Backup Power: Public Implications of Private Substitutes for Electric Grid Reliability
Abstract
Private substitutes for electric grid reliability are common. We study their adoption and distributional implications. We first show that US households buy substitutes in response to a perceived decrease in grid reliability and that higher-income households are more likely to adopt them. We then develop a theoretical model of public provision of grid reliability in the presence of private substitutes that is consistent with these facts. The existence of substitutes increases aggregate welfare and reduces the efficient level of reliability spending. Using a calibrated version of the model, we find that, even though only a few households adopt batteries, most nonadopting households benefit from their availability. Battery adoption reduces utilities’ reliability spending, resulting in lower electricity bills for all customers. Most nonadopting households value these bill savings more than the reduced grid reliability.
Many households purchase private substitutes to improve the reliability of their electricity supply. For example, 14% of US households own a backup generator. Home storage batteries have also become increasingly mainstream as technology has matured and prices have fallen. Advice columns like the New York Times’s Wirecutter now publish guidance on purchasing backup battery systems that help households avoid the costs of power outages (Heffernan 2022).
While power outages impose costs on households, it is also expensive to prevent them. Power outages reduce productivity, destroy goods in cold storage, and can even lead to death—especially for people who warm themselves in unsafe ways or are unable to avoid high temperatures (Barreca et al. 2016; Trevizo et al. 2021). Yet, reducing or eliminating power outages requires expensive investments, resulting in larger utility bills for households. Many US households already struggle to pay their energy bills (Jessel et al. 2019; Barreca et al. 2022; Doremus et al. 2022). Climate change may also increase the cost of providing reliability through more frequent extreme weather, such as wildfires in California and extreme cold in Texas.
Replacing publicly provided reliability with privately provided reliability raises important distributional questions. An unreliable electric grid increases the benefits from purchasing private substitutes. As more households buy substitutes and decrease their dependence on the grid, the grid will not need to be as reliable. In turn, even more households will buy substitutes. This positive feedback loop raises the concern that households without substitutes will be left behind. Which households benefit from private substitutes depends on who is purchasing them and how responsive these purchases are to grid reliability.
This study contributes to answering these questions through two major avenues. First, we document who purchases substitutes and empirically estimate how household purchases respond to major electricity outages. These private substitutes affect the efficient provision of public reliability by reducing the damage from outages. We next develop and calibrate a model of public reliability provision and household behavior and use it to consider the distributional consequences of private substitutes. To our knowledge, this is one of the first studies demonstrating the causal relationship between grid reliability and private reliability investments. We are also the first to investigate the interplay between public and private reliability investments.
To understand who is likely to benefit from private substitutes for grid reliability, we first describe who is buying them and how they are increasingly common. Nationally, 14.3% of households in 2020 had a backup generator, an increase of 2.5 percentage points from 2015. Higher-income households are more likely to own a backup generator, as are households in rural areas. Households with incomes over $100,000 are nearly twice as likely to have a backup generator as households with incomes below this threshold. We find broadly consistent patterns for battery purchases.
Next, we show that a major power outage leads to a significant increase in battery purchases in the following year. We estimate this effect by combining data on battery adoption with data on power outages for the state of California. Our city-level data span 2017–21, a period where California enacted power shutoffs to reduce the risk of wildfires. We estimate two-way fixed effects, event study models of quarterly battery adoption in response to major outages. The impact of the outage on battery adoption starts the next quarter and peaks two quarters after the outage before returning to zero at five quarters. Our point estimates imply that a major outage leads to a 0.8 standard deviation increase in the rate of battery purchases two quarters later. We view these estimates as conservative because households in the control group may also respond to news coverage about outages by purchasing batteries. As long as a power outage causes households to perceive a future outage as being more likely, this finding implies that households buy private substitutes in response to a perceived decrease in electricity grid reliability.
Motivated by these empirical facts, we consider how the efficient level of grid reliability responds to changes in the availability of private substitutes; we view this as informative for how public utility commissions (PUCs) are likely to adjust spending on reliability. Utilities propose investments in grid reliability to PUCs that decide which investments to approve. These commissions are tasked with ensuring reliable electricity service, while also keeping electricity rates reasonable.1 They are often explicitly required to evaluate the costs and benefits of proposed investments. For example, the Florida Public Service Commission is required by law to consider “the estimated costs and benefits to the utility and its customers of making the improvements proposed in the plan” when deciding whether to approve utilities’ plans for storm hardening (Florida Senate 2019). Due to data limitations, we are unable to test empirically how PUCs respond to private investments in reliability.
We develop a theoretical model to describe centralized decisions over grid reliability and how they interface with household decisions to purchase substitutes. In our model, a benevolent planner sets the level of spending on grid reliability at the efficient level for electricity customers. These customers differ in their potential loss from power outages and, hence, in their willingness to pay for grid reliability. Substitutes like generators or backup batteries reduce the loss from an outage. Using the model, we compare two scenarios: one in which a private substitute is available and one in which it is not.
The model has three main results. First, when households purchase private substitutes, the efficient level of grid reliability is reduced. Second, it is not possible for a central planner to improve upon private decisions over whether to buy substitutes. Because households with substitutes continue to contribute to grid reliability, they do not impose costs on others. Third, under many circumstances, some households will be hurt by the existence of substitutes. It is possible for all households to benefit from the existence of substitutes, but it requires that, at a minimum, the average household purchases a substitute.
The households with the lowest willingness to pay (WTP) for reliability, and hence the lowest income households, are not typically harmed by private substitutes. Instead, when households are hurt, it is a group with higher WTP who preferred paying for a higher level of grid reliability. Households with both lower and higher WTP than these households benefit from the substitute. The higher WTP group benefits because they purchase a substitute and enjoy the lower potential loss from outages that it provides. The lower WTP group benefits because they value the electricity bill savings more than the lost reliability.
Using a calibrated model, we find that the overall welfare benefit of battery backup systems to households in California is $45 million per year. This includes a benefit of $125 million to households in the bottom 80% of the WTP distribution, an $18 million benefit to households in the top 0.3% of the WTP distribution, and a $98 million loss to those in between. More effective or cheaper substitutes increase the magnitudes of each of these welfare effects, reflecting larger shifts in the efficient level of grid reliability.
This study relates to the literature on the economic costs of power outages. Much of the economics literature focuses on developing countries where outages are more common.2 It finds that outages lead to small decreases in firm productivity in India and China (Fisher-Vanden et al. 2015; Alcott et al. 2016), and large decreases in sales for firms in sub-Saharan Africa (Cole et al. 2018). Because fewer outages are a challenge for quantifying these costs empirically, the economics literature on outages in developed countries has focused on stated preference methodologies (Carlsson and Martinsson 2007; Blass et al. 2010) or the related question of how electricity market design affects the reliability of electricity supply (Wolak 2021; Elliot 2022; Borenstein et al. 2023). A separate engineering-oriented literature has developed tools to help utilities estimate the costs to customers of outages to evaluate potential investments in grid reliability (see, e.g., Sullivan et al. 2018). These analyses often quantify the value of lost load, a single number that captures the benefits of enhanced reliability (Gorman 2022).
We contribute to this literature in two ways. First, we provide revealed preference evidence on the economic costs of power outages in the United States. We find that power outages increased spending on home storage batteries by over $20 million over a four-year period in California. Two contemporaneous papers also use a revealed preference approach to quantify household willingness to pay to avoid outages. Harris (2022) uses national generator purchases and outage data to find that US households are willing to pay $1.57/kilowatt-hour (kWh) to avoid outages. Similarly, Brown and Muehlenbachs (2023) use variation caused by California power shutoffs and battery purchase data to quantify a revealed preference measure of the value of lost load. Second, our theoretical model highlights how these battery purchases themselves affect households’ willingness to pay for grid reliability. We show that investments in grid reliability have distributional impacts, and these impacts are affected by the availability of private substitutes.
This study also contributes to the literature on the distributional impacts of new energy technologies. These new technologies are often heavily subsidized, so a salient concern is which households are adopting them. Previous work has focused on electric vehicles (Borenstein and Davis 2016; Holland et al. 2019) and distributed generation (Eid et al. 2014; Borenstein 2017; Brown and Sappington 2017). Like Brown (2022), we describe which households adopt home storage batteries. We also investigate how nonadopters are affected via public reliability provision.
Finally, our work is part of a broader literature examining publicly provided goods with private alternatives. Public education is the most prominent example (Glomm et al. 2011). In finding that high and low WTP households benefit while those in the middle do not, this study is similar to Epple and Romano (1996a, 1996b). They find that the availability of private schools causes households at the top and bottom of the WTP distribution to have similar preferences over school spending. However, their result arises in the context of a voting model, while ours has a regulator implementing the efficient level of spending.
1. Background
Electricity grids are complicated networks that require electricity supply to exactly equal electricity demand at every second. Failure to do so can result in blackouts (too little supply) or damage to the infrastructure (too much supply).
Electric grid reliability generally means that this careful balance is maintained with a high level of confidence. Planned or unplanned outages mean that electricity is not available to the end-user. These can be especially costly when electricity is a critical input—for example, at military bases, hospitals, or for residential users who use electricity to power oxygen pumps.
The US electricity grid is increasingly unreliable (Blunt 2022), and this trend is expected to continue as climate change increases the prevalence of extreme weather. Already, wildfires in California (2019), hurricanes on the Gulf Coast (2020), extreme heat in California (2020), and extreme cold in Texas (2021) have left millions of people without power (Canon 2019; Roth 2020; Schwartz 2020; Blunt and Gold 2021). Further, the transition to a grid based on intermittent renewable energy makes it more challenging to provide reliable electricity (Potter 2022).
1.1. Public Spending on Grid Reliability
There are many types of spending that increase reliability. Many outages are caused by weather downing power distribution lines. Utilities that own these lines can undertake regular tree trimming around lines or hire more staff so that downed lines are repaired more quickly. Moving distribution lines underground is another option, though a costly one. Outages can also be caused by inadequate capacity, that is, there is not enough generation supply to meet demand during the highest demand hours. This type of outage is expected to become more common as the grid becomes more dependent on renewable energy (Potter 2022). Investment in additional generation reserves can make this type of outage less likely. Similarly, investment in the transmission system can increase reliability by smoothing regional supply and demand shocks.
Regulated utilities are the entities that typically perform this work, but many expenditures require regulatory approval from a public utility commission.3 Most expenditures that improve reliability are capital investments. Utilities propose these investments, and the regulator decides whether to approve them. For approved investments, utilities are allowed to pass the cost on to consumers in the form of higher rates. Regulated utilities earn a profit on capital investment, so they have an incentive to propose more investment than the efficient level (Averch and Johnson 1962).4
Most PUCs make decisions about reliability spending in the context of their mission, which is to facilitate high-quality service and reasonable prices. Something akin to the efficient level of reliability spending is the goal of this process, and private substitutes including generators and backup batteries may affect these decisions. By choosing what projects to approve, the regulator influences the level of reliability of the electrical grid and is tasked with approving an amount of investment that balances the costs and benefits to consumers. Specifically, most states task the regulator with “assuring that utilities provide reasonable, adequate and efficient service to customers at just and reasonable prices,” and holding utilities to a “resource adequacy” standard, most commonly that blackouts due to insufficient supply occur only once per 10 years. The prices that the regulator approves must allow the utility to recover its costs, including a fair rate of return on its capital (US Environmental Protection Agency 2010).
For reliability spending decisions, the expected payoff is in terms of avoided interruptions of electrical services. Regulators use a variety of methods to assess the benefits of these avoided interruptions relative to the costs. While some regulators do not attempt to monetize these benefits, others do so by using estimates of the cost to consumers of outages (LaCommare et al. 2017; Zamuda et al. 2019). These estimates reflect the willingness to pay of customers to avoid outages as well as lost economic activity during outages. Since private substitutes make customers less reliant on their utility company, they can affect these estimates and thus feed back into the policymaking process.
However, the efficient level of reliability spending need not be the outcome of these policy processes. Regulatory capture or asymmetric information could result in too much investment in reliability. Regulators overly focused on the short run or the time inconsistency problem studied in Lim and Yurukoglu (2018) could result in too little investment. Yet, given the regulator’s stated objective, developments that shift the efficient level of investment will also likely shift the equilibrium level of investment.
Ensuring high levels of grid reliability is expensive. For example, consider capital spending on transmission and distribution, much of which is aimed at maintaining reliability. Deloitte (2016) reports that, for 47 investor-owned utilities representing 89% of the total market capitalization of US utilities, these expenditures were $42 billion in 2015.
All electricity consumers share the costs of reliability spending, with low-income households paying nearly as much as higher income households. Because most expenditures to improve reliability are fixed costs, the efficient two-part tariff would recover them entirely via the fixed charge. In this case, all households would share the costs equally regardless of income.
In reality, utilities in the United States recover fixed costs mostly with volumetric charges (Borenstein and Bushnell 2022). Low-income households use less electricity on average, but there is substantial overlap in the electricity use distributions for low- and higher-income households (Doremus et al. 2022). Even California’s aggressive use of increasing block volumetric charges for cost-recovery only results in modest wealth redistribution and creates substantial deadweight loss relative to the size of the transfers (Borenstein 2012).
Bill assistance programs like LIHEAP at the federal level or California’s CARE program could mean that vulnerable households pay little to none of the cost of reliability spending. Yet, only a small fraction of households receive this assistance. As of 2020, only 4.6% of US households report ever receiving any energy assistance, a category that includes both bill assistance and help fixing broken equipment (US Energy Information Administration 2022).
1.2. Private Substitutes for Grid Reliability
We now discuss two private substitutes for grid reliability: backup generators and batteries.5 Backup generators typically cost between $2,000 and $20,000 dollars and provide small-scale, at-home generation in the event of an outage. They usually use fossil fuels like diesel, natural gas, or propane. They are less efficient than traditional power plants but can reliably operate in the event of an outage. While backup generators are an established technology, ownership has been growing in response to climate change, and 2021 was a record year for generator sales (Phillips 2021).
Batteries like Tesla’s Powerwall are in the same broad cost range as generators. They allow households to store power for use during a power outage. If a household is not connected to the grid, the battery can store electricity production from solar cells for use overnight. Recently, blackouts in the western United States (due, in part, to faulty transmission lines) have caused demand for behind-the-meter batteries to surge (Hering and Copley 2021).
2. Data
The empirical analysis uses three main data sources: survey data from the US Energy Information Administration, battery purchase data from the State of California, and power outage data from Bluefire Studios.
2.1. Backup Generator Ownership Data
We use data from the Residential Energy Consumption Survey (RECS) to describe which households own backup generators (US Energy Information Administration 2022). The RECS is a nationally representative survey conducted every four to six years by the US Energy Information Administration. A novel feature of the RECS is that it collects data on respondents’ electricity billing and use directly from their utilities. The American Housing Survey (AHS) also collects data on backup generators in some survey waves, and patterns of adoption in the AHS are similar to those in the RECS (see appendix sec. C.2) (US Census Bureau 2021).
2.2. Battery Purchase Data
We use data from the State of California to describe which households purchase batteries and estimate how these purchases respond to power outages. These data come from a subsidy program, the Self-Generation Incentive Program (SGIP).6 The SGIP dates from 2001 and was created as a peak-load reduction program in response to the California electricity crisis. Starting in 2014, most program funds were allocated to subsidizing batteries (Center for Sustainable Energy 2021). The subsidy for battery purchase is large, over 30% of total costs on average.7 Thus, we expect these data to cover nearly the universe of battery purchases in periods when SGIP funds were available. We focus on residential battery purchases, which increased rapidly after 2017 (see fig. A1). While it was not necessary to have solar panels to receive the subsidy, nearly all the batteries in our sample (97%) were paired with photovoltaic systems.
During our study period, the program introduced higher subsidies for some customers, and our analysis excludes batteries that received a higher subsidy because of a power shutoff. The change to the program increased subsidies to vulnerable households in fire-prone areas. Households are classified as vulnerable if they are low-income, have medical needs, or rely on an electric pump for drinking water. These vulnerable households can qualify for a higher subsidy if they either (i) reside in a Tier 2 or Tier 3 high fire threat district or (ii) have experienced two or more Public Safety Power Shutoff events (California Public Utilities Commission 2020). To ensure that our results are not driven by a change in the effective price of batteries after power shutoffs, we exclude battery purchases that received this higher subsidy because of a power shutoff. Some of these households would likely have purchased batteries without the higher subsidy, so our estimated effect of outages on battery purchase is conservative.
This change to the program occurred right after significant power shutoffs, and we do two robustness checks to ensure that our results are not due to the timing of this change. The SGIP was changed in late 2019 and began accepting applications for these higher subsidies in May 2020 (California Public Utilities Commission 2019, n.d.–b). Many of the outages in our sample occurred in the fourth quarter of 2019. We first reestimate the model dropping all batteries that received this higher subsidy, as opposed to only those that received it because of a power shutoff. We find a smaller, but still statistically significant, impact of a major outage on battery purchase (fig. A2). As an additional check, we limit our analysis to fire-prone areas that were all treated with the same change to the subsidy.8 This restriction drops 88% of battery purchases. We find larger effects of an outage on battery purchase for this sample, but the coefficients are less precisely estimated and not statistically significant (fig. A3).
The battery analysis also relies on data on the following zip-code characteristics: median household income, number of housing units, and whether the zip code is located in a rural county. We use data on ZCTA-level median income and number of housing units from the US Census’s 2019 American Community Survey. Our definition of rural is based on the 2003 Rural-Urban Continuum codes, and we define all nonmetro areas as rural. We use a data set from the University of Michigan’s Population Studies Center that has matched these county-level codes to ZCTAs (University of Michigan Population Studies Center 2022). Finally, we use a crosswalk between zip codes and cities from
2.3. Outage Data
We also use city-level outage data from Bluefire Studios (Bluefire Studios 2021). These data are collected from utility outage management systems and begin in late 2017 when Bluefire began to track large California utilities. The data end in October 2021. For each city and for each hour, we observe the number of customer-hours of outage.9 We first drop observations without city information or from unspecified unincorporated areas (about 6% of observations). The resulting dataset covers 49 months and expands in coverage over time: it contains roughly 1,050 cities in October 2017 and 1,500 cities by October 2021. Most of the cities that were added over this period were very small.10
3. Motivating Empirical Facts
This section presents two empirical analyses that motivate the theoretical model. We first describe the types of households that buy private substitutes for grid reliability. We then show that major power outages spur adoption.
3.1. Wealthier Households Are More Likely to Purchase Private Substitutes
3.1.1. Backup Generators
More than one in eight US households own a backup generator, and this share has been increasing over time. In 2020, 14.3% of households had a backup generator, an increase of 2.5 percentage points from 2015.11 These national numbers conceal important regional heterogeneity; 18% of households in the Northeast own one, compared to only 10% of households in the West.12
Table 1 shows that wealthier households, those living in rural areas, and those with higher electricity consumption are more likely to own backup generators.13 Column 4 shows that a $10,000 increase in annual income is associated with a 0.4 percentage point increase in the probability that a household owns a backup generator.14 Similarly, increasing annual household electricity consumption by 1 megawatt-hour (MWh) (0.14 SD) is associated with a 0.4 percentage point increase in the probability of owning a backup generator. Living in a rural area is associated with a 15 percentage point higher probability of owning a generator. These results are similar whether we include fewer covariates (col. 3), more covariates (col. 4), or only look at single-family detached houses (col. 5).
| (1) | (2) | (3) | (4) | (5) | |
|---|---|---|---|---|---|
| Household income ($10,000/year) | .007*** | .005*** | .004*** | .004*** | |
| (.001) | (.001) | (.001) | (.001) | ||
| Electricity consumption (MWh/year) | .006*** | .006*** | .004*** | .005*** | |
| (.001) | (.001) | (.001) | (.001) | ||
| Average electricity price (cents/kWh) | .002 | .000 | −.001 | ||
| (.002) | (.002) | (.002) | |||
| Rural | .149*** | .159*** | |||
| (.015) | (.017) | ||||
| Census region fixed effects | No | No | No | Yes | Yes |
| Structure type fixed effects | No | No | No | Yes | No |
| Mean generator ownership | .129 | .129 | .129 | .129 | .151 |
| R-squared | .009 | .017 | .022 | .074 | .068 |
| Observations | 4,828 | 4,828 | 4,828 | 4,828 | 3,752 |
3.1.2. Batteries
We find that households in higher-income areas are more likely to adopt batteries, but rural households are not. Table 2 shows that the correlation between zip-code median household income and new batteries per household is 0.25. This correlation is also evident in the spatial distribution of batteries and income within urban areas; figure A4 shows this pattern for Los Angeles and San Francisco. Conditional on income, adoption is uncorrelated with living in a rural county (table 2, col. 3). Why are rural households more likely to adopt generators, but not batteries? One explanation is that rural areas have historically had less reliable power and residents may already have owned generators by the time batteries became a feasible alternative. The vast majority of batteries are also paired with rooftop solar panels, and solar panels may be less appealing to households in rural areas.
These patterns have also changed over time, likely due to the change to California’s battery subsidy program discussed in section 2. In the first quarter of 2020, the SGIP program began offering much higher subsidies to vulnerable households in areas with high fire risk. Having a low income is one way a household can qualify as vulnerable, and columns 4 and 5 show that the correlation between adoption and zip-code income falls after this change. Yet this correlation is still substantial at 0.18. Rural households, which are disproportionately located in fire-prone areas, are also more likely to adopt batteries after this change to the subsidy program.
3.2. Households Purchase Private Substitutes in Response to a Perceived Decrease in Grid Reliability
We next present evidence that households buy private substitutes in response to a perceived decrease in grid reliability. We assume that experiencing a major power outage makes households believe the grid will be less reliable in the future. Given this assumption, it is sufficient to show that households buy more private substitutes in response to a major power outage. We focus on purchases of batteries because we can link them to outage data at a fine geographic scale. We estimate the dynamic treatment effect of experiencing a major power outage on city-level purchases of residential batteries.
Our model of battery purchases for city i in quarter t is the following:
We define a major outage as a quarter where power is out for at least 3% of customer-hours (as defined in sec. 2). In our sample, 0.9% of city-quarters are classified as having a major outage. Because we use customer-hours for our definition, we do not differentiate between major outages in which all customers lost power for nearly three days within a quarter and those where a smaller share of customers experienced very long outages. Much of the identifying variation comes from wildfire-induced power shutoffs in the fall of 2019. The fourth quarter of 2019 accounts for 54% of the city-quarters with a major outage and for 79% of the city-quarters with a major outage for the sample of cities with only one major outage.
We estimate this model using the interaction-weighted estimator proposed in Sun and Abraham (2021). Because this estimator is designed for an absorbing treatment, we exclude the 7.7% of cities with multiple major outages over the sample. Of the remaining cities, 13.4% experience a major outage. Results are similar if we instead use the two-way fixed effects estimator with leads and lags.
We find that a major outage leads to an increase in battery purchases in the following quarters. Figure 1 plots the estimated βk coefficients and 95% confidence intervals. We do not find an impact on purchases during the quarter that a major outage occurs. There is a positive but statistically insignificant effect in the subsequent two quarters, and we find statistically and economically significant increases in purchases three and four quarters after the outage. The point estimate for the second quarter (9.0) implies that a major outage leads to a 0.8 standard deviation increase in the rate of battery purchase two quarters later. Overall, the estimated quarterly effects are noisy. However, the estimated total effect in the year following the outage is 19.2 more batteries per 10,000 households, which is significantly different from zero. This effect is the same as a city moving from the median to the 91st percentile of annual batteries purchased per 10,000 households. Impact of outages on battery purchases. Estimated coefficients and 95% confidence intervals from model (1). The city-quarters from 646 cities in 2017–21. Sample excludes the 7.7% of cities with more than one major outage. A major outage is defined as one where over 3% of customer-hours are out for the quarter; 0.9% of city-quarters have a major outage. The mean of the dependent variable (batteries per 10,000 housing units) is 2.0, with a standard deviation of 8.4. SE clustered by city. HH = household.
Results are robust to including more lags, keeping cities with multiple outages, and using alternative definitions of a major outage. Figure A5 presents results from four alternative specifications. First, we include more leads and lags in the model. We still find no effect on purchases outside the four quarters following the outage. Second, we keep cities with multiple major outages. For this specification, we define all coefficients relative to the first outage. We find that the estimated effects are similar, though slightly smaller. Third, we reestimate the model using a less stringent definition of a major outage: more than 1% of customer-hours without power. We find a similar pattern but slightly smaller point estimates; we estimate that the total effect on battery purchases in the year following the outage is 15.5 batteries per 10,000 households. Finally, we use a more stringent definition of a major outage: more than 5% of customer-hours without power. The estimated coefficients are similar but slightly smaller than those for our main specification.
We also find similar results in an alternative analysis that uses more granular data on power shutoffs. Appendix B presents results that use an outage measure based on circuit-level data on Public Safety Power Shutoffs. This analysis is at the zip-code level and only for the area of California served by Pacific Gas and Electric (PG&E). Our estimates imply that a major outage caused by power shutoffs leads to 27.1 more batteries purchased per 10,000 households in the following year, and this effect is statistically different from zero, with a .
Of note, because of the possibility of spillovers to the control group, we view these estimates as a lower bound. One possible mechanism for spillovers is that hearing about nearby areas losing power affects the beliefs of households in other cities. Indeed, major power outages can receive considerable media attention. We expect this attention to cause untreated households’ perceptions of grid reliability to decrease and thus make them more likely to adopt a substitute. As a result, our point estimates will be smaller than if the “untreated” households did not have any information about their neighbors.
It is also unlikely that our estimates are simply capturing a shift forward in the timing of purchases. Panel A of figure A5 shows that the estimated effects return to zero after the outage, rather than below zero. Thus, at least in the medium run, these outage-induced purchases appear to be new and additional.
Our estimates suggest that the outages in our sample increased battery spending by over $20 million. We find that a major outage leads to 1,915 more batteries per million households. Over our four-year sample, households experienced 0.61 million major outages, and the mean cost of the battery systems purchased was $21,525 (in real $2022). Thus, the implied outage-induced spending on batteries—by both households and the state of California—was 25.1 million dollars. This number captures only a fraction of the response to outages since batteries are not the only private substitute available: figure A6 shows that the time series of Google searches for a common home backup generator (Generac) mirrors that of searches for a common battery (Tesla Powerwall). Overall, our results imply that outages can lead to economically significant increases in spending on private substitutes.
4. Theoretical Model
We next develop a model to describe centralized decisions over grid reliability and individual decisions over purchasing a private substitute. The planner in our model approximates a regulator that sets the level of reliability by choosing how much spending to approve. The model allows us to compare how the efficient level of grid reliability changes as private substitutes are introduced. We compare the welfare of households in these two scenarios to examine the distributional consequences of private substitutes. We then calibrate the model to features of backup batteries in California to describe the magnitude of these effects.
4.1. Primitives
There is some probability that households will experience a grid power outage. This probability, π(R), is a function of spending on grid reliability, R.15 We assume that this function is twice continuously differentiable and for any R we have that , , and .
There is a set of households of size one who differ in the degree of their loss in the event of a power outage. We denote the size of the loss for household i as Li. This parameter can be thought of as the household’s willingness to pay (WTP) to eliminate the risk of an outage. Thus, it is a reduced form parameter that may capture both differences in underlying preferences and differences in economic resources. We denote the probability density function of the distribution of Li as 𝓁, the cumulative distribution function by ℒ, and its mean by . We assume that all households have a positive WTP.
Households can abate their potential loss by purchasing a private substitute. We assume that the purchase of a private substitute (e.g., a generator or backup battery) is a binary choice. Purchasing a private substitute costs Pλ, where λ represents the effectiveness of the substitute and P represents its price per unit of effectiveness. A household that purchases a substitute would experience a loss of in the event of a power outage, where λ exceeds zero and is less than one. With no substitute it experiences its full loss, Li. We assume that even if everyone purchased a substitute it would still be efficient to have some spending on grid reliability and hence .
4.2. Policymaker’s Problem
We consider the spending decision on grid reliability, R, as chosen by a benevolent planner.16 The planner balances the benefits and costs of providing reliability and chooses the level that maximizes utilitarian social welfare. The planner knows the distribution of losses, 𝓁, but cannot observe an individual customer’s loss Li. As a result, the costs of grid reliability R are shared equally by all customers. This reflects situations in which the fixed costs of the utility are recovered via a flat fee to all households.
We consider this problem for two cases. In the first, households rely exclusively on the grid and cannot purchase private substitutes. In this case, the planner solves the problem
It is possible for either no households or all households to purchase substitutes depending on the substitute’s price per unit effectiveness. Here, we focus on the case where some, but not all, households purchase substitutes. Households with will purchase substitutes. Therefore, will satisfy
It is not possible for the planner to improve upon the private decisions about whether to purchase a substitute. Because households who purchase a substitute continue to pay for the grid, purchasing a substitute does not impose costs on other households, which would create a fiscal externality. As a result, were the planner to simultaneously choose both who purchased a substitute and the level of grid reliability, they would make the same choices as the households do (this is proven in appendix sec. D.2).19
Before discussing the welfare consequences of substitutes, it is worth noting that this model applies only to substitutes that do not fully replace the electrical grid. We assume that households that purchase substitutes remain connected to the grid. This is realistic, as very few households currently go “off grid.”20 However, as substitute technologies continue to develop, households leaving the grid may become more common. This would affect our model in two important ways. First, households with substitutes would no longer care at all about grid reliability. Second, they would no longer contribute to the cost of grid reliability. Hence, our setting is similar to publicly provided goods funded via taxation, like public schooling, where all households contribute even if they choose not to use the public services.
Some outages, like Public Safety Power Shutoffs, are active choices motivated by limiting the risk of wildfires, a factor that is not directly in our model, but accounting for this would not alter the model’s implications. Excluding wildfire risk from the model implicitly assumes that the existence of substitutes would not alter these decisions about when to shut off power to avoid wildfires. However, if households care about wildfires, then private substitutes would lead the planner to want to reduce wildfire risk in addition to electric bills. If all households care equally about wildfire risk, then the main results of the model are unchanged. When substitutes are available, the grid will be less reliable. Utility bills would be higher than our model indicates, but households would be compensated for this by decreased wildfire risk.
We also note that our static model abstracts from possible dynamic responses to changes in available substitutes or grid reliability. Grid reliability is a durable good that depreciates, and policy decisions determine investment in it, not its overall level. A dynamic model may produce interesting transitions between the steady states that are represented by instant responses in our static model.21
4.3. Welfare Consequences
To analyze the welfare consequences of private substitutes to households with different potential losses, we compare two settings: with and without substitutes. If reliability spending were fixed at , the availability of substitutes would be Pareto improving. However, because the efficiency-targeting planner will adjust reliability spending as a result of substitutes’ existence, some households may be harmed.
In principle, there are many possible outcomes. Even if all households do not purchase substitutes, all households can benefit from their existence. One common, if surprising, result is that households with both high and low potential losses can benefit while those in the middle are hurt. Alternatively, it is possible for only households with potential losses below a cutoff to benefit, or for only households with potential losses above a cutoff to benefit. The specific outcome depends on the curvature of the probability of an outage function, π, the distribution of potential losses, 𝓁, and the characteristics of the substitute, λ and P. However, we show one important general result. Appendix section D.3 shows that if the household with the average potential loss does not buy a substitute, then there are households that would be better off if substitutes were not available. While this condition is sufficient for households to be hurt by substitutes, it is not necessary. The average household purchasing a substitute does not preclude households from being hurt.
4.4. Calibration
The previous section shows the range of possible welfare effects of private substitutes but is not informative as to the actual welfare effects of the substitutes that are and may become available to electricity customers. In this section, we fill this gap. To fully specify our model would require knowing the full distribution of willingness to pay to eliminate power outages, the risk of power outages that would result from any given level of reliability spending, and the price and effectiveness of the available substitute. It is not possible to know all of these factors with any degree of confidence, but we calibrate our model to reasonable specifications. We do so in the context of backup batteries in California.
We take the WTP distribution from Sullivan et al. (2018). This distribution is based on surveys of residential customers in the western United States. Utilities asked these customers how much they would be willing to pay to avoid a one-hour outage. We fit a lognormal distribution to the percentiles of this empirical distribution. The resulting lognormal distribution has μ and σ of 0.77 and 1.72, respectively. The surveys are from no later than 2015, prior to the wide availability of backup batteries, so we treat it as the WTP distribution without batteries.
We approximate the probability of outage as a function of reliability spending π(R) with a second-order Taylor polynomial around the current level of spending. We take spending on grid reliability to be $776 per year per household, the estimate of the average household contribution to residual costs among customers of the three largest California investor-owned utilities from Borenstein et al. (2022), in 2022 dollars. To make it comparable to the WTP distribution, we convert it to an hourly rate. We take the probability of outage from the California outage data for the year 2020 (the last full calendar year in our data). It is the empirical probability that at any given time a household in California is facing a power outage. This is approximately 0.0007. Since we assume that the planner is setting reliability spending optimally, this probability determines the first derivative of the π function at this level of spending. The curvature of the function is unknown, so we show results for multiple possibilities.
We consider the welfare effects from the availability of a residential backup battery, like those sold by Tesla. These cost approximately $12,000 and are expected to last for 20 years. We illustrate the effects if owning a battery reduces the WTP to eliminate outages by 50% and show how the results would differ under other assumptions.
Figure 2 shows the resulting welfare effects for households with different WTP to eliminate outages. The horizontal axis shows WTP transformed into percentiles of the distribution. Because WTP is lognormally distributed, mean WTP () is greater than median WTP. If reliability spending is fixed at , individual welfare benefits are given by the blue line. In this case, substitutes are Pareto improving but only benefit households that purchase them. The planner can increase total welfare by adjusting reliability spending. The orange and green lines represent the welfare gains or losses if the planner efficiently adjusts reliability spending under two different assumptions about the curvature of the function mapping reliability spending to the outage probability, π.22 In both cases, the average WTP household does not purchase a substitute, and some households are hurt by the decrease in grid reliability. This includes some households that choose to purchase a battery. Since the grid is less reliable after the social planner adjusts reliability spending, more households purchase batteries than with fixed reliability spending. Calibrated individual welfare gain from batteries. The blue dotted line describes the individual welfare benefit to households with different potential losses (X-axis) from the existence of batteries when reliability spending does not adjust. The solid orange and dashed green lines represent potential welfare benefits with different curvatures of the π function, which result in different . In these calibrations, most households, including those at the top and bottom of the WTP distribution, benefit from the substitute, but some households in between do not.
Under efficient readjustment, low- and high-WTP households are better off when backup batteries are available. High-WTP households are spending more money in total but receiving increased reliability. On the other hand, low-WTP households value the saving on their electricity bills more than the lost grid reliability. Some of those in the middle are hurt; they were previously close to their ideal reliability level but now are buying too little reliability (if they do not buy a battery) or too much reliability (if they do buy a battery in response to decreased grid-level reliability).
In our preferred calibration, described by the green line in figure 2, batteries provide a total net surplus of $45 million per year to households in California. This includes the households in the bottom 79.5% of the WTP distribution who benefit, the next 20.3% of households with higher WTP who are hurt, and the remaining 0.25% with the highest WTP who benefit. The total benefit to the bottom 79.5% of households in the WTP distribution is $125 million per year. On average these households receive a surplus of $11.87 per year. The next 20% of households in the WTP distribution are hurt by the substitute and on average would each be willing to pay $36.66 per year to eliminate the substitutes. The lost surplus to this group totals $98.2 million per year. The top 0.25% of households in the WTP distribution benefit from the substitute and on average receive a surplus of $551 per year, totaling $18 million.
Figure 3 shows how the welfare implications differ for substitutes with different characteristics. Panels A and C describe the effects of a substitute with different values for the substitute’s effectiveness λ. Panels B and D describe the effects of a substitute with different prices per unit effectiveness. In all cases, the group of households with very high WTP who benefit is extremely small. The more effective the substitute is the more it will benefit this group. Households can only choose between buying or not buying the substitute, so effectiveness can be thought of as the “quantity” of substitute that is available to purchase. Thus, when more is available, the high WTP group will purchase it and benefit. Effectiveness has a small effect on the fraction of households who buy substitutes because it only affects purchasing decisions through the decline in grid reliability. A low price per effectiveness benefits the high WTP group less because more households will purchase the substitute and thus grid reliability will decline more, while the amount of loss mitigated for those already purchasing the substitute remains the same. The magnitude of the welfare benefit to the low WTP group and the harm to the middle WTP group are both larger with more effective or cheaper substitutes. Effect of substitute price and effectiveness. Panels A and C display outcomes from substitutes with varying effectiveness while panels B and D vary their price. Panels A and B display the resulting welfare effects while panels C and D show the fractions of people who benefit and are hurt by the existence of batteries. The fraction of people at the top who benefit is at most 0.45% and hence not visible. The π function and 𝓁 distribution follow the calibration in section 4.4 with .
We expect most low-income households to be in the low-WTP group that benefits from the availability of substitutes. Fewer resources results in a lower WTP for all goods, and we see empirically that low-income households are less likely to purchase substitutes.23 We model grid reliability spending as being paid through a flat tax on all households. As discussed in section 1.1, it may be more accurate to model these charges as varying with electricity use. Yet, because household income is not highly correlated with use, this alternative model would yield similar conclusions. An alternative funding structure like paying for grid investments with income tax revenue, as discussed in Borenstein et al. (2022), would alter the policy preferences of all households and make it unlikely that low-income households benefit from private substitutes.
We model private substitutes and do not differentiate between generators and batteries. While the two are similar in many ways, they differ in that home storage batteries may provide power back to the grid during periods of high demand. Thus, private adoption of batteries could also lower the cost of reliability. This feature would not change our main conclusion, that most households benefit from substitute adoption, but it could result in fewer (if any) households being worse off due to the availability of private substitutes.24
5. Discussion and Conclusion
Ownership of private substitutes can both affect and be affected by policy decisions over grid reliability. In section 3.2, we show that power outages cause households to purchase batteries, that is, households purchase substitutes in response to the perceived unreliability of the electrical grid. Had large investments been made to prevent power outages, fewer batteries would have been purchased. More widespread ownership of private substitutes then has the potential to affect policy decisions as PUCs seek to balance meeting the changed needs of their constituents with the impact of investment on electricity rates. When more households own batteries, the marginal benefit from spending on reliability is lower and hence the efficient level of reliability is lower. The missions of PUCs suggest that changes in the efficient level of provision should be reflected in the actual level of spending. However, the extent of this response is an outstanding and important question. We model the welfare effects associated with the efficient level to understand the potential effects of private substitute ownership on all households.
The availability of substitutes and the resulting decrease in spending on grid reliability affects households differently, depending on their valuation for reliability. Households who do not value reliability highly benefit from cost savings on their electric bill. The reduced reliability hurts other households who were served well by the prior level of grid reliability. A small group of households who value reliability very highly purchase substitutes and are better off despite the lower level of grid reliability.
Empirically, we see that high-income households are more likely to purchase private substitutes. This holds true both for backup generators, where we observe these purchasing patterns in the RECS, and for batteries, where we observe these patterns in the SGIP data. This positive correlation between income and willingness to pay for reliability suggests that low-income households are likely to benefit from private substitutes even if they do not buy them.
Three structural factors are likely to increase adoption of private substitutes. First, the price of new substitute technologies like home storage batteries is likely to fall as the technology develops. At the same time, more extreme weather may make the grid less reliable, at least absent larger investments than were previously necessary. Finally, spurred by policies to reduce carbon emissions, more households are expected to use electricity for heating and transportation, potentially increasing their need for reliability. As long as the households buying substitutes continue to use and pay for the grid, those that cannot afford these substitutes are unlikely to be left behind. Yet not all households will benefit, and substitutes’ distributional effects are likely to increase.
Notes
Paul A. Brehm is in the Department of Economics and Environmental Studies Program, Oberlin College ([email protected]). Sarah Johnston is in the Department of Agricultural and Applied Economics, University of Wisconsin–Madison ([email protected]). Ross Milton is at La Follette School of Public Affairs, University of Wisconsin–Madison ([email protected]). We are grateful for financial support from Oberlin College. We thank Maggie Brehm, Dylan Brewer, Shana Cui, Lucas Davis, Erik Johnson, Corina Mommaerts, Edson Severnini, Anna Spurlock, Christopher Sullivan, anonymous referees, and seminar participants at University of Wisconsin–Madison for valuable feedback. We also thank Sam Fechner, Sharan Ganjam Seshachallam, Hannah Keidan, Yifei Liu, and Anna Slebonick for excellent research assistance.
1. For example, the California Public Utilities Commission works to “ensur[e] the provision of safe, reliable utility service and infrastructure at reasonable rates” (California Public Utilities Commission, n.d.–a). Florida’s Public Service Commission’s mission statement begins, “To facilitate the efficient provision of safe and reliable utility services at fair prices” (Florida Public Service Commission, n.d.). Pennsylvania’s Public Utilities Commission works to “ensure safe and reliable utility service at reasonable rates” (Pennsylvania Public Utilities Commission, n.d.).
2. The incentives for public provision of reliability also differ. For example, Jha et al. (2023) find that one cause of blackouts in India is utilities buying less electricity when wholesale prices are high. Regulations prevent utilities in developed countries from not meeting electricity demand simply because the price is high.
3. In the United States, the regulator is a state-level commission with three to seven members. Commissioners are political appointees in about three-quarters of states and elected in the rest (Howe 2019).
4. Some reliability expenditures, such as regular tree trimming, fall under operations and maintenance. Utilities have more discretion over this spending, which they do not profit from.
5. Appendix C.3 discusses microgrids, another substitute that is also increasingly popular. While data are sparser and incomplete, it appears that adoption is highest for wealthier, coastal states. The data also suggest that the number of microgrids in the United States grew rapidly from 2017 to 2020.
6. These data are available at https://www.selfgenca.com/report/public. We use the January 12, 2023, version for our analysis.
7. The 2021 incentive was $150–$200/kWh for residential customers. Higher incentives of $850–$1,000/kWh were available for low-income households and those with critical resilience needs (State of California and the Self-Generation Incentive Program 2023).
8. Specifically, we keep battery purchases from zip-code tabulation areas (ZCTAs; the areas corresponding to zip codes) with at least 99% of their area located in Tier 2 or Tier 3 high fire threat districts.
9. While the total number of customers is reported, this statistic is sometimes incorrectly based on the maximum number of customers that have experienced an outage up to that point in their dataset (confirmed via email on April 25, 2023). We instead proxy for the number of customers with the number of housing units in the city. We construct this number by aggregating up ZCTA-level data from the American Community Survey. For each quarter, we construct total customer hours (used to construct the fraction of customer-hours out) by multiplying the number of housing units by the number of hours that are covered by the outage data in that quarter.
10. Bluefire reports that cities that are present for all 49 months average 8,436 customers in October 2021, while cities that were present for less than 45 months average 612 customers in October 2021. A “customer” generally covers more than one person; it can refer to, e.g., a business, a house, or an apartment building. There are occasional missing months for some cities.
11. Authors’ calculation using the household-level Residential Energy Consumption Survey (US Energy Information Administration 2022). Residents of apartment buildings with more than four units were not asked about generator ownership in 2015, so this statistic is calculated for a sample that excludes residents of apartment buildings with more than four units. Of these respondents, 15.4% owned generators in 2020, up from 12.9% in 2015.
12. For more detail, please see fig. C1, which summarizes data from the 2020 RECS.
13. This table uses data from the 2015 RECS because electricity consumption data are not yet available for the 2020 RECS. Table C1 presents regression results using the available 2020 RECS data; available results are similar to those presented here.
14. We construct household income using the midpoint of $20,000 income bins. Incomes above $140,000 are top-coded into one bin, and we use an income of $150,000 for this group. The standard deviation is $44,500.
15. As discussed in sec. 1.1, this spending could take a range of forms, including investments in transmission reliability and generation adequacy or staffing to make repairs more quickly.
16. The planner in the model most closely corresponds to the state regulator of utilities. While it does not directly determine grid reliability, the regulator can influence it by approving utility investments and setting required levels of resource adequacy. Section 1.1 discusses the goals of the regulator and its ability to influence grid reliability.
17. Note that in cases where the WTP distribution is bounded, will be a solution in any case where exceeds its upper bound and as a result no one buys a substitute when reliability spending is set at . However, there can be another local maximum in which households purchase the substitute and grid reliability is lower.
18. From the implicit function theorem
19. Households that buy home storage batteries usually also install rooftop solar panels. Under net metering, installing these panels can impose costs on other households, but this effect is not due to the substitute.
20. Gorman et al. (2020) show that disconnecting from the grid in favor of self-generation and storage is unlikely to benefit consumers under current utility practices.
21. For example, if we begin in a steady state where private substitutes are prohibitively expensive and then the price decreases, some households will purchase private substitutes. This will induce the policymaker to lower the investment level in the next period. As reliability depreciates, more people will purchase private substitutes, and hence the policymaker will allow further depreciation, until we reach the new steady state.
22. The orange and green lines reflect of −40 and −70 respectively. When this ratio is larger in magnitude, the function’s curvature is greater, and is closer to .
23. We note that there are alternative measures of the marginal benefit of reliability. For households for which outages can result in severe health consequences, the marginal benefit as measured by these expected health costs could be higher than the revealed preference WTP, i.e., WTP could be constrained by ability to pay. Policies like SGIP’s higher subsidies for households with medical needs directly target these households.
24. Households with batteries may also be compensated by the utility for providing this backup power. This compensation would make batteries more affordable for all households but is unlikely to change the relationship between WTP for reliability and adoption that is central to our model.
References
Alcott, Hunt, Allan Collard-Wexler, and Stephen D. O’Connell. 2016. How do electricity shortages affect industry? Evidence from India. American Economic Review 106 (3): 587–624. Averch, Harvey, and Leland L. Johnson. 1962. Behavior of the firm under regulatory constraint. American Economic Review 52 (5): 1052–69. Barreca, Alan, Karen Clay, Olivier Deschenes, Michael Greenstone, and Joseph S. Shapiro. 2016. Adapting to climate change: The remarkable decline in the US temperature-mortality relationship over the twentieth century. Journal of Political Economy 124 (1): 105–59. Barreca, Alan, R. Jisung Park, and Paul Stainier. 2022. High temperatures and electricity disconnections for low-income homes in California. Nature Energy 7:1052–64. Blass, Asher A., Saul Lach, and Charles F. Manski. 2010. Using elicited choice probabilities to estimate random utility models: Preferences for electricity reliability. International Economic Review 51 (2): 421–40. Bluefire Studios. 2021. Historical data. https://poweroutage.us/products .Blunt, Katherine. 2022. America’s power grid is increasingly unreliable. Wall Street Journal, February 18. Blunt, Katherine, and Russell Gold. 2021. The Texas freeze: Why the power grid failed. Wall Street Journal, February 19. Borenstein, Severin. 2012. The redistributional impact of nonlinear electricity pricing. American Economic Journal: Economic Policy 4 (3): 56–90. ———. 2017. Private net benefits of residential solar PV: The role of electricity tariffs, tax incentives, and rebates. Journal of the Association of Environmental and Resource Economists 4 (S1): S85–S122. Borenstein, Severin, and James Bushnell. 2022. Do two electricity pricing wrongs make a right? Cost recovery, externalities, and efficiency. American Economic Journal: Economic Policy 14 (4): 80–110. Borenstein, Severin, James Bushnell, and Erin Mansur. 2023. The economics of electricity reliability. Journal of Economic Perspectives 37 (4): 181–206. Borenstein, Severin, and Lucas W. Davis. 2016. The distributional effects of US clean energy tax credits. Tax Policy and the Economy 30 (1): 191–234. Borenstein, Severin, Meredith Fowlie, and James Sallee. 2022. Paying for electricity in California: How residential rate design impacts equity and electrification. Next 10, San Francisco. Brown, David P. 2022. Socioeconomic and demographic disparities in residential battery storage adoption: Evidence from California. Energy Policy 164:112877. Brown, David P, and Lucija Muehlenbachs. 2023. The value of electricity reliability: Evidence from battery adoption. Working paper. Brown, David P, and David E. M. Sappington. 2017. Designing compensation for distributed solar generation: Is net metering ever optimal? Energy Journal 38 (3): 1–32. California Public Utilities Commission. n.d.–a. Consumer affairs branch. https://www.cpuc.ca.gov/about-cpuc/divisions/news-and-public-information-office/consumer-affairs-branch (accessed December 30, 2023).———. 2019. Decision 19-09-027: Decision establishing a self-generation incentive program equity resiliency budget, modifying existing equity budget incentives, approving carry-over of accumulated unspent funds, and approving $10 million to support the San Joaquin Valley disadvantaged community pilot projects. https://docs.cpuc.ca.gov/PublishedDocs/Published/G000/M313/K975/313975481.PDF .———. 2020. Attachment A: SGIP equity resiliency eligibility matrix—residential customers, version 3. https://www.cpuc.ca.gov/-/media/cpuc-website/files/uploadedfiles/cpucwebsite/content/news_room/newsupdates/2020/attachment-a-sgip-equity-resiliency-eligibility-matrix-for-residential-customers-version-3.pdf .———. n.d.–b. Self-generation incentive program (SGIP). https://www.cpuc.ca.gov/industries-and-topics/electrical-energy/demand-side-management/self-generation-incentive-program (accessed March 15, 2023).Canon, Gabrielle. 2019. California launches investigation into public safety power shutoffs by PG&E, other utilities. USA Today, November 13. Carlsson, Fredrik, and Peter Martinsson. 2007. Willingness to pay among Swedish households to avoid power outages: A random parameter Tobit model approach. Energy Journal 28 (1): 75–89. Center for Sustainable Energy. 2021. SGIP background. https://sites.energycenter.org/sgip/background .Cole, Matthew A., Robert J. R. Elliott, Giovanni Occhiali, and Eric Strobl. 2018. Power outages and firm performance in sub-Saharan Africa. Journal of Development Economics 134:150–59. Deloitte. 2016. From growth to modernization. https://www2.deloitte.com/content/dam/Deloitte/us/Documents/energy-resources/us-er-from-growth-to-modernization.pdf .Doremus, Jaqueline, Irene Jacqz, and Sarah Johnston. 2022. Sweating the energy bill: Extreme weather, poor households, and the energy spending gap. Journal of Environmental Economics and Management 112:102609. Eid, Cherrelle, Javier Reneses Guillén, Pablo Fréas Marín, and Rudi Hakvoort. 2014. The economic effect of electricity net-metering with solar PV: Consequences for network cost recovery, cross subsidies and policy objectives. Energy Policy 75 (C): 244–54. Elliot, Jonathan. 2022. Investment, emissions, and reliability in electricity markets. Working paper. Epple, Dennis, and Richard E. Romano. 1996a. Ends against the middle: Determining public service provision when there are private alternatives. Journal of Public Economics 62 (3): 297–325. ———. 1996b. Public provision of private goods. Journal of Political Economy 104 (1): 57–84. Fisher-Vanden, Karen, Erin Mansur, and Qiong (Juliana) Wang. 2015. Electricity shortages and firm productivity: Evidence from China’s industrial firms. Journal of Development Economics 114:172–88. Florida Public Service Commission. n.d. Mission statement and goals. https://www.psc.state.fl.us/about#MissionAndGoals (accessed December 30, 2023).Florida Senate. 2019. Florida state statute 366.96. https://www.flsenate.gov/laws/statutes/2019/366.96 .Glomm, Gerhard, Bala Ravikumar, and Ioana C. Schiopu. 2011. The political economy of education funding. In Handbook of the economics of education, vol. 4:615–80. Amsterdam: Elsevier. Gorman, Will. 2022. The quest to quantify the value of lost load: A critical review of the economics of power outages. Electricity Journal 35:107187. Gorman, Will, Stephen Jarvis, and Duncan Callaway. 2020. Should I stay or should I go? The importance of electricity rate design for household defection from the power grid. Applied Energy 262:114494. Harris, Robert I. 2022. Willingness to pay for electricity reliability: Evidence from U.S. generator sales. Working paper. Heffernan, Tim. 2022. How to pick a solar panel and battery backup system. New York Times, December 12. Hering, Garrett, and Michael Copley. 2021. Western US blackouts fuel unmet demand for behind-the-meter batteries. S&P Global Capital IQ, August 10. Holland, Stephen P., Erin T. Mansur, Nicholas Z. Muller, and Andrew J. Yates. 2019. Distributional effects of air pollution from electric vehicle adoption. Journal of the Association of Environmental and Resource Economists 6 (S1): S65–S94. Howe, Douglas J. 2019. Governance models of public utility commissions in the United States. Competition and Regulation in Network Industries 20 (3): 229–39. Jessel, Sonal, Samantha Sawyer, and Diana Hernández. 2019. Energy, poverty, and health in climate change: A comprehensive review of an emerging literature. Frontiers in Public Health 7:357. Jha, Akshaya, Louis Preonas, and Fiona Burlig. 2023. Blackouts: The role of India’s wholesale electricity market. NBER Working paper 29610, National Bureau of Economic Research, Cambridge, MA. LaCommare, Kristina, Peter Larsen, and Joseph Eto. 2017. Evaluating proposed investments in power system reliability and resilience: Preliminary results from interviews with public utility commission staff. Report, Energy Technologies Area, Berkeley Lab Lim, Claire S. H., and Ali Yurukoglu. 2018. Dynamic natural monopoly regulation: Time inconsistency, moral hazard, and political environments. Journal of Political Economy 126 (1): 263–312. Pennsylvania Public Utilities Commission. n.d. Pennsylvania public utilities commission: About us. https://www.puc.pa.gov/about-the-puc/ (accessed December 30, 2023).Phillips, Matt. 2021. Climate change calls for backup power, and one company cashes in. New York Times, September 15 Potter, Ellie. 2022. “Grim” FERC reliability outlook sees policy failures, extreme weather as threats. S&P Capital IQ, May 19. Roth, Sammy. 2020. California blackouts are public utilities commission’s fault, grid operator says. Los Angeles Times, August 17. Schwartz, Matthew S. 2020. Hundreds of thousands without power after Hurricane Delta sweeps through South. National Public Radio, October 10. State of California and the Self-Generation Incentive Program. 2023. Self-generation incentive program: Incentive step tracker. https://www.selfgenca.com/home/program_metrics/ .Sullivan, Michael J., Myles T. Collins, Josh A. Schellenberg, and Peter H. Larsen. 2018. Estimating power system interruption costs: A guidebook for electric utilities. Lawrence Berkeley National Laboratory Report 2001164. Sun, Liyang, and Sarah Abraham. 2021. Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics 225:175–99. Trevizo, Perla, Ren Larson, Lexi Churchill, Mike Hixenbaugh, and Suzy Khimm. 2021. Texas enabled the worst carbon monoxide poisoning catastrophe in recent U.S. history. Texas Tribune, August 17. University of Michigan Population Studies Center. 2022. Measures of rurality for zip codes in the United States. http://web.archive.org/web/20220808013116/https://www.psc.isr.umich.edu/dis/data/kb/answer/1102.html .US Census Bureau. 2021. American Housing Survey. https://www.census.gov/programs-surveys/ahs/data.html .US Energy Information Administration. 2022. Residential Energy Consumption Survey. https://www.eia.gov/consumption/residential/index.php .US Environmental Protection Agency. 2010. An overview of PUCs for state environment and energy officials. https://www.epa.gov/sites/default/files/2016-03/documents/background_paper.pdf .Wolak, Frank A. 2021. Long-term resource adequacy in wholesale electricity markets with significant intermittent renewables. NBER Working paper 29033, National Bureau of Economic Research, Cambridge, MA. Zamuda, Craig D., Peter H. Larsen, Myles T. Collins, Stephanie Bieler, Josh Schellenberg, and Shannon Hees. 2019. Monetization methods for evaluating investments in electricity system resilience to extreme weather and climate change. Electricity Journal 32 (9): 106641. Zip-Codes.com . 2021. California ZIP Codes.https://www.zip-codes.com/state/ca.asp#zipcodes .
