements) Plan. The FAIR Plan is a state-mandated insurer of last resort designed to provide basic property insurance to high-risk properties that cannot acquire coverage in the private market. It serves as an important safety net, however, FAIR Plan coverage is typically more expensive and offers limited protection compared to standard policies. As a result, increased reliance on the FAIR Plan may signal instability in the insurance market and create financial burdens for homeowners.

This project examines the relationship between wildfire activity, insurer behavior, and the expansion of FAIR Plan coverage in California. The primary research question is: How does wildfire activity affect insurer non-renewals, and how do non-renewals drive FAIR Plan growth? This question aims to help understand how climate-related risks translate into market outcomes and policy challenges.

To answer this question, this study uses a county-year level dataset covering all 58 counties in California from 2015 to 2021. This dataset combines information from multiple data sources, including the California Department of Insurance (CDI), which provides data on insurer non-renewals and FAIR Plan policy counts. It also includes Cal Fire, which provides measures of wildfire activity such as acres burned, and the U.S. Census Bureau, which contributes socioeconomic controls including housing units, median home value, and median income. The unit of observation is county-year, allowing for analysis of changes within counties over time while accounting for differences across regions.

The main empirical methods will be a fixed effects panel regression and linear regression model. These models will be used to estimate relationships between the variables of interest. County fixed effects control for time-invariant characteristics such as geography, baseline fire risk, and existing economic conditions. Year fixed effects control for statewide shocks, including severe wildfire seasons or changes in state insurance requirements. By isolating variation within counties, this framework focuses on the effect of changing wildfire activity on insurer non-renewals and their subsequent impact on FAIR Plan enrollment. This analysis aims to provide more credible estimates of these relationships by concentrating on variation within counties over time.

The results of this analysis indicate that increased wildfire activity may be associated with higher insurer non-renewals. This suggests that insurers reduce their exposure in response to greater risk. Additionally, higher non-renewal rates are associated with increased FAIR Plan enrollment. This indicates that homeowners who lose their private coverage turn to the FAIR Plan as an insurer of last resort. Overall, the results suggest that climate-related risk is pushing homeowners away from private insurance and toward FAIR Plan coverage. However, there are significant trade-offs associated with this shift, as FAIR Plan policies are often more expensive with less coverage.

The findings of this project have important implications for multiple stakeholders. For the insurance industry, the results highlight the challenges of maximizing profits and managing risk with increasingly severe wildfire incidents. For policymakers, the expansion of FAIR Plan enrollment suggests that the program is functioning as intended for the short-term, however, this is not a long-term solution. The FAIR Plan program is intended as a temporary safety net, and may not be sustainable if homeowners continue to increase reliance on it. For housing markets, rising non-renewals and higher insurance costs may reduce property values, limit mortgage availability, and reduce housing availability in high-risk areas. Overall, this project contributes to the understanding of how environmental risk is associated with and translates to market changes and policy responses. It also provides evidence that can inform future decisions about insurance regulation, climate risk management, and housing policy in wildfire-prone regions.

2 Background and Related Literature

Wildfire risk has become an increasingly important issue for insurance markets in California. As wildfires have become more frequent and severe over time, the cost of insuring high-risk properties has increased. This impacts and creates challenges for insurers, policymakers, and homeowners. In response to increased wildfire risk, insurers may increase premiums or completely withdraw from high-risk areas. When homeowners lose or cannot obtain private insurance, they often rely on “insurers of last resort”, such as California’s FAIR Plan. This state mandated program provides basic coverage to high-risk properties to ensure homeowners can still meet mortgage and legal requirements.

Prior research shows that wildfire risk generates significant economic costs with implications for housing, land use, and risk management. Literature by Bayham et al. (2022) provides a comprehensive review of the economics of wildfires in the United States, documenting the scale of damage and incentive challenges in high-risk areas. At the same time, work by Boomhower, Fowlie, and Plantinga (2023), shows that insurance markets also shape risk mitigation behavior. A key finding is that insurer discounts for risk mitigation can help households learn about ways to reduce their own losses. This suggests that insurance influences how people respond to risk.

More recent work focuses on how climate risk is reshaping housing insurance markets. Keys and Mulder (2025) show that increased climate disaster risk is reflected in insurance and mortgage data, with insurers adjusting coverage and pricing in response to perceived risk and exposure in the long run. Similarly, Fowlie et al. (2025) highlight that climate change is affecting home insurance markets through rising premiums, coverage restrictions, and insurer withdrawal in high-risk areas. These findings support the idea that insurance markets may become less stable as climate risk increases.

California’s FAIR Plan is generally viewed as an important safety net, but also interpreted as a signal of instability in the private insurance market. Auer (2024) emphasizes that as wildfire risk increases, insurers have started to restrict coverage or exit high-risk areas, which has important implications for both policymakers and homeowners. Overall, this literature suggests that insurer behavior plays a key role in how wildfire risk translates to insurance access.

Important gaps still remain in the existing research. Most of the existing literature focuses on either insurance pricing or broad market outcomes, with less emphasis on specifically how non-renewals connect wildfire exposure to the expansion of insurers of last resort. Additionally, there is limited county-level evidence linking wildfire activity, insurer withdrawal, and FAIR Plan enrollment over time. This project contributes to existing literature by using a county-year panel dataset to examine the relationship between wildfire activity, insurer non-renewals, and FAIR Plan growth in California. This provides more evidence on the role of insurer behavior as a connection between climate risk and insurance market outcomes.

3 Data

3.1 Data Sources

This analysis combines data from three primary sources to create a county-level panel dataset with wildfire activity, insurance policy outcomes, and socioeconomic features in California. First, the data on insurer non-renewals and FAIR policy counts are obtained from the California Department of Insurance (CDI). The CDI collects this information to monitor insurer behavior. Their goal is to protect consumers and keep the insurance market stable. This data provides annual counts of non-renewed policies, as well as new and renewed FAIR Plan policies at the county level.

Second, wildfire activity data is collected from the California Department of Forestry and Fire Protection (Cal Fire). This dataset includes measures of wildfire severity in California, such as total acres burned per county and year. Cal Fire collects this data for fire management, risk assessment, and public reporting purposes. It is a widely used source for analysis on wildfires.

Third, socioeconomic characteristics are obtained from the U.S. Census Bureau, specifically the American Community Survey (ACS). The ACS provides annual estimates on features such as median household income, median home value, and total housing units. These variables are included to control for differences in economic conditions and housing markets across counties in California.

All three datasets were accessed online in March of 2026 and merged by county and year. The new dataset allows for a comprehensive analysis of how the relationships between wildfire risk, insurer behavior, and socioeconomic characteristics change over time.

3.2 Unit of Observation

The unit of observation in this project is the county-year. Each row in the dataset represents one California county observed in a given year. The dataset includes 58 counties observed over the period from 2015 through 2021. The final panel contains 438 observations.

3.3 Sample and Scope

The sample includes all 58 counties in California over the period from 2015 through 2021. This time frame captures multiple major wildfire seasons in order to analyze the relationship between wildfire risk and insurance market outcomes. The dataset includes three different variable categories: insurance market outcomes (non-renewals and FAIR Plan policies), wildfire activity (acres burned), and socioeconomic controls (income, home value, and housing units). The final dataset is structured as a county-year panel, where each observation corresponds to a specific county in a given year.

3.4 Data Cleaning

In order to construct the final dataset, it is necessary to clean and filter the data. First, wildfire data is spatially joined with California county boundaries using geographic information system (GIS) methods. Each wildfire perimeter is assigned to the counties it intersects in order to link fires with specific geographic areas. The fire-level data is then aggregated to the county-year level by summing the total acres burned and summing the number of fires.

Second, the wildfire dataset is filtered to include only the years 2014 through 2021. The year 2014 is included to construct a one year lag of wildfire activity, specifically acres burned lagged by one year. This approach links each year’s wildfire exposure to insurer behavior in the following year. The lag structure reflects the fact that insurers do not respond immediately. Instead, they typically adjust their policies in the subsequent plan year after observing wildfire losses from the prior year.

Third, missing values for acres burned and number of fires are replaced with zeros. Counties with zero wildfire activity are still included in the dataset, serving as a baseline group for the analysis.

Next, insurance data from the California Department of Insurance is cleaned and transformed. County identifiers are standardized, and character variables are converted to numeric format where necessary. FAIR Plan coverage is constructed by combining the number of new and renewed FAIR Plan policies into a single measure of total FAIR Plan policies. It is important to note that the number of non-renewals are insurer initiated, and exclude those initiated by consumers to exclusively reflect insurer behavior.

Additionally, socioeconomic control variables are obtained from the American Community Survey and reshaped into the county-year panel format. The control variables are merged with the insurance and wildfire datasets using consistent county-year identifiers.

Finally, a few key variables are transformed for analysis. Log transformations are applied to skewed variables including FAIR Plan policies, lagged acres burned, and non-renewals.

3.5 Key Variables

The primary outcome variables in this analysis aim to capture both insurer behavior and reliance on an insurer of last resort. The first outcome, insurer non-renewals, is measured as the number of policies not renewed by insurers in a given county-year. The second outcome, FAIR Plan policies, is the sum of new and renewed FAIR Plan policies. This represents a total count of participation in the FAIR Plan. The main explanatory variable is wildfire activity, which is measured as total acres burned in a county-year. A lagged version of this variable is used to reflect the timing of insurer responses. Insurer non-renewals are also included as an explanatory variable when the outcome is FAIR Plan policies. Control variables include median household income, median home value, and total housing units. These variables account for differences in economic conditions and housing markets across different counties.

3.6 Data Quality and Limitations

Although this dataset is constructed from credible administrative sources, it is important to note its limitations. First, the measure of wildfire activity is solely based on total acres burned, which may not fully represent perceived wildfire risk or expected future losses. For example, insurers may respond to other variables such as proximity to fires or anticipated risk rather than just realized damage. Additionally, spatially assigning fires to counties based on geographic overlap may cause measurement error, particularly for fires that spread to multiple counties. Second, the insurance data does not include information on premiums, underwriting criteria, or specific policy characteristics. Because of this, the analysis does not reflect pricing adjustments or changes in policy terms, which are notable insurer behaviors. Third, the sample size is limited to 58 counties over seven years, which is a relatively small panel dataset. The limited time span may reduce the statistical power and precision of estimates. Finally, although this analysis controls for year and county fixed effects, the results should be interpreted as identifying strong associations rather than definitive causal effects. Additional factors, such as policy changes or unobserved insurer strategies, may influence the relationships between variables.

3.7 Data Dictionary

fair_policies

Number of policies enrolled in the California FAIR Plan (county-year level).

acres_burned_lag1

Total acres burned by wildfires in the previous year (one-year lag).

voluntary_non_renewed_insurer_initiated

Number of insurance policies not renewed by insurers (excluding consumer-initiated cancellations).

median_income

Median household income in the county (in dollars). Controls for local economic conditions.

housing_units

Total number of housing units in the county. Controls for county size and amount of housing.

median_home_value

Median value of owner-occupied housing units (in dollars). Controls for housing market conditions.

log_fair_policies

Natural log of FAIR Plan policies. Used to reduce skewness.

log_acres_burned

Natural log of wildfire acres burned. Used to reduce skewness.

log_non_renewed

Natural log of insurer-initiated non-renewals. Used to reduce skewness.

3.8 Summary Statistics

df |>
  select(
    fair_policies,
    acres_burned_lag1,
    voluntary_non_renewed_insurer_initiated,
    median_income,
    housing_units,
    median_home_value
  ) |>
  skim()

Data summary

Name	select(…)
Number of rows	438
Number of columns	6
_______________________
Column type frequency:
numeric	6
________________________
Group variables	None

Variable type: numeric

skim_variable	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100	hist
fair_policies	37	0.92	3141.73	12804.26	6	92.00	362.00	1226.00	105875	▇▁▁▁▁
acres_burned_lag1	89	0.80	68212.43	184812.19	0	260.57	2715.21	38612.51	1338779	▇▁▁▁▁
voluntary_non_renewed_insurer_initiated	32	0.93	3335.85	6125.78	11	482.75	1304.00	3157.75	47150	▇▁▁▁▁
median_income	32	0.93	64893.00	20039.63	34974	50412.25	59425.50	75954.25	140258	▇▇▃▁▁
housing_units	32	0.93	242744.11	509451.21	1626	24066.75	78079.00	240604.50	3578801	▇▁▁▁▁
median_home_value	32	0.93	387687.93	221348.23	133300	231600.00	305650.00	479950.00	1225900	▇▃▂▁▁

4 Empirical Strategy / Methods

4.1 Empirical Approach

This analysis uses a combination of fixed effects panel regression and linear regression models to examine the relationship between wildfire activity, insurer non-renewals, and FAIR Plan enrollment. The primary goal is to estimate how changes in wildfire exposure influence insurer behavior, and in turn, how insurer withdrawal affects the reliance on California’s FAIR Plan. A fixed effects structure is suitable for this analysis because the dataset consists of observations from the same counties over time. Counties differ in many ways, such as geography, baseline wildfire risk, and housing market structure. These differences may influence wildfire exposure and insurance outcomes. The model’s county fixed effects control for these time invariant differences across counties. Year fixed effects are included to control for statewide shocks that would impact all counties in a given year, such as particularly severe wildfire seasons or changes in state insurance regulations. Additionally, a predictive modeling component is implemented using a train-test split to evaluate out of sample performance. This assesses how well the model can predict FAIR Plan enrollment and insurer non-renewals based on observed variables.

4.2 Model Specification

The primary empirical model estimates the relationship between wildfire activity, insurer non-renewals, and FAIR Plan enrollment using a two-way fixed effects panel regression. The outcome variable is the logarithm of FAIR Plan policies at the county-year level. The use of a logarithm reduces skewness while still capturing proportional changes in relationships in FAIR Plan enrollment.

\(\log({FAIR\_policies}_{it})\) \(=\) \(\beta_0\) \(+\) \(\beta_1 \log({acres\_burned}_{i,t-1})\) \(+\) \(\beta_2 \log({non\_renewals}_{it})\) \(+\) \(\beta_3 X_{it}\) \(+\) \(\alpha_i\) \(+\) \(\gamma_t\) \(+\) \(\varepsilon_{it}\)

In this model specification, \(i\) indexes counties and \(t\) indexes years. The outcome variable is the log of FAIR Plan policies in county \(i\) at time \(t\). The main explanatory variables are the log of acres burned in the previous year and the log of insurer initiated non-renewed policies. The vector \(X_{it}\) includes three control variables: median household income, housing units, and median home value. County fixed effects (\(\alpha_i\)) control for time invariant county characteristics. Year fixed effects (\(\gamma_t\)) control for statewide shocks impacting all counties in a given year. The error term is denoted by \(\varepsilon_{it}\).

Additionally, alternative specifications are estimated using different combinations of fixed effects, including models with only year fixed effects, only county fixed effects, and both. This helps evaluate how sensitive the results are to different sources of variation.

4.3 Additional Outcome: Insurer Non-Renewals

This analysis also examines insurer behavior directly by estimating models with insurer non-renewals as the outcome variable. In these specifications, lagged wildfire activity is the primary explanatory variable. This helps evaluate whether wildfire risk leads insurers to withdraw from high-risk markets. This two step approach first links wildfire activity to non-renewals and then links non-renewals to FAIR Plan growth, which provides a more complete picture of how environmental risk affects insurance market outcomes.

4.4 Prediction and Model Evaluation

In addition to the panel regression analysis, a predictive modeling technique is conducted using a train-test split. The dataset is divided into a training sample (2015-2019) and a testing sample (2020-2021). The linear regression model is estimated on the training data and then used to generate predictions for FAIR Plan enrollment in the testing time period. Model performance is evaluated using the root mean squared error (RMSE). RMSE measures the average magnitude of errors in the prediction. Lower RMSE values indicate better accuracy in the model’s predictions. This approach helps show how well the model performs on new data and whether the relationships found in the regression are useful for prediction.

4.5 Interpretation and Identification

The coefficients in the regression models are interpreted as correlations between variables rather than causal effects. For example, the coefficient for lagged acres burned represents the relationship between changes in wildfire activity and changes in FAIR Plan enrollment within a county, holding other factors constant. Including county and year fixed effects strengthens the analysis by controlling for unobserved differences across counties and state-wide trends over time. However, this analysis does not use an empirical model that supports causal claims. The results should be interpreted as showing correlational relationships, not causal effects.

5 Results

5.1 Figure 1: Wildfire Activity, Insurer Non-Renewals, and FAIR Plan Enrollment Over Time

ggplot(time_long, aes(x = year, y = value, color = variable)) +
  geom_line(size = 1.2) +
  geom_point(size = 2) +
  labs(
    title = "Wildfires, Non-Renewals, and FAIR Plan Growth Over Time",
    x = "Year",
    y = "Standardized Values",
    color = "Variable"
  ) +
  theme_minimal()

Figure 1 shows a clear pattern over time linking wildfire severity to insurance market responses. Wildfire activity shows variation over time, with a large spike in 2020. This reflects one of the most severe wildfire seasons for California in recent history. This spike is followed by noticeable increases in both insurer non-renewals and FAIR Plan enrollment during the following year.

A key pattern in this graph is the lag between wildfire exposure and insurer behavior. Wildfire activity peaks in 2020, but increases in non-renewals do not occur until the following year. This is consistent with the idea that insurers make decisions after observing losses. Through this lag structure, policy decisions are typically made annually.

At the same time, FAIR Plan enrollment shows a steady upward trend throughout the period from 2015 through 2021. This suggests that as private insurers reduce the number of policies in high-risk areas, homeowners increasingly rely on FAIR Plan policies as an alternative form of coverage. These patterns provide initial visual evidence of interconnected relationships between the key variables: as wildfire risk increases, insurers withdraw from the market, and FAIR Plan coverage increases.

5.2 Figure 2: Geographic Distribution of Wildfire Activity, Insurer Non-Renewals, and FAIR Plan Enrollment in California (2021)

ggplot(map_long) +
  geom_sf(aes(fill = value_scaled), color = "white", size = 0.2) +
  scale_fill_viridis_c(option = "plasma", na.value = "gray90") +
  facet_wrap(~variable) +
  labs(
    title = "Wildfires, Non-Renewals, and FAIR Plans (Standardized, 2021)",
    fill = "Standardized Value"
  ) +
  theme_minimal() +
  theme(
    panel.grid = element_blank(),
    axis.text = element_blank(),
    axis.ticks = element_blank(),
    strip.text = element_text(size = 6, face = "bold")
  )

Figure 2 shows the geographic distribution of wildfire activity, insurer non-renewals, and FAIR Plan enrollment across California counties in 2021. The maps use a color gradient, with darker colors (purple and dark orange) indicating higher values and lighter colors (yellow and light orange) indicating lower values.

Overall, the maps show a clear spatial pattern, where counties with higher wildfire activity also tend to have higher non-renewals and greater FAIR Plan enrollment. These patterns are concentrated in high-risk areas instead of being evenly distributed across the state. This alignment suggests that insurer withdrawal is concentrated in areas with greater wildfire exposure. FAIR Plan enrollment is also highest in the same counties, indicating that the FAIR Plan is being used where private coverage is reduced.

5.3 Primary Fixed Effects Regression Results

stargazer(model,
          type = "text",
          title = "Fixed Effects Regression Results",
          dep.var.labels = "Log(FAIR Policies)",
          covariate.labels = c("Log(Acres Burned, Lagged)",
                               "Log(Non-Renewals)",
                               "Median Income",
                               "Housing Units",
                               "Median Home Value"),
          omit = c("factor"),
          digits = 3)

Fixed Effects Regression Results
=====================================================
                              Dependent variable:    
                          ---------------------------
                              Log(FAIR Policies)     
-----------------------------------------------------
Log(Acres Burned, Lagged)            0.006           
                                    (0.009)          
                                                     
Log(Non-Renewals)                  1.260***          
                                    (0.111)          
                                                     
Median Income                      -0.00001          
                                   (0.00001)         
                                                     
Housing Units                     -0.00003***        
                                   (0.00000)         
                                                     
Median Home Value                  -0.00000          
                                   (0.00000)         
                                                     
Constant                           18.343***         
                                    (2.502)          
                                                     
-----------------------------------------------------
Observations                          401            
R2                                   0.958           
Adjusted R2                          0.950           
Residual Std. Error            0.436 (df = 332)      
F Statistic                111.633*** (df = 68; 332) 
=====================================================
Note:                     *p<0.1; **p<0.05; ***p<0.01

The primary fixed effects panel regression examines how wildfire activity and insurer non-renewals affect FAIR Plan enrollment at the county-year level. It controls for county and year fixed effects as well as socioeconomic variables.

The key result is a strong and statistically significant relationship between insurer non-renewals and FAIR Plan enrollment. A 1% increase in non-renewals is associated with approximately a 1.26% increase in FAIR Plan policies, holding all other factors constant. This suggests a strong substitution effect between private insurance and the FAIR Plan. On the other hand, wildfire activity (lagged acres burned) is not statistically significant once insurer behavior and the fixed effects are included. This suggests that wildfire exposure does not directly translate into FAIR Plan growth, but may affect it indirectly through insurer decisions.

5.4 Alternative Specification: Insurer Non-Renewals as Outcome Variable

stargazer(model2,
          type = "text",
          title = "Fixed Effects Regression Results",
          dep.var.labels = "Log(Non-Renewals)",
          covariate.labels = c("Log(Acres Burned, Lagged)",
                               "Log(FAIR Policies)",
                               "Median Income",
                               "Housing Units",
                               "Median Home Value"),
          omit = c("factor"),
          digits = 3)

Fixed Effects Regression Results
=====================================================
                              Dependent variable:    
                          ---------------------------
                               Log(Non-Renewals)     
-----------------------------------------------------
Log(Acres Burned, Lagged)          -0.009**          
                                    (0.004)          
                                                     
Log(FAIR Policies)                  0.00000          
                                   (0.00001)         
                                                     
Median Income                       0.00000          
                                   (0.00000)         
                                                     
Housing Units                      -0.00000*         
                                   (0.00000)         
                                                     
Median Home Value                  7.348***          
                                    (1.168)          
                                                     
-----------------------------------------------------
Observations                          406            
R2                                   0.984           
Adjusted R2                          0.980           
Residual Std. Error            0.216 (df = 338)      
F Statistic                302.484*** (df = 67; 338) 
=====================================================
Note:                     *p<0.1; **p<0.05; ***p<0.01

Additionally, insurer non-renewals are used as the outcome variable to examine whether wildfire activity is directly associated with insurer withdrawal behavior. FAIR Plan enrollment is not included in this model.

In this model, lagged wildfire activity is statistically significant, but the effect is very small. The coefficient of -0.009 suggests that a 1% increase in lagged acres burned is associated with a 0.009% decrease in insurer non-renewals, holding other factors constant. Although statistically significant, the magnitude of this effect is quite small. When combined with earlier results showing weak and insignificant effects of wildfire activity, this suggests that wildfire exposure does not have a strong direct relationship with insurance outcomes. Overall, this shows that wildfires mainly affect the insurance market through how insurers respond, rather than a direct impact on insurance outcomes.

5.5 Predictive Model Results - Linear Regression

etable(model_year_fe, model_county_fe, model_both_fe)

                                                    model_year_fe
Dependent Var.:                                log(fair_policies)
                                                                 
log(acres_burned_lag1)                            0.0012 (0.0250)
log(voluntary_non_renewed_insurer_initiated)   0.7902*** (0.0536)
median_income                                -2.71e-5** (9.43e-6)
housing_units                                6.56e-7*** (1.56e-7)
median_home_value                              2.21e-6* (8.61e-7)
Fixed-Effects:                               --------------------
year                                                          Yes
county                                                         No
________________________________________     ____________________
S.E. type                                                     IID
Observations                                                  315
R2                                                        0.68590
Within R2                                                 0.64863

                                                   model_county_fe
Dependent Var.:                                 log(fair_policies)
                                                                  
log(acres_burned_lag1)                            -0.0016 (0.0138)
log(voluntary_non_renewed_insurer_initiated)     1.231*** (0.1379)
median_income                                 8.47e-5*** (1.36e-5)
housing_units                                -4.23e-5*** (6.54e-6)
median_home_value                                1.32e-6 (1.84e-6)
Fixed-Effects:                               ---------------------
year                                                            No
county                                                         Yes
________________________________________     _____________________
S.E. type                                                      IID
Observations                                                   315
R2                                                         0.93250
Within R2                                                  0.63438

                                                     model_both_fe
Dependent Var.:                                 log(fair_policies)
                                                                  
log(acres_burned_lag1)                             0.0041 (0.0119)
log(voluntary_non_renewed_insurer_initiated)     1.083*** (0.1169)
median_income                                   -3.28e-6 (1.24e-5)
housing_units                                -3.66e-5*** (4.88e-6)
median_home_value                                -1.9e-6 (1.41e-6)
Fixed-Effects:                               ---------------------
year                                                           Yes
county                                                         Yes
________________________________________     _____________________
S.E. type                                                      IID
Observations                                                   315
R2                                                         0.96332
Within R2                                                  0.39284
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

rmse <- sqrt(mean((actual - preds)^2, na.rm = TRUE))

rmse

[1] 1.002992

A linear regression model is also used to assess how well the variables explain FAIR Plan enrollment. Three baseline models are estimated using only year fixed effects, only county fixed effects, and both fixed effects. Insurer non-renewals are consistently the strongest predictor of FAIR Plan enrollment across all three models.

Wildfire activity is not statistically significant in any of the models. This reinforces the earlier results showing a weak direct relationship between wildfire exposure and FAIR Plan enrollment once fixed effects are included. The model fit improves when county fixed effects are added, indicating that differences across counties are important in explaining the model results.

The model is then evaluated using a train-test split (2015-2019 for training and 2020-2021 for testing). The resulting root mean squared error (RMSE) is 1.003. This indicates moderate predictive accuracy. Overall, the predictive model results reinforce the main findings from the fixed effects regression. Non-renewals are directly correlated with FAIR Plan enrollment, while wildfire activity shows a much weaker direct relationship.

6 Discussion and Implications

The results suggest that insurer non-renewals are the primary driver of FAIR Plan enrollment. Wildfire risk may be an underlying factor, however, its direct effect is weak or inconsistent once county and year fixed effects are included. The strongest and most consistent relationship is between non-renewals and FAIR Plan growth. This suggests that the impacts of wildfires are translated indirectly through insurer behavior. Instead of responding immediately to wildfire activity, insurers may adjust their coverage based on broader risk factors, which leads to increased non-renewals. Households that lose private coverage then turn to the FAIR Plan. The predictive results also support this interpretation, as non-renewals are the strongest predictor across all models.

These findings have important implications for insurance markets and policymakers. They suggest that changes in insurance availability are driven more by insurer responses than by wildfire activity alone. Because of this, policies focused only on reducing wildfire risk may not be sufficient to stabilize the insurance market. The results also emphasize the FAIR Plan’s growth as an insurer of last resort. As non-renewals increase, more households rely on the FAIR Plan, which is typically more expensive with limited coverage. This raises concerns about long term sustainability and access to affordable insurance, especially in high-risk areas. Overall, these findings suggest that insurer withdrawal may reflect future instability in the insurance market, not just current wildfire exposure. This emphasizes the need for policies that consider both risk levels and the factors influencing insurer decisions.

A key recommendation is that policy efforts should focus on reducing insurer non-renewals and stabilizing the private insurance market. One approach is for state regulations to require insurers to maintain coverage in high-risk areas. These regulations could help prevent sharp increases in non-renewals and reduce the number of households forced to enroll in the FAIR Plan. However, this policy change comes with significant tradeoffs. Requiring insurers to remain in high-risk areas may increase their losses, which could lead to increased premiums and reduced profitability. This approach should be used in addition to other measures, such as risk-based pricing and risk mitigation incentives, in order to stabilize the insurance market in the long run. At the same time, policymakers should prepare for continued FAIR Plan growth by potentially expanding coverage to meet demand. However, reliance on the FAIR Plan should still be limited, as it is not a substitute for a stable private insurance market.

7 Conclusion

This project examines how wildfire activity affects insurer non-renewals and how non-renewals, in turn, drive FAIR Plan growth in California. The main research question is: How does wildfire activity affect insurer non-renewals, and how do non-renewals drive FAIR Plan growth? This question is important because increasing wildfire risk has raised concerns about the stability of private insurance markets and the reliance on the FAIR Plan as an insurer of last resort for homeowners in high-risk areas. To answer this, the analysis uses a county-year panel dataset covering all 58 counties in California from 2015 to 2021. The dataset combines wildfire data from Cal Fire, insurance data from the California Department of Insurance, and socioeconomic controls from the U.S. Census Bureau. The empirical approach uses fixed effects panel regressions with county and year fixed effects, as well as linear regression with a train-test split to evaluate predictive performance. The main result is that insurer non-renewals are the strongest and most consistent predictor of FAIR Plan enrollment, while wildfire activity has weak or statistically insignificant direct effects once fixed effects are taken into account. These findings suggest that wildfire risk affects insurance markets primarily through insurer behavior rather than directly translating to insurance outcomes. Insurer withdrawal is the main link between wildfire risk and increased reliance on the FAIR Plan.

Overall, these results suggest that changes in insurance availability are driven by insurer responses rather than wildfire activity alone. This emphasizes the importance of policies that address insurance market participation, not just environmental risk. State efforts that reduce non-renewals or support insurer participation in high-risk areas may be effective in stabilizing the insurance market in addition to mitigating wildfire exposure. Future research could improve on this analysis by incorporating more specific data, such as zip-code or property level insurance outcomes. It could also use causal identification strategies to more clearly estimate the effect of wildfire exposure. In addition, future research could examine the role of pricing changes and regulatory policy on insurer behavior, as these factors were not directly observed in this dataset.

8 References

Auer, M. R. (2024). Wildfire risk and insurance: Research directions for policy scientists. Policy Sciences, 57(2), 459–484.

Bayham, J., et al. (2022). The economics of wildfire in the United States. Annual Review of Resource Economics, 14, 379–401.

Bergé, L. (2023). fixest: Fast fixed-effects estimations (R package).

Boomhower, J., Fowlie, M., & Plantinga, A. J. (2023). Wildfire insurance, information, and self-protection. AEA Papers and Proceedings, 113, 310–315.

California Department of Forestry and Fire Protection (CAL FIRE). (n.d.). Fire hazard severity zones. Retrieved March 2026, from https://www.fire.ca.gov/

California Department of Insurance. (n.d.). Data and analysis on wildfires and insurance. Retrieved March 2026, from https://www.insurance.ca.gov/

California FAIR Plan Association. (n.d.). Key statistics & data. Retrieved March 2026, from https://www.cfpnet.com/

Fowlie, M., et al. (2025). How is climate change impacting home insurance markets? Brookings Institution.

Garnier, S. (2023). viridis: Default color maps from ‘matplotlib’ (R package).

Hlavac, M. (2022). stargazer: Well-formatted regression and summary statistics tables (R package).

Keys, B. (2025). Housing, climate risk, and insurance. NBER Working Paper.

Keys, B., & Mulder, P. (2025). Property insurance and disaster risk: New evidence from mortgage escrow data. NBER Working Paper.

OpenAI. (2026). ChatGPT (GPT-5.5) [Large language model]. Retrieved May 13, 2026, from https://chat.openai.com/

Pebesma, E. (2018). Simple features for R: Standardized support for spatial vector data. The R Journal, 10(1), 439–446.

Rudis, B. (2020). hrbrthemes: Additional themes, theme components, and utilities for ‘ggplot2’ (R package).

State of California. (n.d.). CA geographic boundaries.

U.S. Census Bureau. (n.d.). American Community Survey. Retrieved March 2026, from https://www.census.gov/

Vincent Arel-Bundock. (2023). modelsummary: Data and model summaries in R (R package).

Walker, K. (2023). tigris: Load census TIGER/Line shapefiles (R package).

Waring, E., Quinn, M., McNamara, A., Arino de la Rubia, E., Zhu, H., & Ellis, S. (2023). skimr: Compact and flexible summaries of data (R package).

Wickham, H. (2016). ggplot2: Elegant graphics for data analysis. Springer-Verlag New York.

Wickham, H. (2023). tidyr: Tidy messy data (R package).

Wickham, H., et al. (2019). Welcome to the tidyverse. Journal of Open Source Software, 4(43), 1686.

Wickham, H., François, R., Henry, L., & Müller, K. (2023). dplyr: A grammar of data manipulation (R package).