Energy, Energy, Energy Climate, Regulatory Policy, Technology
April 7th, 2020 61 Minute Read Report by Jonathan A. Lesser, Charles D. Feinstein

Playing with Fire: California’s Approach to Managing Wildfire Risks

In 2017 and 2018, wildfires that swept through northern California caused billions of dollars in property damages, destroyed thousands of structures, and killed more than 120 people. The 2018 Camp Fire—which killed 85 people and destroyed virtually the entire town of Paradise—was found to have been caused by faulty equipment installed and operated by Pacific Gas & Electric Company (PG&E).[1]

While such electric operations–related wildfires have accounted for a relatively small percentage of large wildfires, they appear to have had above-average impacts, accounting for 10 of the 20 most destructive wildfires in the state[2] and four of the 20 largest wildfires.[3] The 2018 Camp Fire, which also burned more than 150,000 acres and destroyed almost 19,000 structures, was by far the deadliest and most destructive in state history.

Because of the dangers posed by power lines and faulty electric equipment, the California legislature approved a protocol for electric utilities to reduce the risk of wildfires by preemptively shutting off electricity in portions of their service territories during dry and windy weather conditions.[4] PG&E’s website recommends that customers prepare “for outages that could last longer than 48 hours.”[5]

The first such “Public Safety Power Shutoff” (PSPS) took place in September 2019, when PG&E cut power to 1,400 residential customers and businesses in Sonoma and Napa Counties. Far larger shutoffs, affecting almost 1 million PG&E customers, took place in October and November.[6] Although Southern California Edison (SCE) and San Diego Gas & Electric (SDG&E) also imposed preemptive shutoffs in autumn 2019, the numbers of affected customers were far smaller than those imposed by PG&E.[7]

PG&E and California governor Gavin Newsom have stated publicly that these preemptive shutoffs are necessary to ensure the safety of Californians. In early October 2019, the governor stated that the shutoffs were “the right decision.”[8] Later in the month, after increasing public anger about the widespread shutoffs, Newsom assailed PG&E for mismanagement, claiming that such mismanagement was about “corporate greed meeting climate change.”[9] The public protests reflect a simple fact: although preemptive shutoffs reduce the likelihood of wildfires caused by electric equipment, they impose significant costs on thousands of individuals and businesses.

This report estimates the full range of societal costs and benefits from reducing the likelihood of electric operations–caused wildfires to determine whether, on balance, a preemptive shutoff policy makes economic sense. We find that, except under extreme assumptions, the costs exceed the expected benefits. The reasons are: (i) the low likelihood of electric operations–caused wildfires; (ii) the relatively low expected costs of such wildfires if they do occur; and (iii) the large costs of shutoffs. We conclude by suggesting alternative approaches and policies to reduce the risks of wildfires.

  • From a societal perspective, the costs to individuals and businesses from preemptive shutoffs of electricity exceed the benefits gained by reductions in the likelihood of a wildfire, unless few customers are affected by the shutoffs.

  • From an electric utility’s perspective, preemptive shutoffs are economically rational. They reduce the utility’s potential liability from a wildfire caused by a failure of, or damage to, electric operations equipment, even if that equipment is working properly, while the utility incurs no costs, other than lost revenues from forgone electricity sales. Hence, preemptive shutoffs are a form of low-cost insurance.[10]

  • The implicit justification for preemptive shutoffs appears to be based on a worst-case assumption that every electric operations–caused wildfire will cause catastrophic damages. But the Camp Fire was an outlier; and outliers are, by their nature, rare events. PG&E has never provided cost-benefit analyses for any of the preemptive shutoffs that have taken place since autumn 2019.

  • The risk of damage from wildfires in California has been exacerbated by environmental policies that prevent the clearing of dead and diseased trees from forests and limit opportunities for controlled burns.[11] These policies increase the potential consequences of wildfires by increasing the quantity of available fuel.

  • Preemptive shutoffs are likely to remain PG&E’s policy of choice for the foreseeable future because the utility is neither trimming trees nor replacing equipment that is in poor condition at a rate sufficient to provide safe operations in a relatively short time. But given public anger, whether state regulators and politicians will allow the company to implement these widespread shutoffs in the future is unknown.

DOWNLOAD PDF

I. Background and Origins of the Preemptive Shutoff Policy

Wildfires are not new to California; hundreds occur each year. The vast majority are small, do little harm, and provide ecological benefits.[12] But wildfires and humans don’t mix well. And in 2017 and 2018, California experienced several deadly blazes, including the 2018 Camp Fire, which killed 85 people—the deadliest wildfire in state history.

Most wildfires are not caused by—or do not even involve—electric utility equipment, such as downed power lines. According to data published by the California Department of Forestry and Fire Protection (CalFire) in 2017, a particularly bad year for wildfires, there were 3,470 such blazes, of which 408 (12%) were caused by electric power equipment. Vehicles were responsible for 309 (9%), burning debris for 437 (13%), and arson for 222 (6%), among other causes.[13] The causes of 871 wildfires (25%) were never determined.

Moreover, wildfires that do involve electric utility equipment are typically quite small. Since 2014, electric utilities have been required to notify the California Public Utilities Commission (CPUC) of all wildfires caused by electric operations.[14] During the five-year period 2014–18, the state’s three investor-owned electric utilities—Pacific Gas & Electric (PG&E), San Diego Gas & Electric (SDG&E), and Southern California Edison (SCE), reported 2,583 fire incidents.[15] Of that total, 506 spread less than three meters (10 feet) from their ignition point. Another 1,538 were less than one-quarter of an acre. Thus, 75% of the fires were no more than one-quarter acre in size. At the other end, 22 fires were larger than 100 acres and 10 were larger than 1,000 acres. Thus, less than four-tenths of 1% of reported wildfires over this five-year period were larger than 1,000 acres. PG&E’s service territory accounted for 77% of the wildfires, SCE accounted for 18%, and SDG&E accounted for 5%.

Over the 10-year period 2009–18, there were 45 large (greater than 300 acres) wildfires caused by faulty electrical equipment and downed power lines (all but one in PG&E’s service territory), accounting for less than 8% of the 654 large wildfires in the state (Figure 1). Of the 45 wildfires related to electric operations, 20 occurred in 2017. In 2018, there were six.[16]

One reason that all but one of the large wildfires caused by electric operations occurred in PG&E’s service territory is that the utility’s service territory is the largest in California: about 70,000 square miles, or about 42% of the entire state, extending from Santa Barbara County in the south to Humboldt and Shasta Counties in the north (Figure 2). Another reason is that much of the company’s service territory encompasses areas in the state identified as having a high risk of wildfires (Figure 3), such as heavily forested areas where trees have suffered from drought and disease. Still another reason: some of PG&E’s transmission and distribution (T&D) equipment in these high-risk areas is more than a century old—well past its expected useful life and thus more prone to failures that can cause wildfires.[17]

In 2017, 20 destructive wildfires in California were caused by faulty electric equipment: 14 in October alone, all of them in PG&E’s service territory. Collectively, these wildfires burned over 228,000 acres of land, destroyed 8,900 structures, and caused the deaths of 44 people (Figure 4). The single deadliest and most destructive, the Tubbs Fire, burned almost 37,000 acres in Napa, Sonoma, and Lake Counties, destroyed more than 5,600 structures, and caused the deaths of 22 people.[18] CalFire determined that 12 of the 14—but not the Tubbs Fire—were caused by faulty equipment owned by PG&E.

Although fewer wildfires occurred in 2018, and fewer wildfires were caused by operational failures of electrical equipment, their toll was greater than in 2017. CalFire determined that the Camp Fire, which started in November 2018 in Butte County, was caused by PG&E power lines. It burned more than 153,000 acres, destroyed almost 19,000 structures, including virtually the entire town of Paradise, and caused the deaths of 85 people.[19] No other wildfire in the state has come close to this number of fatalities.

On average, damages and deaths from wildfires caused by electric power equipment failures or downed power lines in California have been far less severe than the Camp Fire. For example, there were no electric operations–caused fires at all in 2011, although there were 62 wildfires that burned over 300 acres each (Figure 1). During 2009–16, there were, at most, four electric operations–related wildfires per year; in 2018, there were six. During the 10-year period 2009–18, there were, on average, fewer than five electric operations–related wildfires larger than 300 acres annually.[20]

Similarly, on average, electric operations–related wildfires have been far less damaging than the ones in 2017 and 2018. During 2009–18, the average electric related–wildfire burned just over 15,000 acres, destroyed about 700 structures, and led to the deaths of three persons. While this destruction is surely considerable, it indicates that wildfires with the impacts of the Camp and Tubbs Fires are extremes, rather than the norm.

The Camp and Tubbs Fires, however, galvanized the state’s electric utilities—especially PG&E—into taking preventive actions, notably preemptive shutdowns in areas during dry and windy weather conditions when electric related–wildfires are most likely to start, particularly by downed power lines.

The preemptive shutdowns have proved controversial. Although de-energizing the T&D system in areas of a utility’s service territory eliminates the risk of an electric operations–related wildfire starting in those areas, PG&E shutdowns, especially the largest ones in October 2019, affected almost 1 million customers (about 2.7 million people). Many of these customers complained loudly—and subsequently, so did politicians.

Thus, an important policy question is whether such shutdowns, despite the controversy, make economic sense. Do the benefits of reducing the risk of wildfires by power shutoffs outweigh the economic costs borne by customers? To answer that question, we first set out the general conceptual framework for our analysis. Specifically, because the consequences of wildfires have multiple dimensions—physical, financial, and even emotional—we need a framework that can incorporate these different dimensions in a consistent manner so that we can count all the costs and benefits and, in doing so, better answer the cost-benefit question.

II. Measuring Physical Risk and Evaluating Risk-Management Alternatives: A Conceptual Framework

Unlike financial risk—which is associated with adverse outcomes that are measured in dollars—physical risk, such as the risk associated with electric system operations, has several adverse outcomes, or dimensions. These can include the financial impacts of property damage but also deaths and injuries to employees and the general public, damages to wildlife, and environmental impacts.[21]

Evaluating the costs and benefits of alternative risk-management strategies that can have effects across several dimensions requires a method that combines observed levels of the different dimensions (which are also called attributes) into a single numerical measurement. A mathematical model that accomplishes this is called a “multi-attribute value function” (MAVF), which we discuss below in the section “Estimating the Consequences of an Electric Operations–Caused Wildfire”(page 13). A MAVF measures the effects of a risk-mitigation strategy on all the attributes and allows us to evaluate the economic efficiency of different risk-mitigation alternatives, even if risk cannot be expressed in dollar terms.[22]

It is natural to think of different types of risk—economic, financial, personal health, and so forth—that one is exposed to in daily life. These risks are based on the adverse consequences associated with the occurrence of an event, perhaps an accident or a failure of some kind. For example, in the context of electric utility operations, a risky event is one that could arise from those operations (e.g., a pipeline explosion, a wildfire caused by a downed power line) and has consequences that society and individuals are willing to pay to avoid. How much society and individuals are willing to pay depends on how likely the risky event is to occur and how severe the consequences are if it does.

Assessing risk and measuring risk reduction means that risk must be expressed as a number. Most commonly, the risk of an adverse event is defined as the probability that an adverse event will take place, called the “likelihood of failure” (LoF), multiplied by the expected consequences of that event, called the “consequences of failure” (CoF).[23] Therefore, the risk of an adverse event depends on two uncertainties: (i) the uncertain occurrence of the adverse event; and (ii) the uncertain consequences of that event, if it occurs. The most useful way to specify and reduce risk is to separate these two uncertainties and then evaluate alternatives that reduce either or both. In other words, risk can be reduced by reducing the likelihood that an adverse event will take place, reducing the consequences of the event if it does occur, or both. It is the interplay between these two uncertainties that determines the risk associated with an event.[24]

If the consequences of a risky event can be measured in dollars, we can calculate the expected cost of the event. For example, if the likelihood of a wildfire is 10% over a specified period of time and we estimate the expected value of the possible damages—loss of life, damage to structures, the environment, and so forth—to be $10 billion, then the risk of the wildfire event over the specified period of time is LoF x CoF = (0.10) x ($10 billion) = $1 billion.

It’s tempting to evaluate risk in terms of worst-case outcomes. After all, that’s why homeowners purchase insurance: it reduces their potential financial loss. While buying insurance to indemnify against the damages of a worst-case outcome can make sense for an individual in some circumstances, such as the loss of a home, doing so at the societal level leads to a fundamental problem: society does not have sufficient resources to insure against all worst-case outcomes. Moreover, policies that focus on avoiding worst-case events ignore a basic statistical fact: worst-case events are, by their nature, rare.

Estimating the Likelihood that a Wildfire Will Occur

LoF can be estimated based on past experiences, other statistical data, or even an educated guess. For example, a natural gas utility might estimate that there is a 1-in-100 chance of a pipeline rupture event somewhere on its distribution system this year. In other words, the estimated frequency of the rupture event is one every 100 years. The resulting likelihood that there will be at least one such rupture in the next year is 0.00995, or about 0.01.[25]

The likelihood of failure can be expressed using what is called a hazard-rate function. For example, the annual hazard rate is the conditional probability that failure will occur within the next year, given that the failure had not already occurred. A hazard-rate function also can express the dependency of failure probability on the age of a piece of equipment, or “asset,” because for some assets—such as transformers, poles, wires, other electric equipment, vehicles—the likelihood of failure typically increases with age. In addition, if something is known about the condition of a piece of equipment, and the likelihood of failure also depends on its condition, we can specify a condition-dependent hazard rate.[26] For example, old and corroded natural gas pipe is more likely to fail than pipe known only to be old, and old pipe is also more likely to fail than new pipe. Generally, assets in different conditions will have different likelihoods of failure as they age (Figure 5). As shown, the probability of failure typically increases as assets age over time, regardless of their condition (i.e., “good,” “fair,” or “poor”), while at any given age (“years from now” on the horizontal axis), the probability of failure typically increases as an asset’s condition worsens.

In our analysis of wildfire risk, we begin by estimating the base hazard rate for electric operations–caused wildfires. The base hazard rate can be thought of as the average likelihood of failure, either when the asset’s failure is unrelated to its condition or when there is no information about the condition of the asset.

Although electric utilities have reported hundreds of wildfires caused by electrical equipment, the vast majority have been tiny and caused little or no damage. During the five-year period 2014–18, only 33 of 2,583 reported wildfires, just over 1%, were larger than 100 acres, and only 10 of those were larger than 1,000 acres.

Because our analysis of wildfire risk examines the costs and benefits of policies to address large and destructive wildfires (greater than 300 acres burned), our empirical analysis focuses solely on those wildfires. Before 2009, CalFire did not break out wildfires caused specifically by electric utility equipment. Instead, the agency included all such fires under the general category of “equipment related.” According to CalFire, most such equipment-related wildfires are the result of incidents involving vehicles, welding equipment, and so on. Therefore, to estimate the likelihood of large wildfires, we have relied on data published by CalFire for the 10-year period 2009–18. (We do not include 2019 because the preemptive shutoff policy that was put into place presumably reduced the likelihood of wildfires during that year. Including 2019 data, therefore, would have resulted in a downwardly biased estimate of the likelihood of an electric operations–related wildfire.)

During 2009–18, there was an average of 4.5 large wildfires per year caused by electrical equipment failures in the entire state.[27] Using this average, the probability that there will be at least one such electric operations–caused wildfire per year is 98.9%.[28]

Because we are evaluating preemptive shutoff policies that have disrupted electric service to customers for a few days at a time, the next step is to calculate the likelihood of an electric operations–caused wildfire (LoF) on a given day during wildfire season. Over the past 10 years, the earliest recorded electric operations–caused wildfire was Eastern 2, which began in San Benito County on June 4, 2018. It was relatively small, burning 513 acres. It destroyed no structures and caused no deaths. Nevertheless, eight of the 45 electric operations–caused wildfires have taken place in June. The latest date on which an electric operations–caused wildfire occurred was the McCabe Fire, which began on November 22, 2013, and burned just over 3,500 acres. Therefore, we believe that the appropriate time frame over which to calculate the daily probability of an electric operations–caused wildfire is the six months between June 1 and November 30, a period of 183 days.

Using this six-month wildfire season period, but before incorporating specific weather and equipment conditions (addressed in the next section of this report), we can calculate the base hazard rate and the probability of an electric operations–caused wildfire on any given day during wildfire season using the Poisson distribution. The resulting probability is 2.43% for the entire state and 2.38% for PG&E’s service territory.[29] In other words, based on the past 10 years of data (2009–18), on each day during wildfire season under average weather conditions (and ignoring the condition of electric equipment), there is just under a 1-in-40 chance of an electric operations–caused wildfire. Over a threeday period, the probability of an electric operations–caused wildfire somewhere in the state increases to 7.11%.[30] The probability in PG&E’s entire service territory over a three-day period is 6.96%.[31]

Of course, when an electric utility has undertaken a preemptive shutoff, the utility has not shut off electricity to all customers in its service territory. Rather, it has shut down power in counties where the likelihood of a wildfire presumably is highest. Therefore, to examine the impacts of a preemptive shutoff on affected customers, we have to adjust the probability further to account for the likelihood of an electric operations–caused wildfire in the affected area itself.

According to PG&E, the company has—in high firethreat districts—25,598 miles of overhead distribution wires and 5,542 miles of overhead transmission lines, for a total of 30,140 miles of overhead circuits.[32] Before accounting for weather conditions or the condition of the equipment, we assume that it is equally likely for these circuits and the equipment along those circuits to fail. In other words, lacking any other information, we assume that there is an equal likelihood of an electric operations–related failure along the 30,140 miles of overhead circuits in high fire-threat districts.[33] Based on this assumption, the probability of an electric operations–caused wildfire along any given mile of overhead circuit is 0.00008%.[34] Thus, suppose that an area to be preemptively shut down has a total of 1,000 overhead circuit miles. In that case, the probability of a wildfire taking place on a given day in that area is 0.008%, or about a 1-in-1,200 chance.[35]

Finally, we can calculate the probability of a wildfire during a multiday shutoff in a given area. For example, for a planned three-day shutoff in this same area with 1,000 miles of overhead T&D circuits, the probability of a wildfire is 0.239%, or about a 1-in-400 chance.[36]

The Effects of Weather and Electrical Equipment Conditions

So far, we have calculated electric operations–caused wildfire probabilities without considering the effects of specific weather and equipment conditions. To account for weather conditions, several utilities, including PG&E, have developed what they call a “fire potential index” (FPI).[37] For example, PG&E defines FPI as an average of separate indexes of weather and fuel conditions. The company then states that “the Utility FPI is a logistic regression model and is related to the probability of a small fire becoming a large incident.”[38]

For each of the individual indexes, the company assigns a scale of 1 (low) to 6 (extreme). The fuel index comprises three variables: (i) dead fuel moisture (DFM);[39] (ii) live fuel moisture (LFM);[40] and (iii) an enhanced vegetation index (EVI), each having values between 1 and 6.[41] The overall fuel condition index is the average of these three variables. The Utility FPI can take values between 1 and 6. For example, if the weather index is 4 (very high) and the fuel index is 3 (dry), the FPI value is 3.5.[42] The probability of a wildfire is then based on the FPI value.

Because there are no publicly available PG&E documents that describe the actual logistic model used by the company, we do not know how PG&E uses FPI to estimate the probability of a wildfire caused by electric equipment. Moreover, FPI does not account for the condition of electrical equipment. Lacking this knowledge, we instead develop condition-dependent hazard rates based on a matrix of weather/fuel and equipment conditions. Specifically, we assume that weather/ fuel and electrical equipment conditions can each be classified as “poor,” “fair,” or “good.”[43] This results in nine possible combinations. For each combination, we provide a multiplier, which is applied to the expected number of wildfires used in the Poisson probability calculations (Figure 6). For simplicity of communication, we call these multipliers hazard-rate multipliers.

Because we have no data regarding the condition of electrical equipment along individual T&D circuits, Figure 6 shows what we believe are reasonable values for the hazard-rate multipliers. For example, we assume that, under good weather/fuel conditions (e.g., calm winds and wet grounds) and good equipment conditions, the expected number of electric operations–caused wildfires is 10% of the base expected value. We further assume that the expected number of electric operations–caused wildfires will not increase (compared with the base expected value) if electrical equipment is in good condition because PG&E claims that preemptive shutoffs will not be necessary after the company completes repairs on its system. We assume that in poor weather/fuel conditions (e.g., high winds and dry fuel) and poor equipment conditions, the hazard rate for an electric operations–caused wildfire increases by a factor of 10. Thus, in total, our analysis assumes that weather/fuel and equipment conditions can affect the base hazard rate by a factor of 100— either increasing or decreasing it by a factor of 10. (As discussed in the next section, we perform sensitivity studies on these values to determine their impact on whether the costs of preemptive shutoffs exceed the benefits, and vice versa.)

Because the CalFire data identify the specific counties where electric operations–caused wildfires have occurred, by using the multipliers in Figure 6 we can calculate the probabilities of such an event for each county when both weather/fuel and electrical equipment conditions are poor (Figure 7).[44]

As shown in Figure 7, Butte County has suffered the most electric operations–caused wildfires: six over the last 10 years. (Three of those six fires took place in 2017.) Madera and Sonoma Counties had the next highest electric operations–caused wildfire frequency, with four each over the last 10 years. (Three of the four Sonoma County fires were part of a larger fire, Central LNU Complex, that began on October 8, 2017.)

Given the frequency of electric operations–caused wildfires over the past 10 years, column [2] shows that the likelihood of at least one such wildfire per year ranges between 9.5% and 45.1%. Column [3] shows that, on a daily basis during wildfire season, the probabilities range between 0.05% and 0.33%. Under a worst-case scenario of bad weather and all electric equipment in poor condition, the daily probability of an electric operations–caused wildfire ranges between 0.5% and 3.2% (see Appendix B: The Mathematics of Calculating the Likelihood of an Electric Operations–Caused Wildfire).

Estimating the Consequences of an Electric Operations–Caused Wildfire

The consequences of a failure event constitute the second of the two fundamental uncertainties that are present in any risky situation. For example, an automobile accident is a risky situation that entails uncertainty that the event will occur and uncertainty about the consequences. In the present situation, the occurrence of a wildfire as a result of electrical operations and the consequences, if it occurs, are both uncertain.

We separate these two uncertainties because riskmanagement alternatives can address the likelihood that an electric operations–caused wildfire will occur (e.g., replacing equipment that is in poor condition), the consequences of such a wildfire (e.g., clearing away accumulated brush near homes in high wildfire-risk areas and thus reducing property damage), or both. Without separating these two uncertainties, it is not possible to estimate risk reduction consistently.

A wildfire can cause deaths, injuries, loss of service, loss of buildings, subsequent flooding, and environmental damage because the ground cover is gone, as well as loss of economic activity. Measuring the consequences of wildfire must take into consideration all these possibilities.

CalFire provides data on the deaths (civilians and firefighters), acres burned, and structures destroyed and damaged that are associated with large wildfires. During the 10-year period 2009–18, the average number of deaths per electric operations–caused wildfire was three. This average value is affected by the 2017 Tubbs Fire, which caused 22 deaths, and the 2018 Camp Fire, which caused 85 deaths.[45] Excluding the Camp Fire, the average number of deaths per electric operations–caused wildfire over the last decade is slightly greater than one (1.14). Moreover, in 37 of the 45 large electric operations–caused wildfires in this 10-year period, there were no deaths (Figure 8).

Next, we consider the distribution of acreage burned (Figure 9). On average, each electric operations–caused wildfire burned about 15,400 acres. However, of the 45 such fires during 2009–18, 20 burned less than 1,000 acres, and 34 of the 45 wildfires (over 75%) burned less acreage than the average. Again, the overall distribution of impacts is skewed, with the outlier Camp Fire burning 153,000 acres (four standard deviations above the average).

Finally, we consider residential and commercial structures destroyed (Figure 10). Here again, the distribution is skewed by the 5,600 structures destroyed in the Tubbs Fire and the almost 19,000 destroyed in the Camp Fire. In 43 of the 45 fires, fewer than 2,000 structures were destroyed. On average, each electric operations–caused wildfire destroyed 684 structures.[46]

Figures 8–10 show that the distributions of the consequences of electric operations–caused wildfires are skewed, with the 2018 Camp Fire the farthest outlier. Based on this 10-year record, it is not reasonable to assume that all future wildfires will be as destructive as the Camp Fire.[47]

As mentioned previously, to estimate the overall consequence of an electric operations–caused wildfire, we can combine these disparate impacts using a multi-attribute value function (MAVF).[48] A MAVF combines the impacts in each consequence dimension or attribute into a single numerical value with no units attached to it.[49] The conversion is based on a set of weights that capture the relative value of changing the outcome in each consequence dimension from its worst to its best outcome. The single numerical value is the weighted sum of the consequences reported for each attribute. As part of the state’s Safety Model Assessment Proceeding (S-MAP) process (see Appendix A: Utility Risk Management in California), California utilities agreed to develop their own MAVFs.[50]

The Importance of Using the Expected Value of Consequences to Measure Riska

As shown in Figures 8–10, wildfires can have a wide range of consequences. The November 2018 Camp Fire burned over 150,000 acres, caused 85 deaths, and destroyed almost 19,000 structures, making it the most destructive wildfire of any cause in California history. But most electric operations–caused wildfires have been relatively small and caused no deaths. What this means is that many failure events, including wildfires, can have a wide range of potential consequences, from minimal to catastrophic.


Everyone prefers to avoid catastrophic outcomes. Thus, the question arises, when measuring risk and formulating policies to reduce risk, should we focus specifically on worst-case outcomes? The answer is no.


There are two reasons. First, expected values account for all possible outcomes, including the extreme (catastrophic and minimal) ones. Second, owing to the properties of probability distributions, the only mathematically legitimate way to compute the risk reduction provided by risk-management programs is based on the difference between the expected consequences before and after the risk-management program has been implemented. 

 

For example, suppose we define risk as the product of LoF and CoF, where CoF is now measured as a 95th-percentile outcome (i.e., a level that would be exceeded only 5% of the time), denoted CoF(.95), rather than the expected value. Suppose a risk-mitigation program reduces the range of possible consequences while keeping the likelihood of occurrence of the event (LoF) unchanged. Although we can estimate the 95th percentiles of both the initial and the reduced-range probability distributions of consequences (CoF), we cannot directly compare the pre- and post-95th-percentile values of risk, as measured by LoF x CoF. The number found by computing ΔR (.95) = LoF x {[CoF(.95)]PRE – [CoF(.95)]POST} is not the risk of anything because the difference of the 95th percentiles, {[CoF(.95)]PRE – [CoF(.95)]POST}, is not the 95th percentile of {[CoF]PRE – [CoF]POST}; hence it cannot be the risk reduction associated with a risk-mitigation program. Although it is always possible to use ordinary arithmetic to compute such numbers, all comparisons based on those numbers are mathematically and logically meaningless.b


[a] The expected value is the probability-weighted average of all possible outcomes. For example, suppose you flip a fair coin (one for which the probabilities of the coin landing on “heads” or “tails” are both 0.50). If the coin lands on heads, you win $10. If it lands on tails, you lose $5. The expected value of the coin flip is then (0.5) x ($10) + (0.5) x (-$5) = $2.50.

[b] The reason is that one cannot “subtract” one probability distribution from another using ordinary arithmetic and, by doing so, determine another probability distribution. In particular, the difference between two 95th percentiles, for example, is not the 95th percentile of anything meaningful.

Hence, risk reduction cannot be defined in terms of the difference between two 95th percentiles.

However, the difference between two expected values is the expected value of the difference between two uncertain situations (each described by a mathematical object called a “random variable”), which is nothing more than a numerical description of the outcomes of an uncertain situation. Therefore, risk reduction can be correctly and naturally defined as the difference between the risks before and after the application of a risk-mitigation measure. In other words, although it is always possible to subtract two numbers, the result is not always meaningful.

Moreover, it is possible to improve (reduce) worst-case outcomes while worsening (increasing) expected outcomes. It would be odd to conclude that a strategy that reduced the worst-case number of deaths, but increased the expected number of deaths, provided a reduction in risk. In general, there is no coherent way to trade changes in expected outcomes against changes in extreme outcomes. In particular, the notion of an equivalence class of pairs of expected and extreme outcomes is not supported by any theory or set of axioms.

Finally, it is completely incorrect to view this situation as analogous to the Markowitz mean-variance portfolio theory, which constructs an “efficient frontier” of portfolios of risky assets defined by the expected value and the standard deviation of the portfolio return. The efficient frontier is not an equivalence class, and no trade-off of mean and standard deviation is implied by the efficient frontier (other than the obvious notion that in order to get a greater expected return, one must accept greater uncertainty in the return).

 

An alternative approach, which is required to perform a cost-benefit analysis, is to convert all the wildfire-related consequences into dollar values. This is the approach we have taken. For example, the U.S. Environmental Protection Agency (EPA) has estimated the statistical value of life at $7.4 million (2006$), equivalent to about $9.5 million in 2019$.[51] The statistical value of life is a measure of an individual’s willingness to pay for a small reduction in the likelihood of accidental death. It is often used to evaluate the costs and benefits of different policies, such as the benefits of reducing premature deaths from pollution controls or the benefits of automobile safety features.

Estimating Risk and Risk Mitigation

As stated previously, we define risk as the product of the likelihood of a failure event (in this case, an electric operations–caused wildfire) and the expected consequences of a failure event, i.e., R = LoF x E[CoF], where “E” denotes the expected value of the consequences CoF, and “R” denotes risk (see sidebar, The Importance of Using the Expected Value of Consequences to Measure Risk).[52] To determine the amount of risk reduction, DR, associated with a specific action, we apply this equation before and after the risk-mitigation action:

     ΔR = LoFPRE x E[CoF]PRE – LoFPOST x E[CoF]POST , where “PRE” and “POST” refer to the risk levels before and after the risk mitigation, respectively.

The next analytical step is to calculate the efficiency of different risk-mitigation alternatives, including preemptive shutoffs, measured in terms of risk reduction per dollar of expenditure. (In the case of preemptive shutoffs, expenditures are primarily the value of lost electricity to customers, as well as additional expenditures by the utility, including the need to inspect power lines before the power is restored.)

III. Cost-Benefit Analysis of Preemptive Electric Shutoffs

The categories of costs and benefits of a preemptive shutoff policy are summarized below (Figure 11). The costs include, first, the harm done to individuals and businesses from not having electric service for several days, including spoiled food, lack of clean water, and lost revenues. In addition, a preemptive shutoff leads to some individuals and businesses using their own generators, many of which are gasoline-powered. According to CalFire, these generators create their own wildfire risk, with all the potential attendant damages and costs.

Weighing against these costs are the benefits of a preemptive shutoff: avoiding the consequences of wildfires in the shutoff area, as well as avoiding the potential costs of firefighting efforts should a wildfire break out.

To compare these costs and benefits, we must assign values to the different costs and benefits, and adjust for the probability that, but for a preemptive shutoff, a wildfire would break out.

As discussed, we convert all impacts to dollar values. The disadvantage of this approach is that we are forced to assign dollar values to impacts that some argue cannot be reduced to mere dollars. For example, reasonable individuals can differ as to the dollar value of a lost wildlife habitat or a lost species, while some individuals entirely reject the concept of monetizing such impacts.[53] In our view, although money is an imperfect measure of some types of damages, it is the best one we have. Moreover, regardless of how some impacts are valued, society must allocate scarce monetary resources among competing—and often compelling—needs. Money is a reasonable measure on which to base such allocations.

Estimating the Statistical Value of Life

The statistical value of life (SVL) reflects how individuals make decisions that reflect their values about health and mortality risk, including the jobs they work, the types of vehicles they drive, smoking, drinking, and so forth. By evaluating how individuals make trade-offs between engaging in activities that change the likelihood of dying and the monetary rewards of risk-increasing activities, economists have developed SVL estimates.[54] These estimates, in turn, are often used in cost-benefit analyses of policies that affect mortality risk. In essence, SVL is based on willingness to pay for a small reduction in the probability of death.

For example, the U.S. Department of Transportation uses SVL estimates to evaluate safety measures in automobiles. The most recent value adopted by the agency was $9.6 million (2016$),[55] which is equivalent to $10.2 million in 2019$.[56] EPA uses a value of $7.4 million (2006$), which is equivalent to $9.4 million in 2019$, to evaluate the health benefits of environmental regulations.[57] For our purposes, we believe that it is reasonable to use an average of the two values, or $9.8 million, to evaluate the potential costs associated with electric operations–caused wildfires.[58]

Estimating the Value of Lost Electric Service from Preemptive Shutoffs

The most common measure used to estimate the value of lost electric service is called the value of lost load (VOLL). VOLL can be thought of as an estimate of a customer’s willingness to pay to avoid losing electricity for a given period. (VOLL can also be based on a customer’s willingness to accept compensation for a service interruption.) VOLL depends on many factors, including the type of customer (e.g., residential, commercial, or industrial), the duration of an outage (in general, VOLL increases as the duration of an outage increases), the time of year, the number of interruptions the customer has experienced, and lost business revenues and equipment damage.

The simplest measure of aggregate VOLL across all types of customers can be calculated by dividing gross domestic product (GDP) by total electricity consumption. In effect, this can be thought of as a measure of the productivity value of electricity for the economy as a whole. For example, in 2018, the U.S. GDP was about $21 trillion[59] and total retail sales of electricity were just over 3,860 billion kilowatt- hours (kWh).[60] By this measure, the overall VOLL for all customer classes in the U.S. was about $5.44 per kWh. We can also perform the same calculation for California. In 2018, California’s gross state product (GSP) was approximately $3 trillion,[61] and total retail electric sales in the state were about 255.3 billion kWh, implying an overall California VOLL of $11.75 per kWh.

Other methods for estimating VOLL include surveys—asking customers to identify their willingness to pay to avoid outages (or willingness to accept an outage), inferred estimates based on the choices that customers make, and “bottom-up” analyses that estimate the costs to operate a backup generator or, in the absence of a generator, tallying up costs including spoiled food, lost business revenues, hotel accommodations, and lost wages.

For example, a 2013 study used surveys to estimate VOLL for a two-day outage for customers in Austria.[62] The estimated VOLL for residential customers was $18/kWh (2012$) for a one-hour outage, increasing to $22.25 per kWh (2012$) for a 24-hour outage.

That same year, a report prepared for the Maryland Public Service Commission and the National Association of Regulatory Utility Commissioners (NARUC) used a bottom-up analysis. The authors estimated that a four-day outage would impose costs between $726 and $1,104 (2011$) on residential customers who lacked a backup generator, or between $181 and $276 (2011$) per day.[63] Over the three-year period 2016–18, average electricity use by PG&E residential customers was about 16.5 kWh per day.[64] Thus, over a four-day outage, the implied VOLL for residential customers is between $11.00 per kWh and $16.73 per kWh (2011$).[65]

Damages to commercial and industrial customers depend on the nature of the affected operations. The 2013 NARUC study, for example, estimated the bottom- up damages associated with a four-day outage for a large restaurant to be between $27,277 and $36,394 (2011$), including lost revenues, equipment damage, and spoiled food.[66] That study’s average annual electricity use for a single commercial customer is 49 megawatt-hours, which was obtained from data published by the U.S. Energy Information Administration. The corresponding VOLL is between $50.80/kWh and $67.78/kWh (2001$).[67]

Another study in 2013, prepared by London Economics for the Texas independent system operator, ERCOT (which controls the electric generation and dispatch in the state), reported VOLL to be between $0/kWh and $17.98/kWh for residential customers and between $3.00/kWh and $53.91/kWh for commercial and industrial customers.[68] The $0/kWh low-end range value for residential customers makes little economic sense because it implies that customers place no value whatsoever on lost electricity. Although some residential customers might not be willing to pay to avoid a momentary outage that is a mere nuisance, it is unreasonable to assume that a residential customer would be unwilling to pay to avoid a prolonged outage resulting in spoiled food, loss of access to clean water, hotel accommodations, and so forth—or alternatively, to avoid the costs of operating a backup generator.

Although the range of values is large, we believe that it is reasonable to assume a range of VOLL between $10/kWh and $20/kWh for all affected California customers.[69] For a residential customer using about 16 kWh of electricity per day, that translates into a daily cost of between $160 and $320.

Estimating the Value of Structures Destroyed

The value of structures destroyed by a wildfire depends on the type of structure (e.g., residential and commercial buildings, schools, hospitals) and the cost to rebuild them. The 2017 Tubbs Fire destroyed more than 5,600 structures in Napa and Sonoma Counties (Figure 4). According to city officials in Santa Rosa, the fire destroyed 2,834 homes and 400,000 square feet of commercial building space.[70]

The cost of rebuilding a home varies widely within California. For Santa Rosa, the costs were estimated to be about $300 per square foot.[71] According to data published by the U.S. Census Bureau, the median size of a new home sold in 2018 was 2,435 square feet.[72] Even assuming that the average size in California is smaller—say, 2,000 square feet—that translates to $600,000 to rebuild a typical home in Santa Rosa, or $1.7 billion in total rebuilding costs for that city.

For commercial space, we assume the cost of reconstruction to range between $100 and $200 per square foot.[73] Thus, for Santa Rosa, the commercial reconstruction cost is between $40 million and $80 million.

If we assume that the average Santa Rosa home destroyed was 2,000 square feet, the total square footage lost was (2,834) x (2,000) = 5.668 million square feet. Adding the 400,000 square feet of commercial space destroyed, the total square footage lost in the Tubbs Fire was 6.068 million square feet, of which residential square footage represents 93.4% and commercial square footage represents 6.6%. Using the $300/square-foot average cost for residential space and a $200/square-foot value for commercial space, the weighted average cost of reconstruction is $293.41 per square foot. Because we do not know the specific number of commercial structures destroyed, we use the overall average size for each of the 2,834 structures of 2,141 square feet.[74] This implies an average replacement cost per structure of (2,141 square feet) x ($293.41 per square foot) = $628,229.

Estimating the Value of Raw Land Burned

Estimating the dollar value of raw acreage burned in a wildfire is difficult because that value depends on the type of land burned. For example, forested land has stumpage value (trees that can be harvested today or in the future). Forested land and grasslands can also have recreational value and environmental value (habitat for wildlife) and provide protection from floods.[75]

A few studies have estimated the value of different types of raw land. In 2001, the Wilderness Society estimated the value of temperate forests at $122/ acre (1994$), equivalent to $194/acre (2019$).[76] A 2012 study estimated total economic values for woodlands and grasslands at $643/acre and $1,040/ acre (2007$), respectively, equivalent to $1,022/ acre (2019$) and $1,653/acre (2019$), respectively.[77] More recently, a 2014 study estimated values for temperate forest and grasslands at $1,270/acre and $1,686/acre (2007$), respectively, equivalent to $1,553/acre (2019$) for temperate forest and $2,062/acre (2019$) for grasslands.[78]

To be conservative, we adopt the values developed by the 2014 study. Although an electric operations–caused wildfire may be more likely to begin in forested land because of trees falling on wires, we cannot know in advance the mix of land that a wildfire eventually will burn. Thus, it seems reasonable to use an average value of $1,807/acre (2019$) for the cost of land destroyed by a wildfire.

Estimating the Cost of Firefighting Efforts in California

Using CalFire data, we estimated the cost of fire suppression per acre for all large (>300 acres) wildfires over the past 10 years (Figure 12). CalFire does not publish data on the cost of suppressing each wildfire but instead publishes annual suppression costs by fiscal year (October 1–September 30).

According to CalFire, average annual suppression costs per acre have ranged between $654/acre and $9,467/ acre. The weighted average cost per acre (weighted by annual acres burned) is $1,539 (2019$), and the simple average cost per acre is $2,779/acre. Rather than using the weighted average cost per acre, we adopt the simple average value, in order to be conservative.

The Expected Consequences of an Electric Operations–Caused Wildfire

As discussed in the previous section, when estimating risk and risk reduction, it is important to base the analysis on the expected consequences of a wildfire event, rather than worst-case outcomes.

To calculate the CoF value, we use the expected consequences of an electric operations–caused wildfire previously shown in Figures 8–10. Using our estimates of SVL, cost per structure destroyed, and value of raw land burned, the resulting average cost is $529.6 million (2019$) (Figure 13).[79]

To calculate the overall wildfire risk, we multiply this $529.6 million cost by LoF values for a three-day period in the given area. For example, as discussed previously, the probability of an electric operations–caused wildfire during a three-day outage in an area with 1,000 miles of overhead T&D lines is 0.239% if nothing is known about weather and equipment condition.[80] Assuming that all 1,000 miles of T&D lines are in poor condition and that weather conditions are also poor, the resulting probability value is 2.36%, based on the hazard-rate multiplier of 10, shown in Figure 6.[81] This is LoF for an electric operations–caused wildfire when both weather and equipment conditions are poor. Similarly, using the county-level data in Figure 7, LoF values under poor conditions ranged between 0.5% and 3.2% per day in these counties. (Over a three-day period, the respective probabilities are approximately three times larger.)

Using these data, the expected cost of an electric operations–caused wildfire over 1,000 miles of T&D lines in poor condition and in poor weather conditions over a three-day period is (2.36%) x ($529.6 million), or $12.5 million (2019$). The expected cost for Sonoma County—based on the likelihood of 6.35% for an electric operations–caused wildfire in that county over a three-day period[82]—is (6.35%) x ($529.6 million) = $33.6 million. The expected cost over PG&E’s entire high fire-risk service territory is (51.39%) x ($529.6 million), or $272.2 million, assuming that all the company’s T&D equipment was in poor condition and that weather/fuel conditions are poor. If a preemptive shutoff covered one-fourth of all PG&E’s T&D lines in high fire-risk areas, and if all those T&D lines were in poor condition and weather/fuel conditions are poor, then the probability of a wildfire is about 16.5% over a three-day period and the expected cost in poor weather/fuel conditions is just over $87 million.[83]

Estimating the Cost of a Preemptive Shutoff

Next, we consider the costs of a preemptive shutoff using specific data from one such event: PG&E’s October 9–12, 2019, PSPS, which affected 728,300 customers,[84] about 15% of all PG&E customers. Of these 728,300 customers, PG&E reported that 636,000 were residential customers, 81,000 were commercial/ industrial customers, and the remaining 11,300 were “other customers” (almost entirely public street and highway lighting customers).[85]

The preemptive shutoff took place in four phases of differing durations: (i) 89 hours for Phase 1 customers; (ii) 75 hours for Phase 2 customers; (iii) 62 hours for Phase 3 customers; and (iv) 55 hours for Phase 4 customers. According to PG&E, the company sent shutoff notices to 507,000 customers in Phase 1, about 232,000 notices to Phase 2 and 3 customers, and about 42,000 notifications to Phase 4 customers.[86] Using PG&E’s FERC Form-1 data, we calculate average consumption by customer class in 2018, based on total customers and total consumption (Figure 14). The average consumption per day for residential customers was 15.7 kWh; for commercial and industrial customers combined, the average consumption was 222.0 kWh per day. The average consumption for other customers, primarily public street and highway lighting, was 23.2 kWh per day.

Using these averages, the duration of each phase, and a range of VOLL estimates between $10/kWh and $20/kWh, we then calculate a range of outage costs. Because the majority of customers affected were in Phase 1, for which the outage lasted 89 hours, we assume an average duration of 72 hours for all four phases. Figure 15 shows the resulting estimates of lost electric service value.

As Figure 15 shows, we estimate a range of costs associated with the October 9–12, 2019, preemptive shutoff to be between $846.9 million and $1.69 billion. (This ignores the additional potential cost associated with wildfires started by customers using portable generators run by natural gas or liquid petroleum gas.) By comparison, as discussed above, we estimate an expected cost of an electric operations–caused wildfire of $272.2 million in PG&E’s entire high fire-risk service territory under worst-case conditions over a three-day period. Thus, the lost value of electric service to the subset of PG&E customers affected by the October 9–12 PSPS is more than triple the expected cost of an electric operations–caused wildfire in PG&E’s entire high fire-risk territory under poor weather/fuel conditions and assuming that all of PG&E’s T&D equipment is in poor condition. In fact, even if an electric operations–related wildfire was a certainty in the affected area (i.e., LoF = 100%), the expected benefit of a preemptive shutoff— avoiding the expected $529.6 million cost of an electric operations–caused wildfire—would still be less than the lower-bound value $846.9 million cost shown in Figure 15.

Of course, the counties affected by the October 9–12 PSPS are a subset of PG&E’s entire service territory. For example, if we assume that they account for onethird of all T&D lines in high fire-risk areas and that all those lines are in poor condition, then the expected avoided wildfire cost in poor weather/fuel conditions would be about $113 million, roughly one-eighth of the lower-bound estimate of PSPS costs to customers from lost electric service. Moreover, even if VOLL were just $4/kWh, a value far less than the empirical estimates, the cost to customers from lost electric service would be about $339 million, still exceeding the $272.2 million expected cost of a wildfire in PG&E’s entire service territory.[87]

Sensitivity Analysis

We next evaluated various sensitivities to determine if changing individual parameter values would change the cost-benefit results (Figure 16).

Case A assumes that the hazard rate associated with an electric operations–related wildfire is 20 times greater than under average conditions, rather than the basecase assumption of 10 times greater. Case B assumes that the expected impacts of an electric operations–related wildfire are twice the expected CoF value that we previously derived, $529.6 million. Case C assumes that VOLL is half the base-case range.

Figure 17 presents the results of these sensitivity studies. If we assume double the expected wildfire cost and assume that VOLL is less than $6.43/kWh, the cost of the shutoff is less than the avoided wildfire cost. Similarly, if we apply all three sensitivity cases simultaneously, the expected benefit of an avoided electric operations–caused wildfire exceeds the cost of lost service if VOLL is less than $9.55/kWh. This sensitivity study compares the cost of a wildfire for PG&E’s entire high fire-risk territory with the estimated lost service cost for the October 9–12, 2019, wildfire. If we use an estimate that the affected service area represents onethird of PG&E’s entire service territory, the expected benefits of an avoided wildfire would still be less than the shutoff costs, even applying all three sensitivities simultaneously.

Next, we consider the break-even probabilities of occurrence for an electric operations–caused wildfire over a 24-hour shutoff period, based on the expected wildfire impact of $529.6 million and the number of customers affected (Figure 18).

In Figure 18, the top line shows the break-even probability of occurrence of an electric operations–caused wildfire as a function of the number of customers whose power is prematurely shut off. The bottom line shows the break-even probability under Sensitivity Case B, which assumes that the costs of an electric operations–caused wildfire are twice as large. For example, if 300,000 customers are shut off, then as long as the probability of an electric operations–caused wildfire is greater than about 22%, the avoided expected cost of such a wildfire is greater than the cost of the shutoff. If the expected cost of a wildfire is double the $529.6 million, then the shutoff of 300,000 customers makes economic sense if the probability of a wildfire is greater than about 11%.

Figure 18 also shows that the calculated probability of occurrence of an electric operations–caused wildfire over a one-day period in all of PG&E’s service territory is 21.37%, assuming that all T&D equipment is in poor condition and that weather/fuel conditions are also poor. However, the probability of a wildfire in, say, a given county is far less.[88] For example, in Napa County (Figure 7), the probability of an electric operations–caused wildfire on a given day in wildfire season, assuming that all T&D equipment in the county is in poor condition and that weather/fuel conditions are poor, is 1.63%. Under the same conditions, the probability of an electric operations–caused wildfire in Butte County is 3.23%.

We also examine the same break-even results for an assumed 72-hour outage (Figure 19).

For a 72-hour shutoff, if more than 500,000 customers are shut off, then the costs to customers will exceed the expected wildfire cost of $529.6 million, even if an electric operations–caused wildfire is a certainty. Thus, given the 51.39% calculated probability of a wildfire for all of PG&E’s service territory over a 72-hour period, assuming again that all T&D equipment is in poor condition and that weather/fuel conditions are poor, the costs of a preemptive shutoff will exceed the expected avoided wildfire cost if more than about 250,000 customers are affected, or more than about 475,000 customers if the expected wildfire cost is doubled.

We also examine how combinations of the per-kWh VOLL estimates between $0/kWh and $20/kWh, and between 50,000 and 500,000 customers shut off for a 24-hour period, affect the cost-benefit analysis (Figure 20).

In Figure 20, the blue columns represent combinations of per-kWh VOLL estimates and the number of customers shut off for which the expected avoided cost of a wildfire is greater than the shutoff cost. For example, if 50,000 or fewer customers are affected, the avoided wildfire cost is greater than the cost of the shutoff to customers (again, based on the probability of an electric operations–caused wildfire in all of PG&E’s service territory and assuming that all T&D lines are in poor condition) for VOLL up to $20/kWh. Similarly, if VOLL is $2/kWh, the shutoff is justified for any number of customers up to 500,000.

Similarly, consider the same analysis, but based on the probability of a wildfire over a three-day period in Sonoma County (Figure 21). Here, the benefits of a shutoff exceed the costs to customers only under limited circumstances, such as VOLL of less than $8/ kWh (below our low-range estimate of $10/kWh) and fewer than 50,000 customers affected, and VOLL of $2/kWh if 200,000 customers are affected.

Conclusion: The Costs of Preemptive Shutoffs Exceed the Benefits, Except Under Extreme Assumptions

Our analysis shows that the costs of PG&E’s preemptive shutoffs in autumn 2019, which affected hundreds of thousands of customers, exceeded the expected benefits in terms of avoided costs of electric operations–caused wildfires. Only in a sensitivity case where we assume that: (i) a wildfire is 20 times more likely when both weather and equipment conditions are poor; (ii) the cost of a wildfire is double the expected value; and (iii) VOLL is half as large as the low end of the range that we assumed based on the published literature, do the benefits of a preemptive shutoff exceed the expected costs. Moreover, that result is based on the likelihood of an electric operations–caused wildfire in all of PG&E’s service territory, not an area encompassing a few counties.

The cost-benefit calculations also depend on the number of customers affected. A shutdown that encompasses a large geographic area but that affects few customers may be cost-effective, as the analyses show in Figures 20 and 21. Of course, PG&E’s preemptive shutdown policy has proved controversial precisely because it has affected so many customers.

IV. Costs and Benefits from an Electric Utility’s Perspective

The results of our analysis raise a question: If the costs of preemptive shutoffs are likely to far exceed the expected benefits in terms of reduced wildfire costs—except under extreme circumstances—why would California electric utilities adopt this strategy? There are at least three potential answers.

First, under California’s inverse condemnation law, an electric utility can be held responsible for damages caused by its equipment, even if that equipment is not faulty. In the absence of that law, a utility would not be liable for damages if its equipment were functioning properly. This would reduce a utility’s economic incentive to institute a preemptive shutdown. At present, electric utilities are not liable for damages caused by outages such as spoiled food and lost business revenues. Instead, the only direct cost to the utility is lost sales revenues. In other words, from an electric utility’s standpoint, the expected costs of an electric operations–caused wildfire far exceed the costs of reducing the probability of a wildfire to zero by a preemptive shutdown. The forgone sales revenues, which are minimal, can be thought of as the premium of a relatively inexpensive but effective insurance policy. Hence, given inverse condemnation, California electric utilities have a strong economic incentive to institute a PSPS to prevent an electric operations–caused wildfire. Moreover, the damages caused by an electric operations–caused wildfire are likely to be paid by utility shareholders, not ratepayers. A utility that is forced to use its earnings to pay damages will tend to see its share price fall. For example, on November 7, 2019, PG&E’s share price fell 13% after it released its third-quarter 2019 earnings report, which reported a loss of $3.06 per share.[89] That loss was the result of a $2.5 billion pretax charge against earnings to pay claims for wildfire damages in 2017 and 2018.

Second, California’s electric utilities, regulators, and politicians appear to have adopted a risk-averse approach to wildfires that is not supported by sound economic analysis. Instead, they appear to prefer to attempt to drive the probability of a wildfire to zero and face the subsequent wrath of their customers (and their customers’ elected representatives), wrath that they may expect to be temporary and less severe than the fallout from an electric operations–related wildfire. For regulators, a perception that they have failed to regulate the utilities under their purview adequately is likely the most severe consequence. What will happen as a result of that perception is unclear. As for current politicians, presumably being voted out of office is the most severe consequence.

A third explanation, related to the first two, is that electric utilities and state regulators appear to be basing their policy decisions on worst-case outcomes rather than expected outcomes. Such decision making misallocates scarce resources. Although the Tubbs and Camp Fires were destructive and deadly, they are outliers. On average, the damage from electric operations–caused wildfires has been far less severe. Imposing huge and not uncertain costs on electric utility customers in order to spare them far lower expected costs associated with the uncertain occurrence and uncertain consequences of a wildfire makes no economic sense.

V. Alternatives to Preemptive Shutoffs

Our analysis shows that the expected costs associated with a preemptive power shutoff generally are greater than the expected benefits in terms of avoided wildfire costs unless relatively few customers are affected by the shutoff. However, this does not necessarily mean that the optimal strategy is to do nothing. PG&E intends to perform additional tree-trimming and replace aging equipment over the next decade or so; even so, more actions can be taken.

First, both the likelihood and consequences of wildfires likely can be reduced through improved forest management. For example, controlled burns and the elimination of dead and diseased trees are one widely recommended approach for reducing the intensity and spread of wildfires.[90] Controlled burns, however, often have been opposed because the smoke from such fires reduces air quality and adversely affects health, especially in children. However, a 2019 study by researchers at Stanford University Medical School concluded that smoke from wildfires causes greater adverse health effects than controlled burns.[91] In other words, better to have some smoke from controlled burns rather than greater quantities of smoke from wildfires.

PG&E, of course, is not in the forest-management business and, in any event, cannot undertake prescribed burns and forest clearing. But California and the U.S. government can and should do so, and they can begin immediately.

Another possible approach, which PG&E could implement, especially in coordination with the state, would be to use utility personnel to actively monitor electric operations equipment during the hot, dry, and windy conditions that trigger preemptive shutoffs. PG&E and other utilities already use estimates of weather and fuel conditions to determine where to schedule preemptive shutoffs. In areas where tree-trimming is inadequate or power lines and related equipment are aging and in poor condition, active monitoring—akin to the individuals who monitor national forests from watchtowers—can decrease the response time of firefighters and reduce the likelihood that a wildfire will spread and cause major damage. Moreover, monitors need not be confined to stationary watchtowers; they can actively patrol small areas during adverse weather conditions. Moreover, remote sensing can be performed by aircraft, including drones.[92] If PG&E found such monitoring to be feasible, it could be implemented by the next fire season.

Suppose PG&E identifies an area of high wildfire risk because of weather conditions and the condition of its T&D equipment. Instead of a preemptive shutoff, the utility could employ ground personnel trained to recognize and address equipment failures. PG&E might place two individuals to monitor one mile of T&D lines and equipment, each individual taking 12-hour shifts. If there were 2,000 miles of lines and equipment in the area, the company would require 4,000 individuals for monitoring.

Here is a potential scenario: each individual is provided with camping gear and basic fire-suppression tools (shovels, fire extinguisher, satellite phone, etc.). The cost to outfit each individual is $1,000, and he or she is paid $400/day (just over $33 per hour for a 12-hour shift). Finally, each mobilization costs the utility an additional $10 million to undertake, regardless of size. The total cost in lieu of, say, a three-day shutoff period would be $18.8 million. If there were 100,000 customers affected by the shutoff, the implied total VOLL would be $231 million, based on our $10 per kWh estimate of VOLL and an average consumption of 77.1 kWh per day per customer over the entire PG&E system.[93] The break-even VOLL would be only about $18.8 million / (77.1 kWh per day/customer x 100,000 customers x 3 days) or 81 cents per kWh, far below any published estimates. Moreover, the costs of this monitoring presumably would be paid by PG&E ratepayers, as part of the company’s risk-management efforts.

Over time, as PG&E upgraded its system with additional tree-trimming and equipment replacement, and as the state (and federal government) thinned forests and removed deadwood, the need for such monitors would be expected to decrease each year. During T&D system upgrades, moreover, the utility could add switching and sensors. Similar changes are under way outside the PG&E service area, as electric utility T&D systems develop into a so-called smart grid. The idea here is to detect a downed or sparking line, de-energize it, and reroute the electricity using switches to bypass the downed line and all those customers radially connected to that line. There is reason to believe that this will be an important aspect of the solution to the problem in future years.

Appendix A: Utility Risk Management in California

The Safety and Enforcement Division (SED) of the California Public Utilities Commission (CPUC) oversees safe operations by electric and natural gas utilities.a SED’s role took on increased prominence after PG&E’s 30-inch natural gas pipeline exploded on September 9, 2010, in San Bruno, California. The explosion, which took place in a neighborhood 13 miles south of San Francisco, killed eight people. The federal National Transportation Safety Board (NTSB) determined that the explosion was caused by substandard and poorly welded pipe.b

The San Bruno explosion was a catalyst for the development of risk-management policies by the state’s electric and natural gas utilities. In November 2013, CPUC issued a rulemaking to develop a “risk-based decision-making framework” for utilities to evaluate safety and reliability improvements.c

On May 1, 2015, CPUC began its Safety Model Assessment Proceeding (S-MAP), with applications from California’s four investor-owned utilities.d The utility companies’ applications presented their proposed methodologies to evaluate risk and alternative risk-management actions. For the next several years, there was much debate over the actual methodologies proposed by the utilities. In March 2016, CPUC issued a proposed decision, which adopted a so-called multi-attribute framework for measuring risk and ordered the utilities to adopt a uniform methodology to address risk-management issues.e For the next two years, the utilities and intervenors debated the most appropriate multi-attribute framework,f and in May 2018, a settlement was reached among the parties adopting a final methodology. CPUC adopted the settlement in December 2018.g


[a] California Public Utilities Commission (CPUC), “Safety Policy Statement of the California Public Utilities Commission,” July 2014. CPUC’s General Order 167 (“Enforcement and Maintenance and Operation Standards for Electric Generating Facilities”), which took effect in September 2005, states that its purpose is to “implement and enforce standards for the maintenance and operation of electric generating facilities and power plants so as to maintain and protect the public health and safety of California residents and businesses, to ensure that electric generating facilities are effectively and appropriately maintained and efficiently operated, and to ensure electrical service reliability and adequacy.”

[b] NTSB (National Transportation Safety Board), “Pacific Gas and Electric Company, Natural Gas Transmission Pipeline Rupture and Fire, San Bruno, California, September 9, 2010,” NTSB/PAR-11/01, Aug. 30, 2011.

[c] Order Instituting Rulemaking to Develop a Risk-Based Decision-Making Framework to Evaluate Safety and Reliability Improvements and Revise the General Rate Case Plan for Energy Utilities, Rulemaking 13-11-06, Nov. 22, 2013.

[d] Application of San Diego Gas & Electric Company (U902M) for Review of Its Safety Model Assessment Proceeding Pursuant to Decision 14-12-025, et al., Docket Nos. A15-05-002, et al. (collectively, “S-MAP”), May 1, 2015. Documents for these proceedings can all be found on the CPUC website: https://apps.cpuc.ca.gov/apex/f?p=401:1:0.

[e] S-MAP, Interim Decision Adopting the Multi-Attribute Approach (or Utility Equivalent) Features) and Directing Utilities to Take Steps Toward a More Uniform Risk Management Framework, Aug. 18, 2016.

[f] The authors of this report worked with representatives of the ratepayer advocacy group, the Utility Reform Network (TURN), in this CPUC proceeding. This work included the development and presentation of an analytic methodology to measure risk and evaluate and select riskmanagement alternatives. That methodology was originally developed for the Electric Power Research Institute (EPRI) (references available upon request). Among the essential features of the methodology are that the consequences of a utility equipment failure are measured in several different dimensions or attributes and that the effectiveness of a risk-mitigation action is measured by the amount of risk that the action reduces. Before the CPUC proceeding, California utilities were using various ad-hoc approaches to choose which risk-mitigation actions they would implement.

[g] S-MAP, Phase Two Decision Adopting Safety Model Assessment Proceeding (S-MAP) Settlement Agreement with Modifications, Dec. 20, 2018.

Appendix B: The Mathematics of Calculating the Likelihood of an Electric Operations–Caused Wildfire

We calculate the likelihood that an electric operations–caused wildfire will occur in all or part of PG&E’s service territory based on the following five factors: (i) the observed average frequency, f, of 4.4 electric operations–caused wildfires in PG&E’s service territory per year; (ii) the number of days, t, during an assumed wildfire season of 183 days; (iii) the number of miles of T&D lines, m, in the portion of PG&E’s service territory we are evaluating, relative to the 30,140 miles of T&D lines in all of PG&E’s high fire-risk areas; (iv) the condition of the T&D assets in that portion of PG&E’s service territory; and (v) the prevailing weather/fuel conditions in that portion of PG&E’s service territory. The combination of the last two factors defines a hazard-rate multiplier, h, as shown in Figure 6. The calculation uses the well-known Poisson distribution. Hence, LoF can be written as:

LoF = 1 – e [- f ( t / 183 )(m / 30.140) h ]

For example, the likelihood of a wildfire in an area with 1,000 miles of T&D lines in poor condition, over a three-day period when weather/fuel conditions are poor, is:

LoF = 1 – e [ -4.4 (3/183) (1,000/30.140) (10) ] = 0.0236 = 2.36%

This is just over 2.3%, or about a 1-in-40 chance. Over all 30,140 miles of T&D lines in PG&E’s high fire-risk service territory in worst-case weather conditions, if all of the electric equipment is in poor condition, then the likelihood of an electric operations–caused wildfire over a three-day period is:

LoF = 1 – e[ -4.4 (3/183) (30.140/30.140) (10)] = 0.5139 = 51.39%

Endnotes

See endnotes in PDF

Donate

Are you interested in supporting the Manhattan Institute’s public-interest research and journalism? As a 501(c)(3) nonprofit, donations in support of MI and its scholars’ work are fully tax-deductible as provided by law (EIN #13-2912529).