Benefit-Cost Analysis and Risk

Scott Farrow

University of Maryland, Baltimore County

CREATE, University of Southern California

February 27, 2015

Abstract
Motivated by the US Government context with particular attention to the Department of
Homeland Security, this chapter presents issues in meta-model choice and the implementation of
risk empirics for benefit-cost analysis. Many of the meta-methods considered rise from a shared
economic framework but with altered preference or aggregation assumptions. With different
assumptions come different measures and values. The methods evaluated include those of
impact statements akin to Environmental Impact Statements, Economic Impact or Consequence
analyses, standard and advanced decision analysis, and standard and advanced benefit-cost
analysis. If benefit-cost is selected as an appropriate modeling framework, then numerous
subsidiary issues exist and are discussed concerning the conceptualization and implementation of
risk metrics.

Acknowledgments: The author thanks Center on the Risk and Economic Analysis of Terrorism
Events at the University of Southern California for funding and participants at the benefit-cost
workshop at the Toulouse School of Economics for comments. Several people provided the
impetus for this study in including Ali Abbas and Debra Elkins.

1
Introduction

Many government organizations seek risk informed decisions. Evaluating and managing risk is

clearly central to the mission of some agencies such as the Environmental Protection Agency

(EPA) or the Department of Homeland Security (DHS) while risk lurks behind the issues and

mission of other agencies such as the financial viability of Medicare or the management of

offshore oil and gas resources. However, the risk assessment community has tended to separate

itself from management issues by attempting to draw a line between the “scientific” aspects of

risk assessment and the management process.

Managers however, request analytical metrics to inform their decisions which often aggregate,

weight or transform multiple risk metrics from their natural units such a probability and

consequences of fatalities from a terrorist attack or monetized business economic consequences.

The choice of analytical metrics--the raw quantity of risk information, ranked risks, “rational” or

“behavioral” utility metrics, monetized risks, pain indices or others measures--are either

explicitly or implicitly framed by a model. Such models may be “mental” and reside within a

decision-maker or explicitly incorporate a formal structure for decision-making with risk. Each

metric is thus based on explicit or implicit assumptions.

This chapter first compares alternative risk metrics on the basis of their assumptions. No criteria

are used to select a “best” metric (although some biases may appear). The initial sections of this

chapter extend the concept of “meta-choice” analysis (Vining and Boardman, 2006). Those

authors argued that different types of analysis can be characterized along the two dimensions of

2
goal choice and monetization. Meta-choice analysis in their framework then consists of

matching the desired type of analysis with the analytical purpose. This chapter focuses on the

meta-analytic choices surrounding risk. Later sections focus on one type of analysis, benefit-cost

analysis, and more detailed analytic choices to be made about risk for that particular model. The

decision-making context is made more specific by focusing on the US Federal Government and

its Department of Homeland Security (DHS).

Risk Principles and Enterprise Risk Management in the US Government

Guidance or requirements for risk informed decisions appear in numerous sources for the US

Government. Preeminent are laws passed by Congress. Some laws direct that risk in a particular

context be managed to a specific level, for example by requiring flood insurance for all within a

100 year flood plain for those with certain types of mortgages. Other laws may limit the

application of some metrics to inform risk, such as directing EPA to protect the health of

sensitive populations with an adequate margin of safety and to ignore cost in the development of

regulations which have the force of law. Other legislation requires agencies to consider benefits

and costs of risk management choices.

In the absence of clear and direct guidance from Congress, the Executive Office of the President

provides guidelines for agencies managing risk. One recent guideline is the inclusion of

Enterprise Risk Management (ERM) elements in agency annual budget submissions (OMB, A-

11, 2014). This new task in budget preparation can be carried out using methods tailored to each

agency but states that one of the purposes of such analysis is to “Facilitate identification and

3
adoption of opportunities for improvement, including risk management” (OMB, A-11, 2014,

section 270.9). In the context of environmental, health and safety risks, OMB provides

additional guidance while noting that the guidance may also be relevant to other fields (OMB,

2007).

It is useful to consider whether the DHS risk context considered in this chapter falls within the

environmental, health, and safety (ERS) context? Might risks in the DHS domain such as

flooding and biological or nuclear contaminants be considered environmental or health or safety?

Might concern about terrorist attacks on people and the built environment be considered safety

and environmental? Less obvious might be issues about immigration or customs which has some

element of health and safety but other objectives as well. A restricted interpretation of DHS

activities might result in only some activities lying specifically within the EHS context. A

broader interpretation, including the OMB suggestion that the guidance may be relevant to other

fields, could permit the broader interpretation that the OMB risk guidance is relevant to DHS.

What do OMB’s risk principles state about risk management? The two management principles

are stated below1:

1. In making significant risk management decisions, agencies should analyze the

distribution of the risks and the benefits and costs (both direct and indirect,

both quantifiable and non-quantifiable) associated with the selection or

1
Text in italics is a restatement of 1995 principles, while plain type is the 2007 revision which highlights additional
guidance for regulatory purposes.

4
 implementation of risk management strategies. Reasonably feasible risk

management strategies, including regulation, positive and negative economic

incentives, and other ways to encourage behavioral changes to reduce risks

(e.g., information dissemination), should be evaluated. Agencies should

employ the best available scientific, economic and policy analysis, and such

analyses should include explanations of significant assumptions,

uncertainties, and methods of data development.

Agencies should refer to Circular A-4 for updated guidance regarding agency

best practices for regulatory analysis.

2. In choosing among alternative approaches to reducing risk, agencies should

seek to offer the greatest net improvement in total societal welfare, accounting

for a broad range of relevant social and economic considerations such as

equity, quality of life, individual preferences, and the magnitude and

distribution of benefits and costs (both direct and indirect, both quantifiable

and non-quantifiable).

Agencies should refer to Circular A-4 for updated guidance regarding agency

best practices for regulatory analysis. (OMB 2007)

Risk management, as guided by OMB, focuses on risk, benefits and costs (although not all

interpret this to mean benefit-cost analysis as practiced by economists). Further, the objective is

the greatest net improvement in total societal welfare considered in as broad terms as possible.

Additional guidance on regulatory issues is provided in a supplemental publication (OMB, A-4)

and separately on benefit-cost analysis and discounting (OMB, A-94) but the risk guidance is not

5
uniquely focused on regulation, in fact, it specifically calls for non-regulatory approaches such as

incentives and information.

The Department of Homeland Security context

DHS’ many components could be said to be focused on risk and the agency itself was a creation

of Congress shortly after the 9/11 terrorist attacks. DHS is charged with managing a wide range

of natural and human caused risks including those relevant to the Coast Guard, the Federal

Emergency Management Agency, the Transportation Security Agency, Customs and

Immigration, and more.

DHS has regularly prepared risk assessments on an agency wide basis every four years and

numerous assessments on various combinations of risks. A strategic planning and risk office

exists within the office of policy as do other offices which support DHS wide efforts at risk

analysis.

A recent report summarized the task of risk management in the agency as

The safety, security, and resilience of the Nation are threatened by an

array of hazards, including acts of terrorism, malicious activity in

cyberspace, pandemics, manmade accidents, transnational crime, and

natural disasters. At the same time, homeland security organizations must

manage risks associated with workforce management, acquisitions

operations, and project costs.

6
 Collectively, these external and internal risks have the potential to cause

loss of life, injuries, negative psychosocial impact, environmental

degradation, loss of economic activity, reduction of ability to

perform mission essential functions, and loss of confidence in

government capabilities. (DHS, 2011)

Consider the problem faced in trying to synthesize risks across the many components of DHS.

In public documents, DHS has recognized 23 types of strategic risk they seek to manage as

presented in Table 1 (DHS, 2011)2. They have also stated in the Strategic National Risk

Assessment that they are concerned about numerous kinds of consequences falling into six

categories of:

1. loss of life,

2. injuries and illnesses,

3. direct economic costs,

4. social displacement,

5. psychological distress, and

6. environmental impact

2
DHS clearly does not consider this list exhaustive as they state in the same source: “Additional threats and
hazards, such as droughts, heat waves, winter storms, rain storms, and different types of
technological/accidental or human-caused hazards, can also pose a risk to jurisdictions across the country and
should be considered, as appropriate, in preparedness planning. Non-influenza diseases with pandemic
potential and other animal diseases should also be considered. In addition, assessment participants identified a
number of events for possible inclusion in future iterations of the SNRA, including electric grid failure, plant
disease outbreak, and transportation system failure.” (DHS, 2011)

7
while recognizing other or more detailed consequences in other settings (DHS, 2011). To no

surprise, DHS’ problem involves multiple risks with multiple consequences.

Table 1 DHS Threat and Hazard Types: National Strategic Risk Assessment

Threat/Hazard Type

Animal Aircraft as a

Earthquake
Disease Armed
WeaponAssault
Flood
Outbreak Biological

Human Chemical/Biolo
Terrorism

Hurricane
Pandemic Chemical
gical Food
Attack (non-
Space Weather
Outbreak Cyber
food) Attack
Terrorism
Contamination
Tsunami Cyber
against
Attack Attack
Data
(non-
Terrorism
Volcanic Explosives
against
food)
Attack
Eruption Terrorism
Physical
Wildfire Dam Failure
Attack
Infrastructure
Biological Radiological

Chemical
Food Nuclear
Substance

Substance
ContaminatSpill Radiological
Terrorism
Release
Source: DHS, 2011. or
ionRelease Terrorism
Attack

Attack

This context of a variety of risks, consequences and decision-makers will be used to make

specific some of the general implications of choosing different risk models and risk metrics.

8
Risk Management Models and Metrics

In this section, models are distinguished from metrics although some models will be uniquely

associated with some metrics. This approach complements that taken by Vining and Boardman

(2006) who distinguished among models based on their selection of goals (economic efficiency

or not) and monetization (or not) of impacts. That two way breakdown was determined to be too

aggregate for the purposes of this chapter and importantly, Vining and Boardman did not

explicitly address any issues related to risk.

Meta model choice

Models as abstractions from reality are commonly used to inform decision-makers concerned

with risks. Such models occur at least in the political science, public policy, risk analysis,

engineering, economic, and decision-analysis communities.

The models chosen here are not from a comprehensive review of the literature but based on

experience in cross agency practices and academic experience in engineering, public policy, and

economics. Other researchers may wish to add additional models. The models considered here

are:

 Impact analysis as exemplified by a National Environmental Policy Act type of analysis

(CEQ, 2005),

9
  “Standard” Decision Analysis as exemplified by expected utility or an expected outcome

analysis (Keeney and Raiffa, 1976; Clemen and Reilley, 2001; Haimes, 2005).

 “Advanced” Decision Analysis as exemplified by a multi-criteria/multi attribute analysis

(Keeney and Raiffa, 1976; Clemen and Reilley, 2001; Haimes, 2005),

 “Standard” Benefit-Cost Analysis as exemplified by equal utility weighting and expected

values (Boardman, et al., 2011; OMB A-94),

 “Advanced” Benefit-Cost Analysis as exemplified by equity weighting and information

on statistical outcomes (Boardman, et al., 2011; OMB A-4)

 Economic consequence analysis as exemplified by input-output or computable general

equilibrium analysis (Santos, 2006; Rose, 2014)

 Cost Effectiveness Analysis as exemplified by the least cost to achieve a stated objective

(Boardman, et al., 2011l OMB A-94, A-4).

While the lines between some of the analyses can be blurred, the attempt below is to identify

those assumptions, as standardly used, which distinguish the risk management alternative

framings.

Each model is built from conceptual assumptions, implemented with data and quantitative

procedures, and result in one or more performance measures. The performance measures may

integrate several elements, such as expected net present value which aggregates outcomes over

time and reports the probabilistically weighted average, the expected value. The purpose of this

section is to set out the assumptions that distinguish the several models and the standard

10
performance measures. Characteristics of various models as identified by the author are set out

in Appendix Table 1.

Areas in which the models were consistent as generally applied (exceptions may always exist),

included:

1) Ability to define who has standing (and so models could be consistent),

2) That individual preferences were generally modeled as independent on one another,

The distinguishing characteristics among the models, some of which are bundled together in the

discussion to follow for ease of presentation, are: a) whose preferences are being modeled, b)

the goal and associated metric, c) the method of aggregation, including weights if relevant, and

d) risk measures and preferences. Each is discussed briefly below with a focus on those models

with the most divergent approaches.

1. Whose ultimate preferences are modeled? This assumption distinguishes most forms of

decision analysis from benefit-cost analysis. In decision analysis, the analyst is typically

modeling the preferences of one or a few people who are generally considered to be

taking the role of a decision-maker, stakeholder, or representative consumer In contrast,

benefit-cost analysis models the preferences of those in society who have standing. The

question of whose preferences are being modeled is ambiguous for a NEPA type impact

analysis.

2. The goal and associated metrics. The goal or objective function of some model types is

clear while others are more ambiguous. A NEPA type analysis has no unifying objective

11
 function and reports metrics in their natural units (illnesses, crimes, health outcomes,

dollars). Impact analysis typically reports Gross Domestic Product but also dimensions

such as unemployment without a unifying objective. Decision-analysis typically uses

monetary or utility measures for the outcomes with the goal the maximization of the

common metric. The outcomes may be decomposed into attributes. Benefit-cost analysis

aspires to use surplus measures (measures of value taking into account different valuation

for the infra-marginal units) but standard benefit-cost analysis typically applies an

average value which may or may not be based on concepts of surplus. The goal is to

maximize total surplus (net benefits) which economists define as (economic) efficiency.

Advanced benefit-cost analyses often take into account measures of surplus possibly

including option price or other variations involving risk preferences. Regarding

descriptive or other statistics used, the expected value of the preferred measure is

typically reported although increasingly, more information about the statistical

distribution of the risk metric is presented.

3. Method of aggregation, including aggregation weights: There are often two dimensions

of aggregation: 1) for one individual, aggregating across impacts if multiple impacts

exist or 2) aggregating across individuals. Here the assumption is that social impacts,

broadly based, are the object of estimation in contrast to impacts on a select group such

as a select group of consumers, firms, or the government alone. In an impact analysis,

individual impacts such as new jobs are typically aggregated equally across individuals

for impacts in a particular category but no attempt is made to aggregate across impact

categories, as across jobs and environmental impacts. Economic impact, or consequence

analyses, are defined in monetary terms and typically aggregate equally within and across

12
 economic impact categories. With decision analysis, numerous means exist to develop

weights to aggregate impacts for one individual but the standard application is without

aggregation of individual preferences. Sometimes, if a group of people are analyzed then

various aggregates such a mean, median, or minimum distance measure might be

reported but are seldom interpreted to represent the aggregate of society. An exception is

the aggregation of expected monetary values which may be little distinguished from a

benefit-cost analysis based on expected values. Standard benefit-cost analysis uses

monetary values to aggregate both for an individual and across individuals. The latter

typically assumes an equal social and individual value per incremental dollar to each

member of society. Advanced benefit-cost analyses will sometimes weight dollar

impacts to different groups in the aggregation process, often though not necessarily based

on different income groups. (This difference can be thought of as adjusting

simultaneously for societal preferences across the different groups and for individual

differences in the incremental value of income.)

4. Risk Preferences: Risk preferences are here taken to be any of risk neutral, risk loving,

risk averse, or some behavioral risk approach (Eeckhoudt et al., 2005). Such preferences

can be revealed at two stages of an analysis. The first are the risk preferences of

individuals and the second are the risk preferences in any social aggregation. Impact

analysis typically ignores such risk preferences. Some advanced types of economic

consequences analysis may include some behavioral risk effect, but the majority of

economic impact analyses assume a risk neutral approach on the part of both individuals

and in the aggregation process. Consequently the standard risk metric is an expected

value impact, often an expenditure or income. Basic decision analysis often assumes

13
 risk neutrality by analyzing the expected value of a decision. Advanced decision

analyses can consider alternative risk preferences as measured by utility, and then may

aggregate using expected or a non-expected utility approach. Standard benefit-cost

analysis is carried out in the world of certainty for individual outcomes and the

aggregation of individual values is based on the risk neutrality, or expected value, of both

the individual and any Governmental decision-maker. Advanced benefit-cost analyses

may take into account risk preferences in assessing the valuation of individuals, and then

typically although not necessarily present the expected value of such valuations (Arrow

and Lind, 1970; Keeney and Raiffa, 1976; Haimes, 2005).

An assumption which tends to but need not vary across model types includes the treatment of

impacts in different time period which is usually aggregated into a present value metric by

discounting. Benefit-Cost analysis routinely uses discounting of future values or annualization

of values as do many decision-analyses. Discounting may be used less commonly in other

analyses.

These common and disparate dimensions among the models extends the two way monetization

and goal dimensions of Vining and Boardman (2006). The Vining and Boardman choice

problem is graphed is Figure 1 where various choices in the logic model, starting on the left, lead

the analyst to a chosen model on the right hand side and its associated metric. For example, the

sequence of the goal of efficiency and monetization leads to benefit-cost and net present value.

14
Figure 1: Meta-choice logic model of Vining and Boardman

Model choices involving aggregation and weighting refine the Vining and Boardman meta-

analysis. As the left hand side of Figure1 remains the same, Figure 2 highlights additional

choices beyond monetization which lead to finer characterization among the models.

15
Figure 2: Extended meta-choice including aggregation and risk preferences

The distinction among models can be investigated in greater detail should one desire. For

instance, Adler (2012) has developed a related sequence, reproduced as Figure 3, for standard

and non-standard methods of aggregation based on which of several axioms are met by social

welfare functions. The “tree” sequence however is not unique in the sense that some re-ordering

of the axioms (boxes) can lead to a different representation.

16
Figure 3: Axioms and social welfare functions

Benefit-Cost Analysis and the Implementation of Risk Concepts

This section investigates the specific ways in which risk and uncertainty affect the choice of

what is to be measured--the conceptual metric--and the numerical estimate. The conceptual

issues are briefly highlighted from above focusing on benefit-cost analysis and while more

detail is provided on empirical measurement.

17
Risk and BCA

Risk appears in numerous guises in the design and implementation of a benefit-cost analysis

(Farrow and Viscusi, 2011). Risk is here interpreted broadly to include both known and

unknown statistical distributions and the several ways these concepts reveal themselves in

assumptions about metrics to be used and model specification. For instance, an incorrectly

applied constraint that cross price elasticities are zero among some markets introduces a kind

of degenerate or deterministic distribution on a value which may be incorrect.

This section separates levels of benefit-cost analysis into deterministic BCA (DBCA) as that

standardly used in practice and BCA incorporating risk (RBCA) as a more advanced

version3. As will be shown, numerous issues which may appear relevant only in an RBCA

are in fact implicit in the assumptions leading to the restricted DBCA.

First, define a standard DBCA, with Consumer Surplus (CS) and Producer Surplus (PS) as

measures net of payments to government, GR as government revenue and EX as externalities

and other symbols as standardly used. The DBCA objective function, which is to be

maximized by the choice among alternatives or scale, is then

𝐵 −𝐶 𝐶𝑆𝑡 +𝑃𝑆𝑡 +𝐺𝑅𝑡 +𝐸𝑋𝑡

(1) 𝑃𝑟𝑒𝑠𝑒𝑛𝑡 𝑉𝑎𝑙𝑢𝑒 𝑁𝑒𝑡 𝐵𝑒𝑛𝑒𝑓𝑖𝑡 = ∑𝑡= 𝑡 𝑡 𝑡=
𝑡=0 (1+𝑟)𝑡 = ∑𝑡=0 (1+𝑟)𝑡

3
Standard assumptions for a DBCA are in Appendix 2, see also Boardman, et al. (2011).

18
 A standard RBCA objective function is defined as below where πi is probability (here

assumed constant over time, although this is not necessary), D(t) is a discount function, CSR,

PSR and ER are measures which may include risk adjusted values and εit is an additive error

which may vary by outcome or time:

𝐶𝑆𝑅𝑖𝑡 +𝑃𝑆𝑅𝑖𝑡 +𝐺𝑅𝑖𝑡 +𝐸𝑅𝑥𝑖𝑡

(2) 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑃𝑟𝑒𝑠𝑒𝑛𝑡 𝑉𝑎𝑙𝑢𝑒 𝑁𝑒𝑡 𝐵𝑒𝑛𝑒𝑓𝑖𝑡 = ∑𝑛𝑖=1 ∑𝑡=
𝑡=0 𝜋𝑖 ( ) + 𝜖𝑖𝑡
𝐷(𝑡)

Each of these terms and more that are implicit will be discussed for the way in which risk

affects their definition or measurement. The discussions will be arbitrarily divided into the

following categories:

 Meta-choice assumptions within BCA, addressing specification issues related to

means of aggregation and the extent of markets

 Metrics, addressing issues of decision criteria and risk based preferences including

discounting

 Randomness addressing issues of the probability of events, variability in parameters,

random error.

Meta-choice and specification within Benefit-Cost Analysis

The first part of this chapter discussed the meta-choices between major classes of decision-

support models. Within a given analytical choice, such as benefit-cost analysis, there remain

19
several large specification or framing issues for the analysis. A “wrong” assumption on these

issues can create modeling error, or in statistical terms, impose a constraint or degenerate

distribution on the analysis. While assumptions are central to abstraction, they also introduce

specification risk to the extent they are incorrect. Standard macro level framing specification

assumptions for benefit-cost analysis are discussed below:

 General versus partial equilibrium: This choices defines the scope of the terms in the

numerator of equations 1 or 2. Benefit-cost analysis may be carried out in either a

general or partial equilibrium setting (Florio, 2013; Goulder and Williams, 2003).

Choosing a partial equilibrium specification imposes zero cross price elasticity

constraints on economic interactions while simplifying the analysis. Alternatively, if one

carries out a general equilibrium analysis the functional specifications and parameters

may not be accurate; the difficult question is which is relatively more accurate compared

to the “true” model. A specification risk occurs with either choice.

 Aggregation across economic actors: This choice defines the weighting of impacts in the

numerator of equations 1 and 2. The default aggregation behind equations 1 and 2 is

equal and independent social weighting for each individual. This can be explained as

equal social weighting and equal marginal utility of income for all participants; or if one

assumes diminishing marginal utility of income, then equal weighting assumes society

more heavily weights impacts on the wealthy (Nyborg, 2012). The most frequent

alternative to equal weighting subjectively applies heavier weights to impacts on lower

income or other identified populations (e.g. Loomis, 2013). Other weighting approaches

can but seldom do include interdependent weights. A specification risk occurs with any

aggregation choice for the social welfare function.

20
Metrics

Decision criteria metrics

Performance metrics for benefit-cost analysis are generally inferred from criteria for

optimization. The standard optimization problems involve a single social welfare optimizer and

implicitly or explicitly includes the aggregation function (social welfare function) and the risk

perspective of the decision-maker.

 Deterministic benefit-cost analysis seeks to maximize equation 1 with the resulting

optimal solution being the largest present value of net benefits among mutually exclusive

alternative considered. Several variations of this problem exist although often ignored in

practice (Bellinger, 2007). One variation is when the scale of the project can be chosen

such that some benefits and costs vary with scale. In that case the decision criteria

equates the marginal (social) benefit with the marginal social cost or, if there is a budget

constraint, the net marginal benefits may be required to exceed a threshold when there are

multiple projects. Few organizations are without budget constraints which can lead to a

decision metric requiring that a critical ratio be exceeded, in this case the ratio of benefits

to costs must exceed a threshold (Weinstein and Zeckhauser, 1973; Bellinger, 2007).

 Risked benefit-cost analysis (RBCA) introduces numerous alternatives for the

performance metric based on equation 2. The standard approach is that a social decision-

maker who can pool risk across many projects will choose to maximize the expected net

present value (Arrow and Lind, 1970) as given by equation 2, either omitting the error

term or assuming that its expectation is zero. This metric may be used whether or not

21
 risk based valuation has been used for the component measures, such as consumer or

producer surplus. The increasingly frequent practice of reporting the statistical

distribution of net present value and letting a decision-maker choose an appropriate

option or level of stringency implies that a decision-maker will use their own risk

preferences, however created, to make a selection which may not be based on expected

value. Concern for dynamic uncertainty when there are irreversible commitments, as

with unknown prices but fixed long term investment, leads to decision criteria in which

there is a value of information which may be informed by waiting. The standard model

in such settings is stochastic, expected value optimization in which net benefits may need

to exceed a positive threshold, the option value of waiting, in order to take action (Dixit

and Pindyck, 1994).

Preferences and Behavioral Response metrics

Preferences and resulting behavioral responses are the core of benefit-cost analysis. The

monetized meta-choice implied in benefit-cost analysis means that values, sometimes revealed

through choices or surveys, are measured in monetary terms. The literature on the monetization

of values implicit in choices is vast, some of it under topics such as welfare economics (e.g. Just

et al., 2008) or benefit-cost analysis (Zerbe, 2008). Preferences and behavioral responses may be

modeled as rational, often using an expected utility approach (Boardman, et al., 2011). Much

current research is focused on behavioral economics involving an explicit or implicit non-

expected utility approach where responses may be consistent but not be rational (Machina, 1987;

DellaVigna, 2009). Risk preferences can provide an important distinction between DBCA and

22
RBCA. In RBCA and its associated equation 2, consumer preferences are frequently modeled as

risk averse whereby some adjustment to expected cost or benefit is added to the valuation for a

consumer (or producer). While some functional forms are commonly used, there is little

agreement on the appropriate model and empirical parameters of risk for a consumer (Eeckhoudt

et al., 2005). At the same time, for some problems, risk based preference adjustment may or may

not be empirically significant (Farrow, 2014). Although the estimation and monetization of risk

preferences is a research frontier, advances in valuation have led to relatively standard values for

a statistical life in benefit-cost analyses as well as for other risk based outcomes (Boardman, et

al., 2011).

Preferences over time is a long standing dimension of concern. A declining exponential weight

on values in the future (discounted values) has long been standard but behavioral economics and

concern for uncertainty is altering that consensus. Behavioral economic studies have found that

people may overweight (relative to exponential discounting) the near term and also the distant

future (DellaVigna, 2009). Further, debate remains about the impact of incorporating

uncertainty into discount factors although a typical result with uncertainty is that the future is

discounted less than with exponential discounting (Arrow, et al., 2013; Gollier, 2012).

Variability and Randomness

Variability in impacts and valuations are one source of risk in the outcome metric. For instance,

an unconditional mean value for an impact may ignore predictable conditionality of the mean on

socio-economic and other conditioning factors. This is discussed in the literature as the

23
usefulness of a benefit transfer function, which includes adjustments for varying conditions,

instead of an unconditional value (Boyle, et al., 2010). Similarly, if impacts are estimated

separately from values then the impacts may depend on various conditioning factors, such as

distance or income and add further risk.

Fundamental to an RBCA is the estimation of probability, or more accurately, a set of

probabilities and outcomes represented by 𝜋𝑖 and the associated outcomes in equation 2. In
some DHS settings the probability distributions may be reasonably estimated based on historical
data that are sufficient to inform policy. Probabilities of floods of a given size may be such a
case. In other DHS security settings, the estimation of probability distributions is more
problematic although not necessary impossible. Complications may involve the availability of
classified information, the fact that some probabilities may be endogenous (as through game
theory) where the probability of attack may depend on the defenders chosen strategies, and the
infrequent nature of observed outcomes (Farrow, 2007). Some analysts have chosen not to
estimate probabilities but instead estimate a DBCA (or variations thereof) and find a break-even
risk where the DBCA becomes positive as is frequently used in DHS regulatory analyses
(Mueller and Stewart, 2011; Farrow and Shapiro, 2009). In some settings, DHS uses estimates
of probabilities but they are not available to outside researchers. The issue of probability
estimation remains a fundamental challenge for RBCA analysis.

The impact of stochastic randomness or pure error in an RBCA appears in several ways.
Models for components of equation 2 are often build from parameterized models and sometimes
from regression coefficients. Perhaps clearest in the case of regression coefficients, those
coefficients have a statistical distribution based on the random error in the underlying estimation
equation. Such randomness can be included in Monte Carlo type analyses and adds risk to the
outcome measure even if the distribution is symmetric about its mean.

Seldom considered in an RBCA is the degree of model fit associated with a stochastic error term
which may be added as noted in equation 2. For instance, homeland security models may have a
larger random error component that a flood model. Such added variability is a component of the

24
risk of the dependent metric. When Monte Carlo simulation methods are used, analysts may be
able to include a measure of the pure error term in the RBCA if they are willing to place an
estimate on the accuracy of the model (Farrow, 2012).

Conclusion

This chapter emphasizes the ubiquity of risk in benefit-cost modeling. The meta-choice of

benefit-cost analysis itself implies a set of framing assumptions. It is useful to understand the

distinctions between the several major types of descriptive and prescriptive analyses used to

inform homeland security decisions. The model choice can introduce specification error. If a

benefit-cost analysis is chosen as the appropriate analysis, then risk, associated with variability in

outcomes, affects numerous components of the analysis. The analyst must choose whether or not

to: a) use a partial or general equilibrium setting, b) deviate from standard aggregation and

constraint assumptions, c) include conditioning factors which may shift impacts or values, c)

include risk preferences in valuation, d) include uncertainty in discounting, e) use explicit

probability distributions, and f) incorporate the accuracy of the model itself in the variability of

the outcome. Taking all these elements into consideration in detail may lead to the paralysis of

analysis. The challenge to analysts is to incorporate those risk considerations that are central to

the problem without foundering in a morass of analytical uncertainty.

25
Appendix Table 1: Models and Assumptions

Assumptions Impact “Std” Advanced Std Benefit- Advanced Econ. Cost-Effect. Other models,
Analysis Decision Decision Cost Benefit-Cost Consequence Analysis Notes
Analysis Analysis Analysis Analysis Analysis
(eg
MCMA)
Who has Citizens or Citizens or Citizens Citizens or as Citizens or as Citizens or as Citizens or as
standing as otherwise as otherwise or as otherwise otherwise otherwise otherwise
defined defined otherwise defined defined defined defined
defined
Whose Not Decision- Decision- Society Society Not specified Society
preferences specified maker (s) maker (s)
DM
Aggregate Adding up None None or Equal weight Unequal weight Adding up of Equal weight
preferences of natural Some SWF SWF dollar impacts SWF (including
units; group with other quantity
impact may approach dimensions as outcome)
be mental model
distributiona
l info.
Weights None DM or DM “crowd dollar “crowd dollar Dollar Mixture dollar
crowd dollar weighted weighted” , weighted”, weighted by weighted and
weighted surplus surplus; price DM or expert
subjective weighted
distributional
Independent or Independent Independent Independe Independent Independent Independent Independent Taking account of
interdependent nt system risk or
personal interdependent
preferences preferences
Extent of None Limited Usually Usually Usually limited Often large, Usually limited
economic Limited limited but but can be large General Eq. but can be large
interaction but could can be large
across markets be large
Time Unlikely to Constant Constant Discounting Constant or Some static,
discount discounting or no (constant) Variable some
or no discount discounting dynamics/resili
discounting ency

26
 Assumptions Impact “Std” Advanced Std Benefit- Advanced Econ. Cost-Effect. Other models,
Analysis Decision Decision Cost Benefit-Cost Consequence Analysis Notes
Analysis Analysis Analysis Analysis Analysis
(eg
MCMA)
Risk Analysis Impacts in Expected Weighted Monetized Risk preference Impacts in ESPV cost per Movement toward
metric (s) natural units Value cardinal social adjusted ESNPV monetary units unit (issue with estimation of
utility, expected net valued at price multiple statistical
non- present value outcomes) distribution of
expected ESNPV outcomes
utility
Risk Not Risk neutral Risk Risk Neutral Risk pref. of Risk neutral Usually risk Can apply to all of
preference specified or risk neutral or consumers or neutral consumers,
values of risk values firm producers, DM
DM of DM
Objective Not Max EV or Max EU Max expected Max risk Max expected Min Cost
(implicit or specified EU (or non Total surplus adjusted total dollar value,
explicit) EU surplus such as GDP
approache
s)
Risk Decision Decision- Choose Highest Choose Choose largest GDP and Choose least
rules maker or highest EV, expected largest sum of sum of “surplus possibly other cost (benefit
stakeholders EU utility “surplus” as as EV; possible dimensions assumed)
apply EV. to let decision-
mental maker choose
model to
aggregate
Social welfare Not None None Additive, Additive, Additive, Least cost of
function specified unweighted weighted unweighted reaching given
objective (or
most output for
given budget)

Green: all models tend but need not be consistent (for instance, all could apply standing to the same group of people).
Yellow: models vary relatively consistently across model types
Gray: models tend to but need not make different assumptions.

27
Appendix Table 2: Assumptions for deterministic (standard) benefit-cost analysis
Based on the first and second theorems of welfare economics being correct: 1) A PC society will be Pareto Optimal, 2) PO point is not unique.

Topic Assumptions
Consumer preferences and “Rational”
optimum Complete ordering
Transitive
Continuous
Non-satiation
Individual (no inter-personal
utility)
Max utility subject to budget
constraint
Firm optimum Given production function
Input prices
Output price
Max profit
Government activity Responds to market failures
including public goods; access
to either lump sum or optimal
taxation
Nature of goods or services Homogenous goods, no
externalities (see 2nd best), no
taxation that distorts markets.
Market equilibrium Perfectly competitive equi. In all
input and output markets
Time Discount at a constant rate
Risk Standard: none
Advanced: numerous concerns
for risk

28
Topic Assumptions
Scope of analysis (partial or Include those actions where
general equilibrium) dQi/DPj not equal to zero (could
lead to partial or GE analysis)
Social welfare function Additive in individual utility
dW/dUi = a (constant and
uniform)
dUi/dYi= b (marginal utility of
income a constant and uniform)
dUi/dUj = 0, no interpersonal
utility

Typical decision context: Undertake the alternative with

choosing among several the largest PVNB. Alternative
alternatives with no budget decision rules exist as for
constraint constraints or scaling of project.

29
References

Adler, M., Well Being and Fair Distribution, Oxford University Press, Oxford, 2012.

Arrow, K. and R. Lind, Uncertainty and the Evaluation of Public Investment Decisions,

American Economic Review, 60, 364-78, 1970.

Arrow, et al., Determining Benefits and Costs for Future Generations, Science, 341(6144):349-

350, July 26, 2013.

Bellinger, W., The Economic Analysis of Public Policy, Routledge, New York. 2007.

Boardman, A., D. Greenberg, A. Vining and D. Weimer, Cost-Benefit Analysis, 4th ed., Prentice

Hall, 2011.

Boardway, R. and N. Bruce, Welfare Economics. Basil Blackwell, 1984.

Boyle, K., N. Kuminoff, C. Parmeter and J. Pope, The Benefit-Transfer Challenge, Annual

Review of Resource Economics, 2:161-182, 2010.

Chetty, R., Sufficient Statistics for Welfare Analysis: A Bridge Between Structural and Reduced

Form Methods, in Annual Review of Economics, Volume 1, edited by K. Arrow and T.

Bresnahan, Palo Alto: Annuals Reviews, 2009.

Clemen, R.T. and Reilley, T., Making Hard Decisions. Belmont CA: Duxbury Press, 2001.

Council on Environmental Quality (CEQ), Regulations for Implementing the Procedural

Elements of the National Environmental Policy Act.,40 CFR Parts 1500-1508, 2005.

DellaVigna, S., Psychology and Economics: Evidence from the Field, Journal of Economic

Literature, 47(2): 315-72, 2009

Dixit, A. and R. Pindyck, Investment Under Uncertainty, Princeton University Press, Princeton,

1994.
30
Dreze, J. and Stern, N., The Theory of Cost-Benefit Analysis, Chapter 14 in Auerbach, A.J. and

Feldstein, M. (eds) Handbook of Public Economics, Amsterdam: North-Holland and

Elsevier Science Publishers, 1987.

Eeckhoudt, L., C. Gollier, and H. Schlesinger, Economic and Financial Decisions Under Risk,

Princeton University Press, Princeton, 2005.

Farrow, S., The Economics of Homeland Security Expenditures: Foundational Expected Cost-

Effectiveness Approaches, Contemporary Economic Policy, January, pp. 14-26, 2007.

Farrow, S. and S. Shapiro, The Benefit-Cost Analysis of Homeland Security Expenditures,

Journal of Homeland Security and Emergency Management, 6(1):1-20, 2009.

Farrow, S. and W. K. Viscusi, Principles and Standards for the Benefit-Cost Analysis of Safety,

Journal of Benefit-Cost Analysis, 2(3):#5, 2011.

Farrow, S. and R. Zerbe, Principles and Standards for Benefit-Cost Analysis. Edward Elgar,

Cheltenham, UK, 2012.

Farrow, S., A Missing Error Term in Benefit-Cost Analysis, Environmental Science and

Technology, 46(5):2523-2528, 2012.

Farrow, S., A Comparison of Key Benefit Estimation Issues for Natural Hazards and Terrorism:

Ex Ante / Ex Post Valuation and Endogenous Risk, in The Benefit-Cost Analysis of

Security Focused Regulations, ed. C. Mansfield and V.K. Smith. Edward Elgar,

Cheltenham, 2014.

Florio, M., Applied Welfare Economics: Cost-Benefit Analysis of Projects and Policies,

Routledge, 2014.)

Gollier, C., Pricing the Planet’s Future: the Economics of Discounting in an Uncertain World,

Princeton University Press, Princeton, 2012.

31
Goulder, L. and R. Williams, The Substantial Bias from Ignoring General Equilibrium Effects in

Estimating Excess Burden, and a Practical Solution, Journal of Political Economy,

111(4):898-927, 2003.

Haimes, Y., Risk Modeling, Assessment and Management. Wiley, 2005.

Just, R., D. Hueth and A. Schmitz, eds. Applied Welfare Economics, Edward Elgar, Cheltenham,

2008.

Keeney, R. L., & Raiffa, H. (1976). Decision making with multiple objectives preferences and

value tradeoffs. New York: Wiley.

Loomis, J. Incorporating Distributional Issues into Benefit–Cost Analysis, in Principles and

Standards for Benefit-Cost Analysis, ed. S. Farrow and R. Zerbe. Edward Elgar, 2012.

Machina, M. “Choice Under Uncertainty: Problems Solved and Unsolved”, Economic

Perspectives, 1(1):121-154, 1987.

Mueller, J. and M. Stewart, Terror, Security and Money: Balancing the Risks, Benefits and Costs

of Homeland Security, Oxford University Press, Oxford, 2011.

Nyborg, K. (2012). The Ethics and Politics of Environmental Cost-Benefit Analysis. Routledge.

Office of Management and Budget (OMB), Principles for Risk Assessment. Washington, 2007.

Office of Management and Budget (OMB), Circulars A-94, A-4, A-11, Washington, 2013, 2014.

Rose, A. “Macroeconomic Consequence of Terrorist Attacks: Estimation for the Analysis of

Policies and Rules”, in The Benefit-Cost Analysis of Security Focused Regulations, ed. C.

Mansfield and V.K. Smith. Edward Elgar, Cheltenham, 2014.

Santos J. “Inoperability Input-Output Modeling of Disruptions to Interdependent Economic

Systems, Systems Engineering 9(1): 20-34, 2006.

32
US Department of Homeland Security (DHS), Risk Management Fundamentals, Washington,

2011. Available at /www.dhs.gov/xlibrary/assets/rma-risk-management-

fundamentals.pdf .

Vining, A. and A. Boardman (2006) “Metachoice in Policy Analysis”, Journal of Comparative

Policy Analysis, 8(1):77-87.

Weinstein, M. and R. Zeckhauser, “Critical Ratios and Efficient Allocation”, Journal of Public

Economics, 2(2):147-157, 1973.

Zerbe, R., Jr., ed. Benefit-Cost Analysis, Edward Elgar, Cheltenham, 2008.

33