European Journal of Operational Research 107 (1998) 451±469

On the optimal management of project risk

L. Valadares Tavares *, J.A. Antunes Ferreira 1, J. Silva Coelho 1

CESUR-IST, Av. Rovisco Pais, 1000 Lisboa, Portugal

Received 1 July 1996; received in revised form 1 March 1997

Abstract

The uncertainty of project networks has been mainly considered as the randomness of duration of the activities.
However, another major problem for project managers is the uncertainty due to the randomness of the amount of re-
sources required by each activity which can be expressed by the randomness of its cost. Such randomness can seriously
a€ect the discounted cost of the project and it may be strongly correlated with the duration of the activity.In this paper,
a model considering the randomness of both the cost and the duration of each activity is introduced and the problem of
project scheduling is studied in terms of the project's discounted cost and of the risk of not meeting its completion time.
The adoption of the earliest (latest) starting time for each activity decreases (increases) the risk of delays but increases
(decreases) the discounted cost of the project. Therefore, an optimal compromise has to be achieved. This problem of
optimization is studied in terms of the probability of the duration and of the discounted cost of the project falling out-
side the acceptable domain (Risk function) using the concept of ¯oat factor as major decision variable. This last concept
is proposed to help the manager to synthetize the large number of the decision variables representing each schedule for
the studied project. Numerical results are also presented for a speci®c project network. Ó 1998 Elsevier Science B.V.
All rights reserved.

Keywords: Project management; Scheduling; Risk analysis; Finance

1. The concept of project and of project risk
The concept of project has evolved since the
early papers about networking and project sched-
uling (see e.g., [1]), considered nowadays as an in-
terconnected set of activities aiming to achieve
speci®c goals which can be related to a wide range
of objectives, such as the development of a new
*
Corresponding author. Fax: +351 8409884; e-mail:
cesur@civil18-ist.utl.pt.
1
Fax: +351 8409884; e-mail: cesur@civil18-ist.utl.pt.
product, the implementation of an organizational
change or the construction of a new building [2].
This implies that each project is supposed to at-
tain speci®ed levels of indicators which should ex-
press the goals of the institution responsible for the
project. These indicators focus relevant perspec-
tives such as
± duration
± cost
± bene®t
± consumed resources
± quality of the results,

0377-2217/98/$19.00 Ó 1998 Elsevier Science B.V. All rights reserved.

PII S 0 3 7 7 - 2 2 1 7 ( 9 7 ) 0 0 3 4 4 - 5
452 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469
and quite often, these perspectives can be aggregat-
ed into three major indicators: Z ± time (duration,
milestones, etc.); X ± cost (resources, expenses,
etc.); Y ± bene®t (outputs, quality, receivals, etc.).
Thus, the performance of the project will be de-
®ned in terms of these three indicators and a ``tar-
get domain'' will be de®ned by f X 6 LX ; Y P LY ;
Z 6 LZ g:
The bene®ts are due to the use of the system (or
of the changes) produced by the project and they
tend to be more dependent on external factors
(market, demand, competition, etc.) rather than
on the internal development of the project, at least
if the achieved level of quality is assumed to be
constant.
In this case, the study of the project's perfor-
mance can be carried out in terms of the plane
XOZ (Fig. 1) and the target domain is de®ned by
X. Whenever the performance falls outside X, the
project fails to meet its objectives and the proba-
bility of such an event can be considered as a mea-
sure of the project risk. This risk depends on the
uncertainty of the project's components, namely,
the randomness of the duration and of the cost
of each activity, as well as the calendar adopted
to schedule the project's activities.
Research on uncertainty has focused on the is-
sue of duration but has not given enough attention
to the randomness of the costs [3]. Also, the dura-
tion and cost of each activity are usually consid-
ered independent variables but unfortunately this
is not the case of most real projects. Actually, a
longer duration of an activity is usually due to
the need of carrying out additional works or of
overcoming unforeseen problems and therefore,
Fig. 1. Reduced target domain.
additional resources are also required by such
activity.
The research presented in this paper is oriented
to study the project's risk as a function of the un-
certainty of the duration and the cost of each ac-
tivity and in terms of the adopted schedule which
is considered the major decision of this problem.
The adoption of the earliest (latest) starting time
for each activity reduces (increases) the risk of an
overall delay but increases (decreases) the project's
discounted cost and therefore an optimal compro-
mise has to be achieved [4]. Unfortunately, the de-
cision on the schedule implies the selection of a
large number of decision variables which is equal
to the number of activities. A concept is proposed
in the next section ± ¯oat factor ± to help the project
manager through the synthesis of that large set of
variables into a single decision variable. Then, the
optimal value for this factor is studied in terms of
a stochastic model describing the project network.
2. On the synthesis of the project's schedule: A
theorem about the propagation of ¯oats
2.1. The problem
One of the most critical problems in Project
Management concerns the use of the ¯oats of the
activities to schedule a project.
Actually, scheduling a project implies setting up
the starting time, tis of each activity, i ˆ 1; . . . ; N
and by de®nition, each time tis should satisfy (a)
tis …E† 6 tis 6 tis …L†; where tis …E† and tis …L† are the ear-
liest and the latest starting times of i, and (b)
tis ‡ di 6 minj2J …i† tjs ; where di is the duration of i
and tis ; tjs are the adopted schedule times for i
and j. The set J(i) just includes all the activities
requiring the immediate precedence of i as it is
shown in Fig. 2 (using the usual convention Activ-
ity-on-Arc which is adopted in this paper).
Fig. 2. The precedence relationship.
L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 453

Fig. 3. Structure of the decision-aid model.

454 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469

The di€erence between tis … L† and tis … E† is called
the total ¯oat of i,
Di ˆ tis … L† ÿ tis … E†:
The adoption of tis near tis … L† increases the risk of
overall delays but it decreases the discount cost
of the project and the other extreme policy of mak-
ing tis near tis … E† has opposite e€ects. Usually, the
most convenient policy to manage the ¯oat Di
avoids these two extremes but its study is rather
complex due to the too large number of decisions
variables (N starting times) and to the size of the
set of constraints to be satis®ed. In Section 2.2, a
synthetic description of such decisions is proposed
in order that their evaluation and selection will be
performed by easier procedures.
2.2. A theorem on the propagation of ¯oats
The proposed synthesis is described by the fac-
tor a (¯oat factor) with

Fig. 4. The network under study (activities i ˆ 1; . . . ; 16).

Table 1
Case A
Activity Duration Cost
Mode l Lower limit r c l r
1 58.26 67.00 50.66 3.65 2.99 151.00 7.07
2 59.13 75.00 51.42 4.98 4.32 34.00 4.47
3 129.57 149.00 112.67 5.44 2.99 219.00 5.92
4 81.74 94.00 71.08 4.32 2.99 126.00 5.48
5 41.74 48.00 36.29 3.09 2.99 220.00 9.49
6 69.57 86.00 60.49 5.04 3.97 97.00 8.37
7 8.70 10.00 7.56 1.41 2.98 34.00 3.87
8 20.87 24.00 18.15 2.19 2.99 103.00 6.32
9 43.48 50.00 37.81 3.15 2.99 170.00 7.94
10 41.74 54.00 36.29 4.40 4.58 51.00 5.57
11 6.09 7.00 5.29 1.18 2.97 20.00 3.16
12 32.17 37.00 27.98 2.71 3.00 207.00 8.94
13 77.39 89.00 67.30 4.21 2.99 154.00 10.00
14 67.83 85.00 58.98 5.17 4.16 194.00 10.95
15 56.52 65.00 49.15 3.60 2.99 292.00 10.00
16 39.13 45.00 34.03 2.99 2.99 69.00 8.94
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 455

tis …a† ˆ tis … E† ‡ a:Di ;
where 0 6 a 6 1. The latest (earliest) schedule is ob-
tained with a ˆ 1…a ˆ 0†. In order to use tis …a† with
the same a for the whole network it is necessary to
prove that the ®nishing time of i assuming tis …a† is
not incompatible with adopting tjs …a† for any
activity j 2 J …i†. Therefore the following theorem
should be proved.
Theorem of the ¯oat factor. The ®nishing time of i
assuming that i starts at tis …a† is not greater than
the starting time of any j using the same a, tjs …a†,

Table 2
Case B
Activity Duration Cost
Mode l Lower limit r c l r
1 51.07 67.00 50.66 7.31 44.64 151.00 14.14
2 51.71 75.00 51.42 9.97 87.54 34.00 8.94
3 113.58 149.00 112.67 10.90 44.71 219.00 11.83
4 71.66 94.00 71.08 8.64 44.29 126.00 10.95
5 36.58 48.00 36.29 6.20 45.19 220.00 18.97
6 60.86 86.00 60.49 10.07 74.82 97.00 16.73
7 7.62 10.00 7.56 2.83 45.49 34.00 7.75
8 18.30 24.00 18.15 4.35 43.75 103.00 12.65
9 38.12 50.00 37.81 6.29 44.08 170.00 15.87
10 36.49 54.00 36.29 8.77 94.98 51.00 11.14
11 5.33 7.00 5.29 2.39 47.67 20.00 6.32
12 28.21 37.00 27.98 5.41 43.97 207.00 17.89
13 67.85 89.00 67.30 8.40 44.22 154.00 20.00
14 59.33 85.00 58.98 10.31 80.38 194.00 21.91
15 49.55 65.00 49.15 7.19 44.40 292.00 20.00
16 34.31 45.00 34.03 5.97 43.93 69.00 17.89

Table 3
Schedule for the studied network
Activity Precedences Time
ES EF LS LF Start Finish
1 ) 0.00 67.00 8.00 75.00 0.00 67.00
2 ) 0.00 75.00 0.00 75.00 0.00 75.00
3 ) 0.00 149.00 45.00 194.00 0.00 149.00
4 ) 0.00 94.00 100.00 194.00 0.00 94.00
5 1 67.00 115.00 113.00 161.00 67.00 115.00
6 1; 2 75.00 161.00 75.00 161.00 75.00 161.00
7 2; 3 149.00 159.00 208.00 218.00 149.00 159.00
8 3; 4 149.00 173.00 194.00 218.00 149.00 173.00
9 5 115.00 165.00 161.00 211.00 115.00 165.00
10 5; 6 161.00 215.00 161.00 215.00 161.00 215.00
11 6; 7 161.00 168.00 228.00 235.00 161.00 168.00
12 7; 8 173.00 210.00 218.00 255.00 173.00 210.00
13 9 165.00 254.00 211.00 300.00 165.00 254.00
14 9; 10 215.00 300.00 215.00 300.00 215.00 300.00
15 10; 11 215.00 280.00 235.00 300.00 215.00 280.00
16 11; 12 210.00 255.00 255.00 300.00 210.00 255.00
456 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469
h i
with j 2 J(i). This means that tis …a† ‡ di 6 tjs …a†; tjs …a† ˆ tjs …E† ‡ a tjs … L† ÿ tjs …E†
with j 2 J …i† and 0 6 a 6 1:
ˆ …1 ÿ a† tjs …E† ‡ a tjs …L†
Proof. The earliest start time of any activity  
ˆ …1 ÿ a† tis …E† ‡ di ‡ di ‡ a tjs …L†
j; with j 2 J …i†; tjs … E† is given by tjs … E† ˆ tis
… E† ‡ di ‡ di with di P 0 in order to account ˆ …1 ÿ a†tis … E† ‡ …1 ÿ a† di
for other paths converging at the starting node
of j. ‡ a tjs …L† ‡ …1 ÿ a†di
s
The hlatest start time i of i; ti … L† is given by and
s s
ti … L† ˆ minj 2 J …i† tj … L† ÿ di and the ®nishing
   
time of i adopting tis …a† is given by tis …a† ‡ di : tis …a† ‡ di ˆ tis …E† ‡ a tis …L† ÿ tis …E† ‡ di
Then,

Fig. 5. Calendar for a ˆ 0. Notation: h earliest schedule; n latest schedule; M critical activity.

Fig. 6. Calendar for a ˆ 1. Notation: h earliest schedule; n latest schedule; M critical activity.
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 457

ˆ …1 ÿ a†tis …E† ‡ atis …L† ‡ di The deduced result allows the project manager
  to select the same ¯oat factor a for the whole set
s s
ˆ …1 ÿ a†ti … E† ‡ a min tj … L† ÿ di ‡ di of activities.
j2J …i†

ˆ …1 ÿ a†tis … E† ‡ …1 ÿ a†di ‡ a min tjs … L†:

j2J …i† 3. A stochastic model of the project network

Finally, tis …a† ‡ di 6 tjs …a† because a minj2J…i† tjs … L† The adopted model assumes that the duration
6 atjs … L† ‡ …1 ÿ a†di : of each activity is lognormally distributed because

Fig. 7. Distribution of s with a ˆ 0 for (a) Case A and (b) Case B.

458 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469

the lognormal law has properties quite convenient
to model the distribution of the duration of each
activity:
(a) There is a lower bound which corresponds
to the minimal feasible duration.This limit is due
to contractual or technical reasons.
(b) The lognormal distribution is an asymmet-
ric distribution with the mode on the left side of
the expected value, which is a very common feature
of the duration of the activities. The skewness co-
ecient is positive and denoted by c.
(c) The upper quantiles are unbounded. Actual-
ly, the occurrence of uncertain and inconvenient
factors can always delay even further, the duration
of the activity.
Other authors have discussed and selected the
lognormal distribution for project networks (see,
e.g. [5]).

Fig. 8. Distribution of s with a ˆ 0.5 for (a) Case A and (b) Case B.
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 459

The cost of each activity is assumed to be
Gaussian and the hypothesis of signi®cant interde-
pendence between the cost and the duration is
adopted as this is the most common case in real
projects, where a longer duration means additional
work to be carried out or problems to be over-
come, always requiring additional resources. Thus,
the correlation between the cost and the duration,
q, is given and a linear regression is used to gener-
ate the cost in terms of the duration.
The project manager has to make a decision
about the starting time of each activity and such
time can be expressed in terms of a multiplied by
the ¯oat of i, Di , determined by assuming that
the durations of the activities are equal to their ex-
pected values. These scheduled starting times are
denoted by tis …a† and given by tis …a† ˆ tis … E† ‡ aDi :
Therefore, a simulation model can easily gener-
ate a random set of realizations …k ˆ 1; . . . ; K†
of the studied project network, de®ning each

Fig. 9. Distribution of s with a ˆ 1 for (a) Case A and (b) Case B.

460 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469

realization by a duration and a cost for each activ- activities. Actually, the earliest starting time of
ity ‰di …k †; Ci …k †Š; with i ˆ 1; . . . ; N : After selecting any activity j for such a set of occurred durations
the set of values ftis …a†; with i ˆ 1; . . . ; N g, the in realization
 s k can be
s
determined by tj …E†k ˆ
project manager can start the  project and the real maxi2I…j† ti …a†k ‡ di …k† ; where I(j) represents the
set of durations and costs di …k †; Ci …k †; with i ˆ set of all activities directly precedent to j and
1; . . . ; N g will be considered as a realization of tis …a†k is the real starting time of i which is given by
the adopted model.
The occurred durations fdi …k†; i ˆ 1; . . . ; N and tis …a†k ˆ tis …E†k if tis …E†k > tis …a†
k ˆ 1; . . . ; Kg imply that for each i and k, one may
have tjs …a† for j 2 J …i† not feasible for one or more and

Fig. 10. Distribution of v with a ˆ 0 for (a) Case A and (b) Case B.
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 461

tis …a†k ˆ tis …a† if tis … E†k 6 tis …a†:
Thus, a progressive iterative procedure starting
with the activities, i, having an empty I(i), will de-
termine the whole set of tis …a†k ; i ˆ 1; . . . ; N and
k ˆ 1; . . . ; Kg: Therefore, for each realization k,
the total duration and the discounted cost can be
computed.
The support to the process of decision making
concerning the selection of a can then be devel-
oped in terms of the risk estimated from the K gen-
erated realizations of the network, K being a
suciently large number. This risk function, R,
is the probability of falling outside the target do-
main de®ned by (Lx , Lz ). This probability is esti-
mated by the relative number of realizations with
results outside the adopted target domain. For
each target domain, the optimal a is its value min-
imizing R. The study of the total duration is pre-
sented in terms of sk ˆ Tk =T0 where Tk is the

Fig. 11. Distribution of v with a ˆ 0.5 for (a) Case A and (b) Case B.
462 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469

total duration correspondent to the realization k where f is the discount factor, and it is assumed
and T0 is the total deterministic network duration, that the cost of each activity, Ci (k), is allocated
assuming that the duration of each activity is equal to its starting time, tis …a†k . Then, PC will be studied
to its mean. in relative terms by
The analysis of the discounted cost can be made P s
for each k, in terms of the Present Cost, PC, which Ci …k †f ti …a†k
vk ˆ P tis …a0 †
;
is given by i l…Ci † f
X
PCk ˆ
s
Ci …k †f ti …a†k ; where l…Ci † is the average of Ci …k†, and a0 is the
i reference value adopted for a…usually; a0 ˆ 0†.

Fig. 12. Distribution of v with a ˆ 1 for (a) Case A and (b) Case B.
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 463

The estimated v for the realization k is denoted where B is the bene®t received when the project is
by vk . completed. The RPV obtained for the determinis-
Therefore, the target domain will be de®ned in tic network with the duration and cost of each ac-
terms of X ˆ v and Z ˆ s. tivity equal to their means, and with a ˆ a0 , is
Alternatively, the study of the cost can be done denoted by RPV0 , and so the estimated results ob-
in terms of the Relative Present Value of the pro- tained from the simulation can be presented for
ject, RPV, which is de®ned for the realization k by each k through
," #
X s RPVk
RPVk ˆ …k † ˆ … BfTk † Ci …k †f ti …a†k ; wk ˆ ;
i
RPV0

Fig. 13. Estimated mean and quantitiles of s in terms of a for (a) Case A (b) and Case B.
464 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469

where RPVk is the RPV computed for the generat- appropriate a in terms of s and v, using the struc-
ed realization k. It should be noted that w is not a ture presented in Fig. 3.
function of B because
P s
Tk ÿT0 Ci …k †f ti …a†k 4. An application
wk ˆ f Pi tis
:
i l…Ci †f …a0 †
4.1. Results
In this case, X would be made equal to w.
This model is implemented as a decision aid The developed model was applied to an illustra-
to help the project manager to select the most tive project network with the data presented in

Fig. 14. Estimated mean and quantitiles of v in terms of a for (a) Case A and (b) Case B.
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 465

Fig. 4 and in Table 1 (Case A). The parameter q is
assumed to be equal to 0.7.
Alternatively, a network with a higher uncer-
tainty is also studied (Case B) multiplying the pre-
vious standard deviation of the duration and of
the cost by a factor of 2 (Table 2).
The major decision variable is a which varies
between 0 and 1 and the risk function, R, is studied
in terms of the limits of the target domain …Lx; Lz†,
as was previously presented.
The study of this network is initially performed
assuming deterministic data equal to the means of
durations and costs. The result for the start and
®nish times are presented in Table 3. The calendar
is presented in Figs. 5 and 6 for a ˆ 0 and a ˆ 1,
obtaining a total duration of 300 units.
The stochastic simulation is carried out using
the presented model. A sample of 10 ´ 1000 real-
izations is generated and, for Cases A and B, the
estimated distribution of s and v (where
p‰sŠ or p‰vŠ denotes the estimated probability of oc-
currence of s or v within the corresponding inter-
val) are presented in Figs. 7±12 for a ˆ 0, a ˆ 0.5
and a ˆ 1.0.
The analysis of s and v in terms of a is also pre-
sented in Figs. 13 and 14 where the estimated
mean (l) and the 5% and 95% quantiles (Q0:05 ,
Q0:95 ) are plotted.
The estimated results for the average, standard
deviation and quantiles of s and v in terms of a
and of the adopted case are presented in Tables 4
and 5. In these tables, the standard error of such

Table 4
Estimated parameters of the distribution of s
a 0.0 0.5 1.0
Variance LV HV LV HV LV HL
l 1.074 1.266 1.081 1.275 1.102 1.306
SE(l) 0.001 0.011 0.001 0.010 0.001 0.007
r 0.146 0.833 0.143 0.870 0.132 0.782
SE(r) 0.003 0.134 0.003 0.240 0.004 0.099
Q0:95 1.345 2.226 1.350 2.237 1.348 2.279
SE(Q0:95 ) 0.007 0.037 0.005 0.035 0.006 0.030
Q0:05 0.940 0.917 0.952 0.925 0.990 0.968
SE(Q0:05 ) 0.001 0.000 0.000 0.000 0.001 0.000
LV ± Low variance (Case A).
HV ± High variance (Case B).

Table 5
Estimated parameters of the distribution of v
a 0.0 0.5 1.0
Variance LV HL LV HV LV HL
l 0.969 0.957 0.811 0.797 0.686 0.675
SE(l) 0.001 0.002 0.000 0.001 0.001 0.001
r 0.028 0.148 0.026 0.128 0.025 0.110
SE(r) 0.001 0.023 0.000 0.012 0.001 0.012
Q0:95 1.018 1.101 0.855 0.913 0.728 0.780
SE(Q0:95 ) 0.001 0.006 0.001 0.005 0.001 0.004
Q0:05 0.928 0.819 0.765 0.718 0.648 0.603
SE(Q0:05 ) 0.001 0.001 0.019 0.001 0.000 0.001
LV ± Low variance (Case A).
HV ± High variance (Case B).
466 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469

Fig. 15. Estimated risk for Region 1 in terms of a for (a) Case A and (b) Case B.

parameters (SE ( )) are estimated using 10 samples
of 1000 realizations.
Finally, the risk function, R, was computed (for
Cases A and B) in terms of a and using di€erent
bounds for Lx and Lz :
This set of results is presented in Figs. 15±17, al-
lowing the estimation of the optimal a which is giv-
en by the a minimizing R.
Therefore, the recommended values for a are
given in Table 6.

v s
4.2. Comments
Region 1 Lx ˆ 0.6 Lz ˆ 1.0
Region 2 Lx ˆ 0.8 Lz ˆ 1.20 The presented results deserve the following
Region 3 Lx ˆ 1.0 Lz ˆ 1.40 comments:
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469 467

Fig. 16. Estimated risk for Region 2 in terms of a for (a) Case A and (b) Case B.

± The distribution of s always has a positive
skewness, showing that the Gaussian assumption
of PERT is far from acceptable for this type of ex-
ample. However, the distribution of v is reasonably
symmetrical. It should be noted that these results
were obtained assuming a lognormal law for the
duration of the activities and a linear Gaussian re-
gression for the cost in terms of the duration.
± The level of variance adopted is crucial as the
assumption of a higher variance for the durations
of the activities increases the e€ect of a very signif-
icantly. Actually, the increase of the duration and
the decrease of the present cost is much clearer
when a increases for Case B than for Case A. This
e€ect is particularly strong for the skewness and
the upper quantile.
± The presented results for the risk, R, show
that the concept of optimal a can be estimated
through the presented model. The risk gain due
to adopting the optimal a grows with the uncer-
tainty of the data as would be expected and its
value also depends signi®cantly on the bounds
adopted and on the level of variance (Case A
or B).
468 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469

Fig. 17. Estimated risk for Region 3 in terms of a for (a) Case A and (b) Case B.

Table 6 cost and duration associated with the project's

Optimal value or interval for a network.
Region Case A Case B These magnitudes are modelled in terms of the
1 0.8 0.7 scheduled starting time of each activity and the
2 0.4±0.8 0.3±0.7 concept of a ¯oat factor is introduced as a decision
3 0.1±1.0 0.3±0.9 aid to help the project manager to select the most
convenient schedule. The advantage of using this
concept stems from the possibility of using the
5. Conclusions same factor throughout the whole project, as has
been proved in this paper.
The concept of project risk was proposed and A model is proposed to study the project risk in
de®ned in this paper as a function of the discounted terms of the ¯oat factor (risk function).
 L.V. Tavares et al. / European Journal of Operational Research 107 (1998) 451±469
An application is presented showing how this
model is a useful instrument to support the deci-
sions of the project manager through the estimation
of the distributions of total cost and duration and
through the estimation of the project risk. The risk
can easily be estimated in terms of the ¯oat factor
and its optimal value can therefore be computed.
References
[1] J.E. Kelly, Critical path planning and scheduling mathe-
matical bases, Operations Research 9 (1961) 246±320.
469
[2] L.V. Tavares, A review on the contributions of opera-
tional research to project management, EURO 14 (1995)
67±82.
[3] S.E. Elmaghraby, A guided tour through some recent
developments, European Journal of Operational Research
82 (1995) 383±408.
[4] L.V. Tavares, A stochastic model to control project
duration and expenditure, European Journal of Opera-
tions Research 78 (1994) 262±266.
[5] B. Dean, S. Mertel, Jr., L. Roepke, Research project cost
distribution and budget forecasting, IEEE Transactions on
Engineering Management 16 (4) (1969) 176±191.