1-s2.0-s0038012106000280-main
TRANSCRIPT
-
7/29/2019 1-s2.0-S0038012106000280-main
1/15
Socio-Economic Planning Sciences 42 (2008) 3145
A multiobjective model for the selection and timing of public
enterprise projects
Andre s L. Medagliaa, Darrell Huethb,c,, Juan Carlos Mendietab, Jorge A. Sefaira,b
aCentro de Optimizacion y Probabilidad Aplicada (COPA), Departamento de Ingeniera Industrial, Universidad de los Andes;
A.A. 4976, Bogota, DC, ColombiabCentro de Estudios sobre Desarrollo Economico (CEDE), Facultad de Econom a, Universidad de los Andes; A.A. 4976,
Bogota, DC, ColombiacUniversity of Maryland, Department of Agricultural and Resource Economics, 2200 Symons Hall, College Park, MD 20742, USA
Available online 7 September 2006
Abstract
In theory, a public utility company improves the quality of community life through its projects and actions. However,
project selection and prioritization by these companies are highly complex processes. To assist company planning
managers in such processes, we propose a mixed integer programming model that selects, from a bank of projects, which
are worthy of investment. The question of timing is also addressed. The model maximizes a weighted sum of normalized
economic and financial net present values and a social impact index. It simultaneously satisfies a set of precedence relations
among projects, the earliest and latest project start dates, exogenous budget limits, and endogenous project cash flow
generation. We illustrate the models effectiveness using an example constructed from a case study of a major Latin
American water and sewage company.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Project selection; Multicriteria decision-making; Public sector decision-making; Public investment decisions; Mixed integer
programming
1. Introduction
Most public enterprisesincluding water, sanitation, transportation, and energy supply utilitiesface a
common problem: the current budget available for those investment projects that could potentially beundertaken during a planning horizon is insufficient to initiate all projects during the first year. Adding to the
complexity of the decision process are technical limitations such as earliest and latest start dates and
precedence relations between specific projects. Moreover, there may be substantial external political pressure,
and internal bureaucratic support, for specific projects.
ARTICLE IN PRESS
www.elsevier.com/locate/seps
0038-0121/$- see front matterr 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.seps.2006.06.009
Corresponding author. University of Maryland, Department of Agricultural and Resource Economics, 2200 Symons Hall, College
Park, MD 20742, USA.
E-mail address: [email protected] (D. Hueth).
URLs: http://copa.uniandes.edu.co, http://economia.uniandes.edu.co.
http://www.elsevier.com/locate/sepshttp://localhost/var/www/apps/conversion/tmp/scratch_3/dx.doi.org/10.1016/j.seps.2006.06.009mailto:[email protected]:http://copa.uniandes.edu.co,mailto:http://copa.uniandes.edu.co,mailto:[email protected]://localhost/var/www/apps/conversion/tmp/scratch_3/dx.doi.org/10.1016/j.seps.2006.06.009http://www.elsevier.com/locate/seps -
7/29/2019 1-s2.0-S0038012106000280-main
2/15
Two additional considerations are as follows:
(1) Some projectsfor example, those related to potable water supplies and electricitymay be at least
partially self-financing in that consumers can be billed for them; while otherssuch as reforestation of
hillsides and wetland restorationsmust be funded from other sources. Thus, to some degree, the budget
is endogenous.(2) Unlike the private sector, in which maximizing net financial benefits or returns to shareholders is generally
the sole criterion, the public sector must also consider social equity, economic, and political criteria.
The question thus arises as to what types of decision support systems (DSS) are currently available for
public agency managers in such a complex environment. Theoretical and applied research in economics and
operations research has provided substantial guidance in this area.1 In economics, McGuire and Garn [1]
building on the project evaluation work of Eckstein [2], Marglin [3], and othersincorporated equity
considerations into a project selection model by assuming that one can construct weights for each projects net
benefits to each income group, where the weights depend on each groups income and employment levels. In
contrast, Freeman [4] proposed that public administrators past decisions be used to construct a societal
welfare function that would allow project selection based on the marginal social value of each affected groupsincome. However, as Just et al. [5, p. 41] correctly pointed out, little hope [apparently] exists for determining
a societal welfare function on which general agreement can be reached.2 Thus, as explained below, in the
current paper we do not attempt to specify societal preferences but, rather, adopt a weighted set of widely
accepted criteria as an objective function.
The operations research literature has naturally focused more on alternative models and solution
algorithms. For example, Benjamin [6] proposed a goal-programming model for public-sector project selection
in Trinidad and Tobago in which the goals are stipulated by the program manager. This model, which was
applied to the energy sector, included economic, financial, social, and political factors but did not address
project scheduling despite the authors recognition that scheduling is the second phase in public-sector
investment problems. Much earlier, Lee and Sevebeck [7] used goal programming in their aggregative model
for municipal economic planning. The model was applied to a small problem from the municipal government
of Blacksburg, Virginia. More recently, Chan et al. [8] used a goal-seeking methodology within a capital
budgeting framework in considering technology modernization by the US Army.
It is important to note that the approaches taken in these earlier studies all demand that managers specify a
policy by explicitly setting goal levels for several criteria.3
Some researchers have used the analytical hierarchy process (AHP) to help managers identify their
priorities. For example, Barbarosoglu and Phinass [11] project selection tool for the Istanbul Water and
Sewerage Administration used AHP and mixed integer programming (MIP) to include social, political and
economic criteria. After first using AHP to quantify tangible and intangible attributes, and obtain an
aggregate weight for each project, they used the resulting weights in the objective function of the project
scheduling MIP model. Son and Min [12] also combined AHP and integer programming to solve a capital
budgeting problem in the US electrical power industry, taking financial and regulatory (environmental)
constraints into account.Whereas these AHP-based approaches can include hard-to-quantify factors, the methodologys demanding
pairwise comparisons tend to limit the size of the project bank. For instance, the above two experiments
ARTICLE IN PRESS
1The economics literature is now, however, somewhat dated. Economic research on public investment criteria peaked during the era of
large dam construction, which ended in the mid-1970s when the sites for such projects became scarce and environmental concerns brought
a halt to most U.S. Army Corps of Engineers projects.2The search for an acceptable social welfare function for use in public investment decisions has been long and largely unsuccessful. Such
investigation has been characterized by three major approaches: (1) the subjective proposal of a specific analytical form; (2) the axiomatic
construction of a social welfare function from widely accepted axioms; and (3) a moral justice-based approach that distinguishes between
an individuals personal and moral preferences.3Due to go-no-go decisions in some project selection problems, integer goal programming can be viewed as an alternative methodology
[9] to that adopted in the current paper. Extensive reviews of techniques and applications of goal programming are provided by
Schniederjans [9] and Tamiz et al. [10].
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 314532
-
7/29/2019 1-s2.0-S0038012106000280-main
3/15
considered only nine and three projects, respectively. Moreover, the authors made no attempt to quantify the
economic cost of including social, environmental, and political factors.
Regardless the approach, Chan [13] has shown limited adoption by government units of those project
selection and capital budgeting models proposed in the operations research literature. Specifically, he found
that only 25% of 484 Canadian municipal governmental units surveyed used some form of capital budgeting
procedure or payback analysis, a finding echoed in similar research for the US [13]. If such procedures are littleused in these two developed countries, their frequent use in developing countries would seem to be much less
likely. In particular, we have found little evidence of current use of capital budgeting models by Latin
American public utilities. Rather, these companies tend to conduct economic and financial evaluations project
by project, after which agency administrators meet and attempt to reach the greatest possible consensus on
project selection and scheduling in light of any additional information.
The decision-making scenarios outlined above present a number of obvious problems. First, the significant
amount of subjectivity in the scheduling process may allow more powerful department administrators
inordinate influence over outcome(s). For example, Zanakis et al. [14] argued that, in non-profit organizations,
selection criteria could be influenced by political issues to the exclusion of technical considerations. Moreover,
the replacement of technical criteria by intuition in the decision-making process can reduce organizational
competitiveness [15].
Second, a project-by-project assessment may fail to consider a project evaluations dependence on the startdate. That is, some projects whose economic evaluation in the current time period is unfavorable may, as a
result of expected population and income growth, fare much better at a later date. Third, the process includes
no logical incorporation of social considerations and little or no way to estimate the economic opportunity
cost of achieving stated social goals. Finally, and most importantly, in most cases, such a process fails to
provide the decision maker with a current and complete economic efficiency ranking of all projects under
consideration.
To address such problems, this research aims to expand the set of decision-making tools available to public
agencies for project selection and scheduling. Specifically, it provides a multiobjective mixed integer linear
program (MOMILP) that can be embedded in a user-friendly DSS to extend the available project selection
models in a number of directions.
First, rather than attempting to specify a particular social welfare function or specific goal levels, we use themore modest but widely accepted approach of allowing the decision maker to find the economically efficient
solution, and then assess the opportunity cost of achieving social or financial goals. Second, our model considers
precedence relations among projects, including the key feature of positive and negative precedence gapswhich
enables user modeling of soft precedence constraintsseemingly missing from the project selection literature.
Third, the models inclusion of specific earliest and latest start dates simulates the impact of political
decisions. Fourth, the model allows for endogenous cash flow generation in addition to the exogenous budget,
an important feature in that early investment in some projects could generate sufficient resources to partially
fund future projects. Lastly, application of the model in a case study shows it to be computationally efficient
for portfolios containing a large number of projects.
The remainder of the paper is organized as follows: Section 2 presents a general multiobjective optimization
model for public utility project selection and scheduling; Section 3 briefly explains the data requirements, data
sources, and methods of data preparation for the general model; Section 4 presents the results of a series of
model runs using a numerical example built on our experience with the Empresa de Acueducto y
Alcantarillado de Bogota (EAAB) (Bogota Water and Sewer Utility Company);4 and Section 5 summarizes
the work, and suggests future extensions of the proposed model.
2. A general model for optimal selection and timing of projects
In this section, we present a general model that can be applied to any public enterpriseincluding water,
sanitation, transportation, and energy supply utilities. The proposed formulation includes a set of candidate
ARTICLE IN PRESS
4Due to non-disclosure agreements, we are not at liberty to report results of the runs of the EAAB case study. However, the portfolio of
water and sewer projects, on which this paper is based, contained 170 projects.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 3145 33
-
7/29/2019 1-s2.0-S0038012106000280-main
4/15
projects, P; the minimum and maximum number of projects to be included in the portfolio, KL and KU,
respectively; and a set of precedence relations between the projects, A. That is, if project iAPprecedes project
jAP, then (i,j)AA, whose precedence gap is, in turn, represented by gij 2 Z1. If gij 0, the last investment
period of project i should be over before the investment cycle of project jbegins; ifgijo0, an overlap of up to
gij investment periods will be allowed; and, if gij40, the last investment period of project ishould be at least gij
periods away from the first investment period of project j.The life of project i is denoted by vi, and its investment life (i.e., the number of periods up to the last
investment period) by ui. Trepresents the planning horizon for investment decisions t 2 f0; . . . ;Tg, while l1,
l2 and l3 are the a priori weights given by the decision maker to the respective economic, financial, and social
factors, respectively 0pllp1;P3
l1ll 1. The normalized economic, financial, and social indices are,
respectively, I1it, I2it and I
3it if project iAP starts in year t t 2 ft
i ; . . . ; minft
i ;T ui 1gg. Finally, cik and bik
are the investment cost and benefit in period kfor project i k 2 f0; . . . ; vi 1g; ti and t
i are the earliest and
latest respective start dates for project i(iAP); and r0t is the available budget for year t t 2 f0; . . . ; Tg. Without
loss of generality, we assume that r0T 0.
The model identifies those projects to be selected, and the starting date of each. The binary decision
variable, yit, takes a value of 1 if project iAPstarts on year t t 2 fti ; . . . ;minft
i ; T ui 1gg and a value of
0, otherwise. The model requires an auxiliary set of binary decision variables, xikt, that take the value of 1 if
period k k 2 f0; . . . ; vi 1g for project iAP is assigned to year t in the planning horizont 2 fti ; . . . ; t
i vi 1g, and a value of 0, otherwise. Let rt X0 be the available resources in year t
t 2 f0; . . . ; T 1g. We assume, without loss of generality, that r0 0.
The proposed mixed integer program for optimal investment planning is outlined below:
maxX3
l1
ll
X
i2P
Xminfti ;Tui1g
tti
Ilityit, (1)
subject to
Xminfti ;Tui1g
tti
yitp1; i 2 P, (2)
KL X
i2P
Xminfti ;Tui1g
tti
yitpKU, (3)
yit xi;k;tk; i 2 P; k 2 f0; . . . ; vi 1g; t 2 fti ; . . . ; minft
i ;T ui 1gg, (4)
yjtpXtuigij
t0 ti
yit0 ; i;j 2 A; t 2 fti ; . . . ; t
i g, (5)
rt1 rt r0t X
i2P
X
k0;...;vi1
bik cik xikt; t 2 f0; . . . ;Tg, (6)
rtX0; t 2 f0; . . . ;T 1g, (7)
yit 2 0; 1f g; i 2 P; t 2 fti ; . . . ; minft
i ;T ui 1gg, (8)
xikt 2 f0; 1g; i 2 P; k 2 f0; . . . ; vi 1g; t 2 fti ; . . . ; t
i vi 1g. (9)
As shown in (1), the models objective seeks to maximize the weighted sum of the normalized economic,
financial, and social values of the portfolio. The group of constraints in (2) allows the model to select at most
one start date for each project. The bound constraints in (3) allow the number of selected projects to fall
between a lower and an upper bound. If these two bounds are equal, the number of projects to be selected is
ARTICLE IN PRESS
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 314534
-
7/29/2019 1-s2.0-S0038012106000280-main
5/15
fixed. The group of constraints in (4) articulates decision variables y with auxiliary variables x. If a given
project starts at a specified date, the corresponding y variable takes the value of 1. Thus, the periods of that
project are assigned to given dates in the planning horizon through activation of the corresponding x
variables.
The precedence conditions are accounted for in (5). The constraints in (6) are responsible for meeting the
exogenous budget and incorporate reinvestment of endogenous resources generated by other projects. Finally,the group of constraints in (7) establishes non-negativity conditions for the available resources at the end of
year t, while those in (8) and (9) force variables x and y to be binary.
3. Data sources and methodology
This section discusses the input requirements and methodology for the general model presented in the
previous section. The technical data for the projects, including precedence relations, gaps, earliest and latest
start dates, and project lifetimes, are available from the public utility. Data on the exogenous budget and
expected financial cash flows are also available from this source. Implementing the model requires the
construction of three sets of indiceseconomic I1it, financial (I
2it), and social I
3it.
Data for construction of the financial indices of net present values for each project and each possible start
date are readily available from the utility company. In addition, they can usually provide current population
figures, population projections, and current market interest rates.5
Construction of the economic net present value indices requires significantly greater effort than that needed
for the financial indices since private costs and benefits require adjustment to reflect those of society.
Estimation of the economic benefits and costs follows the same general principles and guidelines suggested
by the economic project evaluation literature [1618]. Thus, to begin, values of utility services produced must
be based on the consumers marginal willingness to pay rather than on government regulated utility rates.
Second, the prices of inputs in distorted markets (e.g., subsidies, tariffs and union power) must be adjusted to
their appropriate shadow prices. Third, the economic values of outputs of non-market goods (e.g., ecosystem
service flows and public health benefits) must be estimated following the non-market valuation techniques
found in [1921] and other sources.
Because time and money constraints normally preclude a detailed individual benefit cost analysis for each of
a large set of projects, maximum use of benefits transfer (BT) methodology [19,21] is recommended. In its
simplest form, this methodology, which has seen a rapid evolution in the literature, takes valuation estimation
results from one site and uses them, reasonably calibrated, for the site of that project or policy being
evaluated.6
Perhaps the best known of the growing number of available Internet sites giving BT information is the
Environmental Valuation Reference Inventory (EVRI) maintained by Environment Canada (http://
www.evri.ec.gc.ca/evri/), which provides free access to Canadian citizens and US academics. Unfortunately,
although this source contains a large database on environmental and health benefits, most of the empirical
results are from studies in the US, Canada, and other developed countries. Fortunately, the ENVAL database
created by the Environmental Protection Agency of New South Wales focuses on Australian and SoutheastAsian studies that contain more valuation results from the developing world (http://www2.epa.nsw.gov.au/
envalue/). For an inventory of benefits associated with water projects in particular, the interested researcher is
referred to the University of Californias Beneficial Use Values Database (http://buvd.ucdavis.edu).
The most subjective data we must consider in running our model are those for construction of the social
objective indices, which explicitly recognize equity considerations rather than economic efficiency. Possible
metrics for measuring social impact include: (1) The percentage of beneficiaries below the poverty line; (2) the
impact on income distribution, as measured by the well-known Gini coefficient [23, p. 30] for project location;
ARTICLE IN PRESS
5If the utility does not have census and population projection data, other government agencies will have to be consulted. These sources
may also need to be approached to build the social indices discussed below.6A more sophisticated approach to BT is to perform a meta-analysis [22] (i.e., a statistical regression analysis of a large number of
studies) and estimate the contribution of various variables to the projects benefit values.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 3145 35
http://www.evri.ec.gc.ca/evri/http://www.evri.ec.gc.ca/evri/http://www2.epa.nsw.gov.au/envalue/http://www2.epa.nsw.gov.au/envalue/http://buvd.ucdavis.edu/http://buvd.ucdavis.edu/http://www2.epa.nsw.gov.au/envalue/http://www2.epa.nsw.gov.au/envalue/http://www.evri.ec.gc.ca/evri/http://www.evri.ec.gc.ca/evri/ -
7/29/2019 1-s2.0-S0038012106000280-main
6/15
(3) percentage of the project workforce that is unskilled; and (4) the percentage of project costs paid for by
people with personal income below the poverty line.
The first two of these social impact metrics are likely to be correlated and thus have similar implications;however, the third and fourth indicators can vary from the others and might therefore be considered separate
objectives. In our case study, we measure social impact directly by the number of people below the poverty line
(BPL) who will benefit from the project.
Once construction of the social objective indices is complete, the database of normalized economic and
financial net present values, and the social indices, are ready for use as input to the proposed optimization
model.
4. A numerical example based on a case study
The numerical example provided in this section is based on our experience with the Bogota Water and Sewer
Utility Company (EAAB, for its acronym in Spanish). As mentioned above, because of legal considerationswe are unable to report the actual results of our case study here. The current example, however, demonstrates
typical model outputs as well as how public utility management can use the model for project selection and
scheduling.
We consider a public utility company with 10 projects (p1,y, p10), whose earliest and latest start dates are
shown in Table 1, and with a time horizon of 8 years. As Table 1 illustrates, projects p3, p5, p6, p7 and p9 have
no flexibility with regard to their start dates, whereas the remaining projects do. The yearly exogenous budgets
are given in Table 2. These data are carefully constructed to reflect representative infrastructure and wetlands
restoration projects for EAAB.
The projects precedence relations guarantee that the investment cycle of a given project cannot start before
that of another project. In this example, the investment cycle of project p10 must end before the investment
cycle of project p8 can begin, and projects p1 and p5 must precede project p4. The model allows for gapsbetween projects in the optimal solution.
The construction of the social index vector begins with the data shown in Table 3 on the impacted
population and number of people below the poverty line for each project.
Table 4 presents the economic and financial net present values for each project, assuming each begins on the
first year of the planning horizon.7 It should be noted that the economic and financial performance for each
project varies. For instance, from an economic standpoint, projects p1, p7 and p8 perform poorly but are
financially attractive, while from a social standpoint, the best projects are p1, p6 and p7. Obviously, decision-
making is complicated for the planning manager or analyst facing project selection and scheduling under such
multiple competing criteria.
ARTICLE IN PRESS
Table 1
Earliest and latest start dates
Project ti ti
p1 2004 2011
p2 2004 2011
p3 2004 2004
p4 2006 2011
p5 2005 2005
p6 2005 2005
p7 2004 2004
p8 2004 2007
p9 2005 2005
p10 2004 2006
7It should be noted that since the start date is flexible for a subset of projects (p1, p2, p4, p8 and p10), each possible start date in this
subset will have its own set of financial and economic indices.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 314536
-
7/29/2019 1-s2.0-S0038012106000280-main
7/15
4.1. The optimal portfolio
It should first be noted that to produce the optimal solution to the MOMILP (see Fig. 1), the model uses a
precedence gap of zero (i.e., gij 0, i;j 2 A) between projects having precedence relationships. Not only does
the resulting solution show an ordering that satisfies both the earliest and latest start dates, but the decision
maker has chosen a portfolio containing all 10 projects KL KU 10, when equal weights are given to the
economic, financial, and social criteria (i.e., each l 33.3%).
Fig. 2 shows the budget and optimal investment for the entire planning horizon. It is worth noting that the
model allows for the accumulation of resources, that is, resources that are not invested in a given year are
ARTICLE IN PRESS
Table 2
Exogenous budget
Year Budget US$ millions
2004 2,500
2005 5,000
2006 5,500
2007 5,500
2008 9,000
2009 5,500
2010 6,500
2011 8,500
Table 3
Social performance
Project Population Population below
poverty line
% of population below
poverty line (%BPL)
p1 400,000 49,757 12.44
p2 120,000 12,838 10.70
p3 110,000 10,846 9.86
p4 450,000 54,832 12.18
p5 220,000 24,466 11.12
p6 720,000 91,979 12.77
p7 320,000 43,602 13.63
p8 345,000 36,355 10.54
p9 320,000 28,054 8.77
p10 320,000 28,054 8.77
Table 4Economic, financial, and social project evaluation
Project Economic
NPV
Financial
NPV
%BPL Economic
ranking
Financial
ranking
Social
ranking
Average
ranking
p6 2910.7 555.8 12.77 2 4 2 2.7
p4 382.4 1838.5 12.18 5 1 4 3.3
p7 6843.0 871.5 13.63 10 2 1 4.3
p5 13261.2 84.9 11.12 1 7 5 4.3
p1 143.8 117.2 12.44 7 6 3 5.3
p2 120.9 184.0 10.70 6 5 6 5.7
p8 4967.7 703.3 10.54 9 3 7 6.3
p3 1616.3 556.4 9.86 3 9 8 6.7
p9 1281.5 1308.3 8.77 4 10 10 8.0p10 188.6 247.4 8.77 8 8 9 8.3
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 3145 37
-
7/29/2019 1-s2.0-S0038012106000280-main
8/15
carried over to the next year. This carryover means that investment in a given year can exceed the amount of
exogenous budget allocated for that year.
4.2. Sensitivity to objective weights (ls)
A primary goal in a public utility companys decision-making is the economically efficient use of public
funds. However, pursuing pure economic efficiency may negatively affect financial or social performance. On
the other hand, pursuing a better financial or social performance often implies an efficiency loss, measured by
the decrement in economic net present value resulting from the increased weight assigned to financial and
social factors. Therefore, studying the magnitude of these tradeoffs is of vital importance for the decision
makers in charge of public investment.
As Fig. 3 illustrates, various changes occur when the economic and social importance is varied but the
financial weight (l2 0) remains fixed. If the only objective is to achieve economic efficiency (point A), the
maximum economic net present value of 18 billion dollars is reached by benefiting a small number of people
below the poverty line (272,000). Increasing this number to 344,000 implies a sacrifice of 0.15 billion dollars in
economic efficiency (point B); that is, a cost of 2083 dollars/person. However, benefiting an additional 36,000
people implies an economic efficiency loss of 11 billion dollars (point C), or a per capita cost of 305,556
ARTICLE IN PRESS
Fig. 1. Optimal timing of the 10-project portfolio using a precedence gap of zero.
0
1
2
3
4
5
6
7
8
9
10
2004 2005 2006 2007 2008 2009 2010 2011
US$Millions
Investment Budget
Fig. 2. Budget and optimal investment.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 314538
-
7/29/2019 1-s2.0-S0038012106000280-main
9/15
dollars. The controversial decision of moving from point B to point C may imply a harsh battle between those
interested in economic efficiency and those in favor of social efficiency.
The relation between the economic and financial net present values is shown in Fig. 4 using two different
levels for the social factor. The first level places no value on social performance (l3 0), whereas the second
assigns a weight of 50 percent to the social factor (l3 0.5). At point A (where only economic efficiency
matters), the economic and financial net present values reach 17.8 billion dollars and 75 million dollars,
respectively. At point B (where the financial factor is valued higher), the financial net present value reaches 1.3
billion dollars with an economic efficiency sacrifice of 1.2 billion. Moving from point B to point C implies an
increment in financial net present value of 0.6 billion dollars with an economic loss of 1.6 billion. The most
dramatic loss in economic efficiency occurs in a move from point C to point D, where a gain of 1.2 billion in
financial net present value is paid at a sacrifice of 11.6 billion dollars in economic efficiency. Reducing the
ARTICLE IN PRESS
-
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
20,000
230,000 280,000 330,000 380,000
Millions
Population BPL
Economic
NPV
2=0
(1=1, 3=0)
(1=0, 3=1)
Fig. 3. Tradeoff between economic efficiency and social impact.
-
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
20,000
- 500 1,000 1,500 2,000 2,500 3,000 3,500
Millions
Millions
Financial NPV
E
conomicNPV
3=0 3=0.5
(1=1, 2=0)
(1=0, 2=1)
(1=0, 2=0.5)
A
B
C
D
Fig. 4. Tradeoff between economic efficiency and financial profits.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 3145 39
-
7/29/2019 1-s2.0-S0038012106000280-main
10/15
importance of the economic criterion (i.e., increasing the importance of the social criterion) implies a financial
gain from a loss in economic efficiency, a trend that explains the curvature of the relation, l3 0.5.
The relation between financial and social performance is shown in Fig. 5 with the importance of the
economic factor fixed (l1 0). First, the analysis gives full importance to the financial factor but none to the
social. The maximum financial net present value reaches 3.1 billion dollars, affecting 341,882 people below the
poverty line. Improving the life of 38,900 additional peoplethat is, providing additional public utility
services to 380,782 people below the poverty lineimplies a financial sacrifice of 1.7 billion dollars or 44,971
dollars per person, a price that the public utility will most probably be unwilling to pay.To more realistically explore the tradeoffs as the objective weights (~l) change, we next expand the bank of
projects to 100 and show the full set of efficient solutions for all feasible combinations of~l together with the
portfolios economic and financial values (shown in Figs. 6 and 7, respectively).
Figs. 6 and 7 illustrate the following important findings:
(1) Stepwise surface. Critical values exist for the parameters (~l) such that once they are reached, the optimal
timing and ordering changes significantly. This variation (represented graphically as stairs) is reflected in
abrupt changes in the economic and financial value of the portfolio resulting from small changes in the set
of~l.
(2) Robustness. There exist some stable areas in which changes on the importance of the criteria (economic
and financial) do not affect the optimal timing and ordering of projects. It is useful for the decision maker
to know, that changing the values of~l in these areas (valleys) do not generate changes on the optimal
timing and ordering nor on the economic or financial values of the portfolio. This illustrates how the
model could be used as a negotiation tool in a political environment.
4.3. Sensitivity to technical constraints
By relaxing the constraints on the earliest and latest start dates, projects are free to move within the
planning horizon. Projects find their optimal placement in time according to their best contribution to the
objective function. Due to political pressure, projects are often started near the beginning of the planning
horizon. By advancing the latest start date, the model understands that the project must be accomplished
sooner. Hence, the model looses flexibility by being more constrained, and there is a sacrifice on the value of
the objective function.
ARTICLE IN PRESS
-
500
1,000
1,500
2,000
2,500
3,000
3,500
330,000 340,000 350,000 360,000 370,000 380,000 390,000
Millions
Population BPL
FinancialNPV
1=0
(2=1, 3=0)
(2=0, 3=1)
A
B
Fig. 5. Tradeoff between financial profits and social impact.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 314540
-
7/29/2019 1-s2.0-S0038012106000280-main
11/15
To illustrate this principle using the 10-project example, we free the latest start date for projects p3 and p9
previously, scheduled to begin in 2004 and 2005, respectivelyso they are postponed and scheduled to start in
2010 and 2008, respectively (see Table 1). After this change, the objective function increases in 6.6%. For this
scenario, we set the precedence gap to zero.
ARTICLE IN PRESS
Fig. 6. Normalized economic value surface and contours with varying ~l (where l3 1 l1 l2) for the 100-project example.
Fig. 7. Normalized financial value surface and contours with varying ~l (where l3 1 l1 l2) for the 100-project example.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 3145 41
-
7/29/2019 1-s2.0-S0038012106000280-main
12/15
Finally, as shown by the circle in Fig. 8, changing the precedence gap between projects p5 and p4 (from 0
to 1) allows the investment periods of the two projects to overlap while all other precedence relations remain
definitive (with a gap of 0). Allowing this overlap increases the objective function in 1.7%.
4.4. Sensitivity to budget changes
In this experiment, we assume that the exogenous budget for 2006 is cut dramatically from 5.50 million
dollars to 0.55 million dollars. As Fig. 9 shows, neither the original budget nor the partial cumulative budget
(which picks up the slack of uninvested resources) is able to cope with the aggressive investment plan.
Under a purely economically efficient scenario, allowing endogenous resources translates into a net benefitof 18.06 million, a positive effect of the financial cash flows from earlier projects enabling p2 to start in 2005
rather than being pushed later along the planning horizon (i.e., 2007).
5. Concluding remarks and future research
In this paper, we have introduced a new multiobjective mixed integer programming model for optimal
project selection and scheduling that is especially geared toward public sector companies. This model provides
a much-needed decision support tool for public company planning managers who must decide which projects
to select from an existing project bank and when to invest in them. The proposed model maximizes a weighted
sum of normalized economic and financial net present values and a social impact index (i.e., the percentage of
people below the poverty line). Moreover, it simultaneously satisfies the following important criteria: a set of
precedence relations among projects, earliest and latest project start dates, exogenous budget limits, and
endogenous project cash flow generation.
We have illustrated the model using a case study based on our experience with a major water and sewage
company in Colombia, South America. We conducted a sensitivity analysis that: measured the impact of
changes of the weights on normalized economic and financial net present values, as well as on the social index;
measured the impact of relaxing the earliest and latest start dates on the optimal timing; and measured the
impact of endogenous resource generation in a low-budget scenario on the optimal investment plan. We also
explored the opportunity cost of decisions made because of political pressure, a common occurrence for public
utility companies in Latin America.
The models value as a support tool for public utility company planning managers is evidenced by several
characteristics First, it permits optimal selection and ordering in an ex ante scenario, Second, it not only
allows planning managers to defend the investment strategy that best suits the utilitys goals, but it
ARTICLE IN PRESS
Fig. 8. Effect of changing the precedence gaps of projects p5 and p4.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 314542
-
7/29/2019 1-s2.0-S0038012106000280-main
13/15
quantitatively measures the opportunity cost of political decisions. Finally, it greatly facilitates the planning
process of public companies that must invest public resources using a social welfare orientation.The model also provides several avenues for future research. For example, extended models might capture
the essence of different utility companies, such as those that want to incorporate risk into their decisions. The
authors of this paper have developed project selection models for the private sector that incorporate risk
[2426], much work on this aspect remains to be done for the public sector.
One characteristic of this model is that it takes all prices as exogenous for each project. In some settings, the
size of utility company projects can be sufficiently large to affect input or output prices and in these cases we
would need to have demand and supply relationships incorporated in the model rather than point estimates of
prices. These considerations would create market interdependencies among projects in addition to the
technical interdependencies we have modeled here. We leave these considerations to further research.
Acknowledgments
The constructive comments of the Editor-in-Chief, Dr. Barnett R. Parker, and two anonymous reviewers
greatly improved the paper. As always, any and all errors are fully attributable to the authors. We thank
Dawn Mazzanti, from Dash Optimization for providing us with access to Xpress-MPs optimization products
under the Academic Partner Program. Finally, we express our appreciation to Diego Rentera, Ne stor Botero,
Oscar Pardo, and the planning team from the Gerencia de Planeamiento y Control at Empresa de Acueducto
y Alcantarillado de Bogota for many helpful comments along the way.
References
[1] McGuire M, Garn H. The integration of equity and efficiency criteria in public project selection. Economic Journal1969;79(136):88293.
[2] Eckstein O. Water resource development: the economics of project evaluation. Cambridge, MA: Harvard University Press; 1958.
[3] Marglin SA. Public investment criteria. Boston: MIT Press; 1967.
[4] Freeman III AM. Income distribution and public investment. American Economic Review 1967;57(8):495508.
[5] Just RJ, Hueth DL, Schmitz A. The economics of public policy: a practical guide to policy and project evaluation. Cheltenham, UK:
Edwin Elgar Press; 2004.
[6] Benjamin C. A linear goal-programming model for public-sector project selection. Journal of the Operational Research Society
1985;36(1):1323.
[7] Lee SM, Sevebeck WR. An aggregative model for municipal economic planning. Policy Sciences 1971;2(2):99115.
[8] Chan Y, Di Salvo JP, Garrambone MW. A goal-seeking approach to capital budgeting. Socio-Economic Planning Sciences
2005;39:16582.
[9] Schniederjans MJ. Goal programming: methodology and applications. Boston: Kluwer Academic Publishers; 1994.
[10] Tamiz M, Jones DF, El-Darzi E. A review of goal programming and its applications. Annals of Operations Research
1995;58(1):3953.
ARTICLE IN PRESS
0
2
4
6
8
10
12
14
2004 2005 2006 2007 2008 2009 2010 2011
US$
Millions
Investment New Budget Partial Cumulative Budget
Fig. 9. New budget and optimal investment without endogenous resources.
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 3145 43
-
7/29/2019 1-s2.0-S0038012106000280-main
14/15
[11] Barbarosoglu G, Phinas D. Capital rationing in the public sector using the analytic hierarchy process. Engineering Economist
1995;40(4):31541.
[12] Son HJ, Min KJ. Capital budgeting process for electric power utilities: an analytic hierarchy process approach. International Journal
of Energy Research 1998;22(7):67181.
[13] Chan YL. Use of capital budgeting techniques and an analytic approach to capital investment decisions in Canadian municipal
governments. Public Budgeting & Finance 2004;24(2):4058.
[14] Zanakis SH, Mandakovic T, Gupta SK, Sahay S, Hong S. A review of program evaluation and fund allocation methods within theservice and government sectors. Socio-Economic Planning Sciences 1995;29(1):5979.
[15] Benli O , Yavuz S. Making project selection decisions: a multi-period capital budgeting problem. International Journal of Industrial
Engineering 2002;9(2):30110.
[16] Castro R, Mokate KM. Evaluacio n econo mica y social de proyectos de inversio n. Bogota , DC: Ediciones Uniandes; 2003.
[17] Fontaine E. Evaluacio n social de proyectos. 12th ed. Santiago de Chile: Ediciones Universidad Cato lica de Chile, Editorial Alfaomega; 1999.
[18] Little IMD, Mirrlees JA. Project appraisal and planning for developing countries. New York: Basic Books; 1974.
[19] Freeman III AM. The measurement of environmental and resource values: theory and methods. Washington, DC: Resources for the
Future; 2003.
[20] Haab TC, McConnell KE. Valuing environmental and natural resources: the econometrics of non-market valuation. New horizons in
environmental economics. Cheltenham, UK: Edward Elgar Press; 2002.
[21] Rosenberger RS, Loomis JB. Benefit transfer of outdoor recreation use values. A technical document supporting the Forest Services
Strategic Plan. US Department of Agriculture, Forest Service, Rocky Mountain Research Station; 2000, available at: http://
www.fs.fed.us/rm. Last accessed on February 13, 2006.
[22] Van Den Bergh JCJM, Button KJ, Nijkamp P, Pepping GC. Meta-analysis in environmental economics. Economics andenvironment. Boston: Kluwer Academic Publishers; 1997.
[23] Sen AK. On economic inequality. Oxford: Clarendon Press; 1997.
[24] Medaglia AL. An evolutionary algorithm for project selection problems based on stochastic multiobjective linearly constrained
optimization. In: Graves SB, Ringuest JL, editors. Models and methods for project selection: concepts from management science,
finance, and information technology. Boston: Kluwer Academic Publishers; 2003.
[25] Medaglia AL, Graves SB, Ringuest JL. Multiobjective evolutionary approach for linearly constrained project selection under
uncertainty. European Journal of Operational Research, in press, available at: http://dx.doi.org/10.1016/j.ejor.2005.03.068.
[26] Sefair JA, Medaglia AL. Towards a model for selection and scheduling of risky projects. In: Bass EJ, editor. Proceedings of the 2005
Systems and Information Engineering Design Symposium; 2005.
Andre s L. Medaglia is Associate Professor of Industrial Engineering, Universidad de los Andes, Bogota , Colombia and director and
founding member of the Centro de Optimizacio n y Probabilidad Aplicada (COPA; the Center for Applied Optimization and Probability).
He received a B.S. in industrial engineering from Universidad Javeriana, Bogota , Colombia, an M.Sc. in industrial engineering from
Universidad de los Andes, and a Ph.D. in operations research from North Carolina State University, Raleigh. Prior to his current position,
Professor Medaglia was an optimization specialist, Supply Chain Center, SAS Institute Inc., Cary, NC. His current research interests
include multiobjective evolutionary optimization and optimization applications to logistics and project selection. Professor Medaglias
research has appeared in Annals of Operations Research, Automation in Construction, European Journal of Operational Research, Fuzzy Sets
and Systems, and Interfaces. He currently serves as Associate Editor of Journal of Industrial and Management Optimization and
International Journal of Management Science and Engineering Management; and as regional VP for Central and South America for the
Institute of Industrial Engineers.
Darrell Hueth is Professor of Resource Economics, University of Maryland, College Park, and Profesor titular, Universidad de Los Andes,
Bogota , Colombia. He holds a B.S. in accounting and M.S. in applied economics, both from Montana State University, Bozeman, and a
Ph.D. in agriculture and resource economics from the University of California, Berkeley. Professor Hueth has held academic
appointments at Oregon State University, University of California (Berkeley), and the University of Rhode Island. His research areas
include environmental and resource economics, project and policy evaluation, and the economics of biotechnology. His most recentpublication is Welfare Economics of Public Policy: A Practical Approach to Project and Policy Evaluation with Richard Just and Andrew
Schmitz (Edward Elgar Publishing, 2004). The predecessor to this book won the Publication of Enduring Quality Award from the
American Agricultural Economics Association (AAEA). Professor Hueths research has appeared in such journals as American Economic
Review, Quarterly Journal of Economics, American Journal of Agricultural Economics, Western Journal of Agricultural Economics,
Economic Inquiry, American Economic Review, and Journal of Agricultural and Resource Economics He has served as co-editor of the
Western Journal of Agricultural Economics.
Juan Carlos Mendieta is Instructor and Research Associate, Centro de Estudios para el Desarrollo Econo mico (Center for Studies in
Economic Development, CEDE), School of Economics, Universidad de los Andes, Bogota , Colombia. He received a B.Sc. in agronomic
engineering from Universidad Nacional Agraria de Nicaragua, Managua, and an M.Sc. in environmental and resources economics from
the Joint Program between the University of Maryland, College Park, and Universidad de los Andes. His research interests include non-
market valuation and transportation economics.
ARTICLE IN PRESS
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 314544
http://www.fs.fed.us/rmhttp://www.fs.fed.us/rmhttp://dx.doi.org/10.1016/j.ejor.2005.03.068http://dx.doi.org/10.1016/j.ejor.2005.03.068http://www.fs.fed.us/rmhttp://www.fs.fed.us/rm -
7/29/2019 1-s2.0-S0038012106000280-main
15/15
Jorge A. Sefair is Research Assistant of Industrial Engineering, Universidad de los Andes, Bogota , Colombia. He received a B.Sc. in
Industrial Engineering from Universidad de los Andes, and is a B.A. student in Economics and M.Sc. student in Industrial Engineering at
the Universidad de los Andes. He is engaged in several research and consultancy projects at the Centro de Estudios para el Desarrollo
Econo mico (CEDE) and the Centro para la Optimizacio n y Probabilidad Aplicada (COPA). His research interests include project
selection problems and optimization applications in industrial engineering.
ARTICLE IN PRESS
A.L. Medaglia et al. / Socio-Economic Planning Sciences 42 (2008) 3145 45