Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

MODELS & METHODS FOR

PROJECT SELECTION
Concepts from Management Science, Finance and
Information Technology
INTERNATIONAL SERIES IN
OPERATIONS RESEARCH & MANAGEMENT SCIENCE
Frederick S. Hillier, Series Editor Stanford University

Weyant, J. / ENERGY AND ENVIRONMENTAL POLICY MODELING


Shanthikumar, J.G. & Sumita, U. / APPLIED PROBABILITY AND STOCHASTIC PROCESSES
Liu, B. & Esogbue, AO. / DECISION CRITERIA AND OPTIMAL INVENTORY PROCESSES
Gal, T., Stewart, T.I, Hanne, T. / MULTICRITERIA DECISION MAKING: Advances in
MCDM Models, Algorithms, Theory, and Applications
Fox, B.L. ! STRATEGIES FOR QUASI-MONTE CARLO
Hall, R.W. / HANDBOOK OF TRANSPORTATION SCIENCE
Grassman, W.K.! COMPUTATIONAL PROBABILITY
Pomerol, J-e. & Barba-Romero, S. / MULTICRITERION DECISION IN MANAGEMENT
Axsater, S. / INVENTORY CONTROL
Wolkowicz, H., Saigal, R., & Vandenberghe, L. / HANDBOOK OF SEMI-DEFINITE
PROGRAMMING: Theory, Algorithms, and Applications
Hobbs, B.F. & Meier, P. / ENERGY DECISIONS AND THE ENVIRONMENT A Guide
to the Use ofMulticriteria Methods
Dar-El, E. / HUMAN LEARNING: From Learning Curves to Learning Organizations
Armstrong, IS. / PRINCIPLES OF FORECASTING: A Handbookfor Researchers and
Practitioners
Balsamo, S., Persone, V., & Onvural, R.I ANALYSIS OF QUEUEING NETWORKS WITH
BLOCKING
Bouyssou, D. et al. / EVALUATION AND DECISION MODELS: A Critical Perspective
Hanne, T. / INTELLIGENT STRATEGIES FOR META MULTIPLE CRITERIA DECISION MAKING
Saaty, T. & Vargas, L. / MODELS, METHODS, CONCEPTS and APPLICATIONS OFTHE
ANALYTIC HIERARCHY PROCESS
Chatterjee, K. & Samuelson, W. / GAME THEORY AND BUSINESS APPLICATIONS
Hobbs, B. et al. / THE NEXT GENERATION OF ELECTRIC POWER UNIT COMMITMENT
MODELS
Vanderbei, R.I / LINEAR PROGRAMMING: Foundations and Extensions, 2nd Ed.
Kimms, A / MATHEMATICAL PROGRAMMING AND FINANCIAL OBJECTIVES FOR
SCHEDULING PROJECTS
Baptiste, P., Le Pape, C. & Nuijten, W. / CONSTRAINT-BASED SCHEDULING
Feinberg, E. & Shwartz, A / HANDBOOK OF MARKOV DECISION PROCESSES: Methods
and Applications
Ramfk, J. & Vlach, M. / GENERALIZED CONCAVITY IN FUZZY OPTIMIZATION
AND DECISION ANALYSIS
Song, J. & Yao, D. / SUPPLY CHAIN STRUCTURES: Coordination, Information and
Optimization
Kozan, E. & Ohuchi, A / OPERATIONS RESEARCH/ MANAGEMENT SCIENCE AT WORK
Bouyssou et al. / AIDING DECISIONS WITH MULTIPLE CRITERIA: Essays in
Honor of Bernard Roy
Cox, Louis Anthony, Jr. / RISK ANALYSIS: Foundations, Models and Methods
Dror, M., L'Ecuyer, P. & Szidarovszky, F.! MODELING UNCERTAINTY: An Examination
of Stochastic Theory, Methods, and Applications
Dokuchaev, N. / DYNAMIC PORTFOLIO STRATEGIES: Quantitative Methods and Empirical Rules
for Incomplete Information
Sarker, R., Mohammadian, M. & Yao, X. / EVOLUTIONARY OPTIMIZATION
Demeulemeester, R. & Herroelen, W. / PROJECT SCHEDULING: A Research Handbook
Gazis, D.C. ! TRAFFIC THEORY
Zhu, J. / QUANTITATIVE MODELS FOR PERFORMANCE EVALUATION AND BENCHMARKING
Ehrgott, M. & Gandibleux, X. / MULTIPLE CRITERIA OPTIMIZATION: State of the Art Annotated
Bibliographical Surveys
Bienstock, D. / Potential Function Methodsfor Approx. Solving Linear Programming Problems
Matsatsinis, N.F. & Siskos, Y. / INTELLIGENT SUPPORT SYSTEMS FOR MARKETING
DECISIONS
Alpern, S. & Gal, S. / THE THEORY OF SEARCH GAMES AND RENDEZVOUS
Hall, R.W./HANDBOOK OF TRANSPORTATION SCIENCE - 2nd Ed.
Glover, F. & Kochenberger, G.A / HANDBOOK OF METAHEURISTICS
MODELS & METHODS FOR
PROJECT SELECTION
Concepts from Management Science, Finance
and Information Technology

by

Samuel B. Graves
Boston College

Jeffrey L. Ringuest
Boston College

with
Andres L. Medaglia

SPRINGER SCIENCE+BUSINESS MEDIA, LLC


Library of Congress Cataloging-in-Publication Data

A C.I.P. Catalogue record for this book is available from the Library of Congress.

Graves, Samuel B. & Ringuest, Jeffrey L. / MODELS & METHODS FOR PROJECT
SELECTION: Concepts /rom Management Science, Finance & Information Technology

ISBN 978-1-4613-5001-9 ISBN 978-1-4615-0280-7 (eBook)


DOI 10.1007/978-1-4615-0280-7

Copyright © 2003 by Springer Science+Business Media New York


Originally published by Kluwer Academic Publishers, New York in 2003
Softcover reprint of the hardcover 1st edition 2003

All rights reserved. No part ofthis work may be reproduced, stored in a retrieval
system, or transmitted in any form or by any means, electronic, mechanical,
photocopying, microfilming, recording, or otherwise, without the written
permission from the Publisher, with the exception of any material supplied
specifically for the purpose of being entered and executed on a computer system,
for exclusive use by the purchaser of the work.

Permission for books published in Europe: permissions@wkap.nl


Permissions for books published in the United States of America: permissions@wkap.com

Printed on acid-free paper.


This book is dedicated to our families for their love and
continuing support.
TABLE OF CONTENTS

DEDICATION v
TABLE OF CONTENTS VB
PREFACE Xl
CHAPTER!
THE LINEAR MULTIOBJECTIVE PROJECT SELECTION
PROBLEM
1.1 Introduction 1
1.2 An Example from the Literature 3
1.3 Towards a More General Multiobjective Formulation 7
1.4 A Second Example 9
1.5 Summary and Conclusions 11
References 15
CHAPTER 2
EVALUATING COMPETING INVESTMENTS
2.1 Introduction 19
2.2 Adjusting for Time Alone 19
2.3 Adjusting for Time and Risk 22
2.4 Conclusions 27
References 30
CHAPTER 3
THE LINEAR PROJECT SELECTION PROBLEM: AN
ALTERNATIVE TO NET PRESENT VALUE
3.1 Introduction 31
3.2 An Example 32
3.3 The Behavioral Implications ofNPV 33
3.4 Multiple Objective Decision Methods 35
3.5 Conclusions 38
References 40
CHAPTER 4
CHOOSING THE BEST SOLUTION IN A PROJECT SELECTION
PROBLEM WITH MULTIPLE OBJECTIVES
4.1 Introduction 41
4.2 Some Early Approaches 42
4.3 A Matching and Grouping Approach 46
4.4 A Stochastic Screening Approach 54
4.5 Conclusions 62
References 64
CHAPTERS
EVALUATING A PORTFOLIO OF PROJECT INVESTMENTS
5.1 Introduction 65
5.2 Examples 67
5.3 Conclusions 74
References 76
CHAPTER 6
CONDITIONAL STOCHASTIC DOMINANCE IN PROJECT
PORTFOLIO SELECTION
6.1 Introduction 77
6.2 The Model 78
6.3 Summary and Conclusions 89
References 93
CHAPTER 7
MEAN-GINI ANALYSIS IN PROJECT SELECTION
7.1 Introduction 95
7.2 The Model 101
7.3 Conclusions 114
References 117
CHAPTER 8
A SAMPLING-BASED METHOD FOR GENERATING
NONDOMINATED SOLUTIONS IN STOCHASTIC MOMP
PROBLEMS
8.1 Introduction 119
8.2 Stochastic, Nondominated Solutions 123
8.3 Sampling Approaches to Solving MOMP Problems 125
8.4 Computational Issues 126
8.5 Summary and Conclusions 134
Appendix 8.1
Example SAS Code 135
References 144
CHAPTER 9
AN INTERACTIVE MULTIOBJECTIVE COMPLEX SEARCH FOR
STOCHASTIC PROBLEMS
9.1 Introduction 147
9.2 Direct Search Methods 149
9.3 Applying Complex Search to Multiobjective Mathema-
tical Programming.Problems 152
9.4 An Example of Multiobjective Complex Search 155
9.5 Conclusions 158
References 160

V111
CHAPTER 10
AN EVOLUTIONARY ALGORITHM FOR PROJECT SELECTION
PROBLEMS BASED ON STOCHASTIC MULTIOBJECTIVE
LINEARLY CONSTRAINED OPTIMIZATION
10.1 Introduction 163
10.2 Stochastic Multiobjective Linearly Constrained
Programs 164
10.3 Multiobjective Evolutionary-Based Algorithm 166
10.4 Computational Examples 174
10.5 Summary and Conclusions 183
Appendix 10.1
Input File for the Algorithm Parameters for the
SMOLCP Example 185
Appendix 10.2
Java Program that Defines the First Objective Function in the
SMOLCP Example 187
References 188

INDEX 191

IX
PREFACE

The project selection problem is one that has been given much atten-
tion in the literature. In the project selection problem the decision maker is
required to allocate limited resources across an available set of projects, for
example, research and development (R&D) projects, information technology
(IT) projects, or other capital spending projects. In choosing which projects
to fund, the decision maker must have some concrete objective in mind, e.g.,
maximization of profit or market share or perhaps minimization of time to
market. And in some cases the decision maker may wish to simultaneously
satisfy more than one of these objectives. Most often, these multiple objec-
tives will be in conflict, resulting in a more complicated decision making task.
The decision maker may be able to partially fund some projects, or
conversely some projects may involve a binary decision of fully funding or
not funding at all. The decision maker may also have to resolve issues of in-
terdependency-that is that the value of funding an additional individual proj-
ect may vary depending upon the success or failure of projects that are already
in the portfolio. The decision maker then must take all these factors into ac-
count in seeking an appropriate project selection model, choosing a method-
ology which evaluates the appropriate objective(s), subject to relevant re-
source constraints as well as constraints relating to projects with binary (full
or none at all) funding restrictions.
There is a considerable body of literature describing an abundant va-
riety of models designed for the project selection problem. For our purposes
here, the literature can be broken down into two main streams: that which we
will label the traditional management science stream and that which we will
call the financial modeling stream. The first stream, the management science
literature, derives largely from mathematical programming treatments along
with some use of classical decision theory. In order to use these approaches it
is usually assumed that the existing decision alternatives (projects) are rea-
sonably well-known and that the necessary information for modeling these
alternatives is at hand at the initiation of the planning process. The majority
of the management science models treat the decision process of choosing a set
of new projects to form a wholly new portfolio. But some of the models we
will present also address the problem of adding one or more new projects to
an already existing project portfolio. Most of the research in this body of lit-
erature is confined to decisions which are made at one point in time, that is,
the models are static in the sense that they represent a one-time decision to
assemble or analyze a given portfolio.
An important junction in the decision making process occurs when
the decision maker chooses the appropriate objective(s). If a single objective
(e.g., market share) is chosen, then the problem may be handled with ordinary
mathematical programming techniques that have been used in project plan-
ning models for some time now. If, however, the decision maker wishes to
pursue several objectives simultaneously (e.g., maximization of revenue in
each of several future time periods), some form of multiobjective program-
ming will be needed. It is our belief that this multiobjective case is the more
realistic one, thus, in this book, we will show several applications of multiob-
jective programming to the project portfolio problem.
A key assumption of the mathematical programming models above is
that all relevant information about the projects is known. However, this may
not always be true. Some allocation decisions must be made in the presence
of uncertainty. Uncertainty may exist concerning the ultimate result of a proj-
ect (e.g., the amount of revenue) or the success or failure likelihood may be
known only as a probability distribution. Uncertainty may be represented by
probability distributions around the coefficients in the objective function or in
the constraints. In this book we will illustrate treatments for each of the above
forms of uncertainty. We will, however, assume that adequate information is
available to represent these projects in the model. The required information
may be in the form, for example, of a probability of project success or a prob-
ability distribution around a coefficient in the objective function (e.g. project
return).
When we are dealing with uncertainty and multiple objectives, we
may need to resort to the use of stochastic dominance criteria to screen a set
of solutions. Stochastic dominance is appropriate for all probability distribu-
tions and is minimally restrictive with respect to thedecision maker's utility
function. In this monograph there are several forms of stochastic dominance,
which are of interest. First order stochastic dominance simply compares the
cumulative distribution functions for two projects and makes the choice on
this basis alone. The first order criterion is applicable to all decision makers
with monotone utility functions; that is, decision makers who prefer higher
returns to lower ones and/or those who prefer less risk to more risk. In some
instances, the first order criterion does not yield an unambiguous choice. In
these cases it may be necessary to resort to second order stochastic
dominance. The decision calculus here is based on the area between the two
cumulative distribution curves. This second-order criterion is appropriate for
a narrower class of decision makers, those who are risk-averse. We will also
in some cases apply a conditional stochastic dominance criterion. Conditional
stochastic dominance analysis identifies dominant and nondominant projects
conditioned on the projects, which make up the current portfolio. This
criterion requires no explicit knowledge of the decision maker's utility
function and is applicable to all risk averse decision makers. Finally, in some
cases we will apply a stochastic dominance criteria which compares

xii
alternatives based on the expected value and the probability of achieving
desired levels of one or more measures.
The second main stream of literature that we wish to consider here is
that which derives from financial portfolio research, some of which can be
applied directly to the project selection model. The earliest such methodology
is the mean-variance model, which compares two projects according to their
mean and variance. Any project which has a higher mean return for a given
variance or a lower variance for a given mean will be preferred. The mean-
variance approach is limited in its practical applicability because it involves
rather strict assumptions about the decision maker's risk orientation, and be-
cause it may require a large number of pairwise comparisons if there is a large
number of projects under consideration.
Another source of financial literature is the traditional financial opti-
mization models in which the variance in the portfolio returns is minimized
subject to a constraint on expected return. This approach, however, requires
solution of a non-linear optimization problem (The portfolio variance is non-
linear.) which may be impractical when there are large numbers of projects to
consider. A more recent treatment of the project selection problem deriving
from the financial literature is the mean-Gini approach. This, like the mean-
variance criterion, is a two-parameter method. That is, only two parameters
must be estimated for each R&D project. (Mean and variance for the mean-
variance approach; mean and Gini coefficient for the mean-Gini approach).
The Gini coefficient, like the variance, is a measure of dispersion in outcomes
or investment risk. However, when the Gini is used, as opposed to the vari-
ance, preferred portfolios may be designed based on a simple heuristic. In the
mean-Gini analysis--as will be shown later in this work--there is one impor-
tant difference between the financial application and the project portfolio ap-
plication. In the financial application the necessary probability distributions
for each security are unknown and are estimated from sample (i.e., market)
data. For project portfolio applications the probability distributions (describ-
ing various levels of success) tend to be simple discrete distributions, permit-
ting (in principle) complete enumeration of all possible outcomes in the port-
folio.
Largely in parallel to (and distinct from) the R&D project portfolio
selection literature is a body of work describing Information Technology (IT)
portfolio selection. The objectives of this body of work have much in com-
mon with the R&D portfolio modeling work and we will assume in most of
this work that our models apply equally well to R&D or IT project selection
problems. Essentially the problem here is to find the optimal set (portfolio) of
IT projects when resources are constrained. The greatest difference between
the IT models and the R&D models is the heightened importance of project
interdependencies in the IT models. In IT project applications, as opposed to
R&D project applications, there is, due to the very nature of the projects, an
Xlll
increased incidence of interdependency. For example, two IT projects may
share some identical sections of computer code. They may share as well
hardware such as workstations and networks. And whereas R&D interaction
modeling is typically pairwise, realistic IT modeling requires that higher-or-
der interdependencies (among three or more projects) be represented.
Project selection models that capture the various characteristics de-
scribed above (interdependencies, uncertainty and the ability to partially fund
or the requirement to either not fund or fully fund projects) may be quite
complicated mathematically. These characteristics can lead to complicated
mathematical programs that include one or more objectives that may be linear
or nonlinear, deterministic or stochastic and with variables that are real, inte-
ger, or binary. Appropriate solution procedures for these complex mathe-
matical programs are also needed. We will present these as well.
This monograph is intended to pull together in a single publication the
latest work in this field. It is not intended as a survey, but rather as a vehicle
for establishing some unity in the field of project selection modeling. The
models presented here rely heavily on mathematical programming but also
draw from decision theory and finance. Our intention is to present models
that are broadly applicable in the project selection context, to describe the as-
sumptions and limitations of these models, and to provide solution method-
ologies appropriate for solving these models. The chapter outline below traces
out the main themes of the book.

CHAPTERS 1-3: CRITERIA FOR CHOICE

Chapters 1-3 investigate the effect of the choice of optimization crite-


ria on the results of the portfolio optimization problem. Chapter 1 lays out
the multiobjective linear programming approach to the project selection
problem. The multiobjective approach is contrasted with the goal program-
ming approach, which had been used in earlier applications. This chapter
shows that the multiobjective formulation of the problem is superior to the
earlier goal programming approach in that the multiobjective technique yields
several nondominated solutions to the problem, in contrast to the single solu-
tion revealed by the goal programming approach. The multiobjective ap-
proach is recommended here as a more general approach that will reveal all
nondominated solutions.
Chapter 2 diverges from the discussion of optimization models for
project selection to introduce a discussion of appropriate methods for adjust-
ing for time and risk in the project selection problem. Projects and their as-
sociated revenue streams typically last for a number of years and all projects
involve some level of risk. In order to compare two projects with different
time profiles, we need methods for adjusting for time and for risk. This
xiv
chapter shows some weaknesses of the traditional net present value (NPV)
calculations (using the discount rate and the risk-adjusted discount rate) and
shows how to use generalized NPV to avoid some of these weaknesses.
Chapter 3 continues the investigation of NPV as a decision criterion
and returns to the discussion of optimization models by showing how mul-
tiobjective linear programming can be used as an improvement over tradi-
tional NPV. The chapter shows that the NPV formulation of the project se-
lection problem is a special case of optimizing a multi-attribute value func-
tion, and that the NPV formulation may impose undesirably strict assumptions
about the decision maker's preferences over time. Multiobjective linear pro-
gramming is then used in place ofNPV to solve the project selection problem.

CHAPTERS 4-7: RISK AND UNCERTAINTY

Chapters 4-7 deal with uncertainty in the project selection problem.


Most of the models developed in this section are based on the assumption that
a probability distribution is known or can be estimated to deal with uncer-
tainty in some parameter of the project selection model. The multiobjective
models used in Chapters 1 and 3 result in a set of nondominated solutions
from which the decision maker must choose a single preferred solution. The
process of making this choice may not be an easy task.
In Chapter 4 several methods are shown for screening nondominated
solutions. One of these which introduces the problem of uncertainty is cov-
ered in detail. That method uses a multiobjective model as in the earlier
chapters and adds the additional complications associated with uncertainty.
We solve the multiobjective portfolio problem (with profit and market share
as objectives), yielding a list of nondominated solutions from which the deci-
sion maker must choose. Next we establish goals, or desired levels of
achievement for each objective. Then, assuming we can estimate the prob-
ability distributions describing each objective, we calculate the probability of
attaining the goals for each of the nondominated solutions.
In Chapters 5 and 6 we again diverge from the discussion of optimi-
zation models to examine the importance of decision context on the appropri-
ate analysis of risk. In Chapter 5 we use concepts from decision theory to
directly address the treatment of risk and uncertainty. We show that the tra-
ditional methods for treating risk tend to introduce a bias into the project se-
lection decision. This bias results from the common practice of analyzing
each project in isolation, rather than considering the risk-reducing effects
which result from aggregation of diverse projects into the same portfolio.
This chapter demonstrates that managers who separately analyze only the next
project on the horizon without considering the risk-mitigating effects of ag-
gregation will tend toward excess timidity in decision making.
xv
Chapter 6 provides a more sophisticated treatment of risk that consid-
ers the decision context and that is applicable to the project selection problem.
Using an approach taken from financial modeling, this chapter shows a prac-
tical method of developing a nondominated portfolio of risky projects based
on the criterion of conditional stochastic dominance. The method is simple
and highly intuitive, requiring only the estimation of two parameters, the ex-
pected return and the Gini coefficient. The chapter demonstrates a successful
application of this technique to a real-world R&D portfolio and shows that it
is a practical method for screening large numbers of candidate portfolios to
discover those which are nondominated.
Chapter 7 continues the use of the Gini. In this chapter, however, we
set up a branch and bound heuristic, which is based on the mean return and
the Gini coefficient of each project portfolio. This heuristic produces a set of
solutions (portfolios) which are nondominated in the mean-Gini sense. The
results of this branch and bound heuristic are then plotted with the return on
the vertical axis and the associated Gini value on the horizontal axis. The
points on this graph are a mean-Gini efficient frontier. We then screen the
points to find those which are stochastically nondominated.

CHAPTERS 8-10: NON-LINEARITY AND


INTERDEPENDENCE

These chapters deal with problems of non-linearity and interdepend-


ence as they arise in the project selection problem. The ability to handle non-
linear problems allows the application of the methodology to a far wider
range of problems. Similarly, the ability to model interdependence between
projects, as noted in the discussion ofIT models above, is an important step in
generalization. Chapters 8, 9 and 10 present solution methodologies, which
can be used to solve these most general project selection models.
Chapter 8 presents a method for generating nondominated solutions
for stochastic multiobjective mathematical models of the project selection
problem, which is applicable to both continuous and zero-one variables. The
method is based on the assumption that the objective coefficients are random
variables with probability distributions that are known or can be estimated.
The method shown in this chapter generates solutions that are nondominated
in terms of the expected value of each objective and the probability that each
objective meets or exceeds a specified target value. The method in Chapter 8
is most applicable to integer 0,1 problems and is limited in the real variable
case to problems with relatively few variables. This limitation is addressed in
Chapter 9.

XVI
Chapter 9 shows another approach to the stochastic non-linear prob-
lem, this one having non-linearity in both the objective function and the con-
straints. The method presented in Chapter 8 uses random sampling to identify
a set of feasible solutions for the project selection problem. It is shown that
this random sampling can be computationally burdensome so that the method
presented in Chapter 8 is limited to a small number of real valued variables.
In Chapter 9 complex search is applied to the project selection problem. In
complex search only the initial feasible solution is generated randomly. Then
subsequent solutions are found using a systematic search. In this way the
computational burden is greatly reduced. The example problem shown in
Chapter 9 illustrates a model with a non-linear objective function and several
non-linear constraints, which are solved by relying on the progressive defini-
tion of the decision maker's preferences.
Chapter 10 concludes our treatment of non-linearity and
interdependence in the project selection problem. This chapter presents a new
algorithm that treats the project selection problem in cases of uncertain
objectives, partial funding, and interdependencies in the objectives. The
method shown here is based on a multiobjective evolutionary algorithm and
on concepts from linear programming and presents the decision maker with a
very good approximation of the true efficient frontier. The
algorithm is able to solve project selection problems modeled as
multiobjective linear programs and multiobjective non-linear programs with
linear constraints.
We would like to acknowledge here our indebtedness to individuals
who have contributed to the research in this volume. In particular we want to
acknowledge Randy Case, with whom we have co-authored work in this area
and whose data is used in several places here. We are also indebted to the
many editors and referees who have helped to sharpen and clarify the research
we have performed over the past ten years. Finally, we want to thank
Suzanne Proulx for technical assistance in producing this manuscript.

xvii

You might also like