3
Indirect Losses of Natural Disasters

Introduction

Indirect losses of natural disasters, or losses resulting from the consequences of physical destruction, have not been measured, studied, and modeled to the same extent as direct losses (the monetized losses of physical destruction). Recent unprecedented business interruption losses—$6.5 billion in Northridge (Gordon and Richardson, 1995) and a staggering $100 billion of interruption losses in the 1995 Kobe earthquake—have focused attention on the need for more intensive scientific study and measurement of these indirect losses. Evidence to date suggests that the proportion of indirect impacts increases in larger disasters, and thus may constitute a larger fraction of total losses and damage in large disasters than in smaller disasters (Gordon and Richardson, 1995 and Toyoda, 1997).

By their nature, indirect losses are harder to measure than losses stemming directly from physical damage. For example, a ruptured power line is readily observed and the cost of its repair evaluated. Far less obvious are losses such as those of industries that are forced to close down because they lack critical power supplies, firms with power that lose business because suppliers or buyers lacked power, and firms that lose business because employees of firms affected by the power outage have reduced incomes and consequently spent less. Compared to a natural disaster's direct effects, indirect losses are more difficult to identify and measure, and are generally spread over a much wider area.

Additionally, there are almost no programs or processes in place to draw upon in measuring indirect losses. Two exceptions to this observation are business interruption insurance and unemployment insurance. The usefulness of these data are limited, as many firms do not carry business interruption insurance, and that many indirect effects may not qualify for reimbursement under such insurance. Similarly, unemployment insurance data do not adequately reflect employment and income losses that may occur in the wake of a natural disaster. For many, proving eligibility can be troublesome; for others, the key impact is not unemployment per se but reduced work and income that does not qualify for program assistance. In both situations, the coverage problem



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 35
--> 3 Indirect Losses of Natural Disasters Introduction Indirect losses of natural disasters, or losses resulting from the consequences of physical destruction, have not been measured, studied, and modeled to the same extent as direct losses (the monetized losses of physical destruction). Recent unprecedented business interruption losses—$6.5 billion in Northridge (Gordon and Richardson, 1995) and a staggering $100 billion of interruption losses in the 1995 Kobe earthquake—have focused attention on the need for more intensive scientific study and measurement of these indirect losses. Evidence to date suggests that the proportion of indirect impacts increases in larger disasters, and thus may constitute a larger fraction of total losses and damage in large disasters than in smaller disasters (Gordon and Richardson, 1995 and Toyoda, 1997). By their nature, indirect losses are harder to measure than losses stemming directly from physical damage. For example, a ruptured power line is readily observed and the cost of its repair evaluated. Far less obvious are losses such as those of industries that are forced to close down because they lack critical power supplies, firms with power that lose business because suppliers or buyers lacked power, and firms that lose business because employees of firms affected by the power outage have reduced incomes and consequently spent less. Compared to a natural disaster's direct effects, indirect losses are more difficult to identify and measure, and are generally spread over a much wider area. Additionally, there are almost no programs or processes in place to draw upon in measuring indirect losses. Two exceptions to this observation are business interruption insurance and unemployment insurance. The usefulness of these data are limited, as many firms do not carry business interruption insurance, and that many indirect effects may not qualify for reimbursement under such insurance. Similarly, unemployment insurance data do not adequately reflect employment and income losses that may occur in the wake of a natural disaster. For many, proving eligibility can be troublesome; for others, the key impact is not unemployment per se but reduced work and income that does not qualify for program assistance. In both situations, the coverage problem

OCR for page 35
--> Some beachfront property became uninhabitable due to damage from Hurricane Fran in 1996. (Photo courtesy of FEMA.) is exacerbated by the complexity of extracting the information from existing sources. Business interruption reimbursements may be lumped with other types of insurance payments. In the case of unemployment insurance, it may be difficult to separate claims attributable to the disaster and claims that would have arisen as a consequence of typical business and economic cycles. Limited available sources of data and the often high cost of primary data collection have led to attempts to measure indirect losses using statistical models of the type that have long been utilized for economic forecasting and economic impact analysis. A modeling approach is also potentially able to project expected future outcomes over a period of years, and estimate indirect losses associated with a particular actual event. The forward-looking capability is critical for developing simulation models for planning mitigation and emergency responses. Recent studies evaluating model-based estimates suggest that the models designed for traditional economic forecasting and impact analysis do not accurately estimate indirect effects that occur in the wake of a natural disaster. These models must be substantially revised in order to be reliable in estimating indirect effects. Prospects of their long-run cost-effectiveness compared with primary data collection helps justify the research and testing necessary to make the needed revisions. This chapter identifies types of indirect effects and critiques current methods of measuring indirect losses, particularly existing modeling methodologies. It also describes ways in which models can be more usefully employed to generate reasonable estimates of indirect losses. Our recommendations cumulatively constitute an agenda that addresses the current

OCR for page 35
--> lack of information on indirect effects of natural disasters. As we believe that the Bureau of Economic Analysis could best assume responsibility for the collection of data on direct losses, we conclude that the BEA should also be charged with implementing the recommendations outlined in this chapter. Many of these recommendations call for new studies, surveys, and research. Our knowledge of the losses exacted by natural disasters in the United States is fragmented and woefully incomplete. If significant strides are to be made in reducing those losses, sufficient funding will be required to support the studies and research described in this report. Types of Indirect Losses In the short-term, disasters can produce indirect losses and gains. Losses include: Induced losses in sales, wages, and/or profits due to loss of function. The inability to operate may derive from either direct physical damage to commercial structures or from infrastructure failure. Input/output losses to firms forward-linked or backward-linked in production to businesses closed as a result of direct physical damage or infrastructure failure. Slowdowns or shutdowns are induced by reductions in demands for inputs and supplies of outputs from damaged firms. Spending reductions from the income losses triggered by firm closures or cutbacks—so-called multiplier, or ripple, effects. Employees of the firms experiencing reduced production and sales suffer income losses and subsequently curtail their own expenditures, initiating a new round of firm cutbacks. In addition, disasters may generate short-term gains from: Changes in future production, employment, and income and/or changes in these flows outside the damaged area (and the ripple effects thereof). Current production outside the immediate area of impact or future production within the affected region may compensate for initial disaster-induced losses. Income gains outside the impact area to owners of commodities inflated in price by disaster-induced shortages. Both agricultural commodities lost in a disaster and construction materials demanded during reconstruction are particularly likely to generate these windfall profits outside the region. Positive economic stimuli of jobs and production generated from cleaning up and rebuilding and the multiplier effect of those increases.

OCR for page 35
--> Disasters also have longer-term indirect impacts: altered migration flows, changes in development and housing values resulting from changes in insurance costs, reduced consumption (if borrowing occurred to repair and replace damaged structures and goods), and altered government expenditures that derive from new patterns of migration and development. From a very broad temporal and spatial perspective, the net indirect economic impacts of disasters may be zero1. Though, this may seem counter-intuitive, measured over the entire economy, the negative and positive effects may cancel out. Still, precisely because the winners and losers are different groups of individuals and businesses, redistributional indirect impacts of disasters are not zero. These are three key reasons for identifying and measuring indirect impacts of disasters: (1) to inform plans for assistance to disaster victims; (2) to value mitigation measures; and, (3) to plan emergency response programs. Indirect losses of concern to (1) and (3) are losses that occur in the immediate region of impact near the time of the event. To the extent that mitigation costs are to be borne primarily by persons and firms in the immediate area of potential impact, then region-specific net loss savings are the pertinent impacts. Even if the mitigation is federally funded, region-specific savings may still be more relevant than total savings. Assuming federal aid to immediate victims continues, there is a legitimate societal interest in preventing those immediate losses. The valuation of mitigation measures should logically include long-run regional impacts (like the delayed responses to nationwide drought in the 1930s and 1950s), but the substantial passage of time between disaster and impact renders measurement of these phenomena particularly formidable. RECOMMENDATION 3–1: Measurement, study and modeling of indirect losses of natural disasters should concentrate on those losses that occur in the region of impact near the time of the event. The geographic boundaries and the time horizon over which the measurement of indirect losses should occur need to be defined and standardized. The remainder of this chapter expands upon certain themes included in Recommendation 3–1. Current Methods of Measuring Indirect Losses Two methods of ex post measurement of indirect flow losses have been identified by Brookshire et al. (1997). The first relies upon surveys of businesses and households (primary data), and the second utilizes secondary data such as 1   More detailed descriptions of types of indirect flow impacts and examples of offsetting effects are presented in West, 1996.

OCR for page 35
--> tabulations of insurance claims, small business loans, and other forms of disaster relief. In fact, however, there are no mechanisms to systematically ensure that surveys are conducted, nor is there a standard survey format. As a result, no data base exists that would allow the calibration of sophisticated simulation models of indirect losses to study low-probability events with potentially very large indirect losses. The lack of such a data base has in turn inhibited development of simple, rule-of-thumb relationships that might permit efficient estimation of indirect losses for many purposes. RECOMMENDATION 3-2: The agency charged with collecting the direct loss data should commission surveys for the collection of detailed indirect economic loss data from recent disasters, and establish a program for consistently collecting such data on future disasters until a secondary methodology for ''standard" disasters can be validated (see Recommendation 3-3). Once an adequate indirect loss data base is established, such survey data should be collected for all future disasters that have initial total losses (based on model projections) that exceed $10 billion (in 1998 dollars). Because survey data collection is relatively expensive, even when the survey has a narrowly limited time and location, it would be desirable to develop a method based on secondary data to use in ex post estimation of indirect losses for most natural disasters. Initially, however, it is necessary to build a data bank of estimated losses from primary survey data to validate the indirect methods. Once an adequate data base has been established, continued surveys of major events are essential to better understand the significance of indirect losses in larger disasters. RECOMMENDATION 3-3: A study to validate alternative techniques for estimating indirect losses from secondary data should be conducted. This study would test proposed methods using the primary survey data collected pursuant to Recommendation 3-2. Recommendations 3-2 and 3-3 address necessary measurement of indirect losses after natural disasters. However, ex post measurement by itself does not directly address the three primary purposes noted above for quantifying indirect effects. Determining appropriate amounts of resources for victims of disasters cannot wait until after a disaster (when a survey might be conducted). Valuing mitigation requires estimation of expected loss savings over time. Measurement of actual losses from one particular event contributes only limited information for that purpose. Finally, planning emergency response necessarily must precede a disaster. These ex ante approaches require a modeling methodology that permits forecasting (or simulation) of indirect losses. Standard regional economic forecasting or impact models have been used to "predict" indirect losses of

OCR for page 35
--> natural disasters. These include input-output (I-O) impact models (e.g., Rose and Benavides, 1997; NIBS, 1997; Boisevert, 1992; Cochrane, 1997), computable general equilibrium models (Brookshire and McKee, 1992; Boisevert, 1995), and simultaneous equation econometric models (Ellison et al., 1984; Guimares et al., 1993; West and Lenze, 1994). The evidence to date suggests that such models appear to overstate both indirect regional economic losses from natural disasters and indirect regional economic gains from reconstruction. For example, using historical analogies to other earthquakes, Kimbell and Bolton (1994) estimated that reconstruction following the Northridge earthquake would add 20,000 jobs to the Los Angeles economy over four quarters. This estimate is far below the 270,000 job gain predicted by Cochrane et al. (1996) for the entire rebuilding period (i.e., approximately 3 years) using an input-output methodology. Actual data for the Los Angeles area following the earthquake are not consistent with such a large positive impact (Bolton and Kimbell, 1995). Similarly, West (1996), analyzed regional econometric model simulations of the impact of Hurricane Andrew published in West and Lenze (1994) and concluded they were clearly too high, not by a whole order of magnitude, but perhaps by 70 to 85 percent. The core of the problem with statistically based regional models is that the historical relationships embodied in these models are likely to be disrupted in a natural disaster. Temporary or emergency measures taken after disasters are not characteristic of usual socioeconomic conditions, and are therefore not reflected in the model. Economic resiliency can be expected from changes in historical regional production functions, changes in historical purchase and sale patterns, temporary reassignment of labor from outside the area, increased overtime of labor in shortage, and temporary housing arrangements (such as doubling up with relatives or residing in a hotel). In short, regional economic models have been developed over time primarily to forecast future economic conditions or to estimate the effects of a permanent change (e.g., the opening or closing of a manufacturing plant). The abruptness, impermanence, and often unprecedented intensity of a natural disaster does not fit the event pattern upon which most regional economic models are based. The models are thus inappropriate for simulating natural disaster losses. There has been relatively little analysis on how to modify these models in order to increase their accuracy for disaster loss analysis. Secondary regional data currently available on sales, employment, wages, and income following natural disasters provide an opportunity to test possible model modifications, but this testing has not been systematically undertaken. RECOMMENDATION 3-4: A study to test regional economic model modifications for disaster loss analysis should be conducted. Such a study would utilize secondary regional data currently available on sales, employment, wages, and income following natural disasters.

OCR for page 35
--> This recommendation focuses on efficiently using currently available secondary data, a critical first round of work that logically precedes collection of new primary data for supporting such model-based approaches. Optimal Methods for Measuring Indirect Economic Flow Effects The study suggested in Recommendation 3-4 would provide an important component of a model system for measuring indirect economic flow effects. But to use such a model for planning emergency response and valuing mitigation activities, one needs a microsimulation model to generate a timeline of regional commercial/industrial closures (or cutbacks) that trigger indirect losses. A microsimulation model simulates the behavior of individual units, such as businesses or households. It contains a set of rules that define the behavior of each unit (the rules may be probabilistic). The model provides information on an exogenous event (such as a natural disaster), allows the individual units to respond, and then aggregates the results from those units to estimate the impacts on the economy, the market, or industries and businesses. The model may also provide iterative feedback to the individual units. The type of microsimulation model we envision, focused on specific buildings and structures, has five major components: a regional data base; an event-to-loss mapping capability; an emergency infrastructure repair response algorithm; a private commercial/industrial repair response algorithm; and a residential reconstruction algorithm. Data and techniques available for developing these five components differ, and the need to fill in critical information and methodological gaps leads to the remaining recommendations of this section. The first component of the suggested microsimulation model, a regional data base, should contain a fully geocoded inventory of structures and infrastructure capable of identifying commercial/industrial closures from specific structure and lifeline (roads, water, and electricity) losses. Critical documented aspects of each structure include: typical input-output linkages to the regional economy; potential substitutes from outside the regional economy for traditional regional input-output linkages; infrastructure critical and feasible for bringing employees to work; in the case of retailing and service establishments, infrastructure critical and feasible for bringing customers to other places of business; numbers of employees and wages of employee; and profit income in the region.

OCR for page 35
--> Much of the data on firm location, employment, and wages are currently available from reports already required by the federal-state cooperative program on unemployment insurance. Considerable study has also been done on modeling building and lifeline performance in natural disasters. The missing links are how lifeline losses and building losses determine closures. Kiremidjian et al. (1997) estimated the effects of water system loss on Palo Alto, California, for two scenario events. These studies illustrate the importance of lifeline availability for economic functionality, but there are no comprehensive models that systematically relate these phenomena. Similarly lacking is research on the role of building damage in determining indirect losses. Some survey data on lifeline resiliency have been collected. Several years ago, the Disaster Research Center (DRC) of the University of Delaware conducted an extensive survey of businesses in the Memphis region. Specific questions were asked regarding the degree of dependency on different lifeline services and the amount of time that each business could operate without full service. This information needs to be extended beyond lifelines and to other parts of the United States to better understand regional resiliency to natural disasters. RECOMMENDATION 3-5: A range of businesses in different regions of the United States should be surveyed to determine their resiliency to: (1) physical building damage (including feasibility of short-term relocation); (2) loss of infrastructure and utilities (roads, bridges, electricity, water, gas), and; (3) loss of traditional suppliers and markets. Results should be verified by statistical methods (to examine consistency of results by nature of business and size of business) and engineering methods (to determine process-determined lifeline needs). Such a survey is critical if existing models of physical damages for determining indirect losses are to be effective. Accurate estimation of indirect losses requires clear knowledge of the levels of direct damage and business resiliency. We now consider the remaining features of the suggested microsimulation model. The second module would ideally determine which commercial/industrial closures will occur due to direct damage to buildings, loss of lifelines to businesses, and loss of lifelines critical for transporting employees and/or customers. The model's third element would determine how long these losses of functionality occur. This depends on the timeline for infrastructure repair. The fourth part of the model characterizes private commercial/industrial repair response. Repair time is critical to implementing the ideal model's third and fourth modules (Chang et al., 1996; Shinozuka et al., 1997; Shinozuka et al., 1998). However, this parameter is one of the more uncertain parameters in the modeling process. There are many ways in which a damaged lifeline can be repaired.

OCR for page 35
--> Thus, in the best cases outages may last only a few minutes or hours (Lopez et al., 1994); but in the worst, interruptions can last several months, as after the 1995 Kobe earthquake (Takada and Ueno, 1995). Experience teaches us that such losses are initially, primarily a function of restoration time, but then tend to increase exponentially as restoration time is extended. Similarly, there are very limited data on length of time to recover full use of a commercial/industrial building. The Applied Technology Council (1985) provides heuristic estimates used in NIBS (1997) and Kiremidjian et al. (1997). However, there are no formal simulation models of restoration that parallel simulation models of physical damage. Restoration depends not just on physical damage but also on the capacity of the construction industry and the ability to move needed materials and labor to the disaster area. RECOMMENDATION 3-6: A formal restoration model that utilizes available technical and economic data and is consistent with observations from actual natural disasters should be developed. Development of such a model may well uncover additional data needs. These likely would relate to physical and economic aspects of the construction industry and could be incorporated into questionnaires for existing surveys of construction firms. The second, third, and fourth modules of the suggested microsimulation model jointly determine a timeline of regional commercial/industrial closures (or cutbacks) that trigger indirect losses. Wage and profit income lost from closures are estimated from data in the regional data base. Broader measurement of indirect losses must also include a fifth component, residential reconstruction. This module translates damage to residential structures into a reconstruction/rebuilding profile, which depends upon: (1) residential damage; (2) available funds for rebuilding; (3) any "Jacuzzi®" effects (enhancing the original structure beyond reconstruction); and, (4) decisions to abandon totally destroyed structures and migrate from the region. Given income and reconstruction flows simulated in the microsimulation model, the modified regional impact model can be used to simulate multiplier or ripple effects. Given an actual disaster, the system can forecast indirect losses for purposes of planning regional aid. Equally important, it should be used in the context of a probability distribution of disasters to evaluate mitigation proposals and to improve the efficiency of emergency response efforts. RECOMMENDATION 3-7: Research should be conducted on linking a comprehensive indirect loss model to a probabilistic physical damage catastrophe model, for purposes of evaluating mitigation and improving the efficiency of emergency response programs. In sum, the recommendations outlined in this chapter suggest a mix of primary data collection, more intensive use of available secondary data, and

OCR for page 35
--> development of new modeling techniques that will permit significant, cost-effective improvement in measurement and prediction of indirect losses.