KYBERNETIKA — VOLUME 31 ( 1 9 9 5 ) , N U M B E R 2, P A G E S 151-174

MULTISTAGE STOCHASTIC PROGRAMS:

THE STATE-OF-THE-ART
AND SELECTED BIBLIOGRAPHY 1

JITKA DUPAČOVÁ

The paper gives a brief introduction into the problems of multistage stochastic program­
ming with emphasis on the modeling issues (Section 2) and on the contemporary numerical
advances (Section 3). Extensive classified bibliography is contained in the last Section.

1. INTRODUCTION

Mathematical modeling of economic, ecological and other complex systems with the
goal to analyze them and to find optimal decisions ha§ been studied for many years.
The challenging problems connected with running market economies, of realistic
approaches to environmental protection, etc., are that the decisions are to be made
under uncertainty. Therefore the traditional deterministic optimization models are
limited in practical applications because the models parameters (future demands,
interest rates, water inflows, resources, etc.) are not completely known when some
decision is needed. A typical approach of substituting expected values for all random
parameters can lead to inferior solutions that discredit both the model designer and
the use of optimization methods.
Moreover, in controlling or analyzing complex systems, various levels of uncer­
tainties have to be taken into account: besides of requirements for proper treatment
of nonhomogeneity of raw input materials, volatility of prices, demands or of water
inflows one is asked to cope with future development of factors essential for running
the system such as interest rates or innovations of technological progresses and to
hedge against legislative changes and complete or partial changes of economic and
other policies. In principle, these uncertainties can be modeled by various ways and
one of them is stochastic programming.
Stochastic programming gives a probabilistic interpretation to the above men­
tioned uncertainties. It deals with optimization problems in which random parame­
ters are explicitly spelled out and allows for incorporation of risk into optimization.
1
This work was supported by the Grant Agency of the Czech Republic under Grant
No. 402/93/0631
152 J. DUPAČOVÁ

It originated in the late fifties, cf. seminal papers by Beale [Rl], Dantzig [R3], Tint-
ner [R13] and Charnes and Cooper [R2]; some one dimensional examples of stochastic
programming can be traced even in earlier papers on inventory, maintenance, etc.
The general formulation of stochastic programming problems, cf. [155], reads:

Minimize Ef0(x,ui) (1)

subject to E{/,(t5,w)} < 0, i ={,...,k

E{h(x,u)} = 0 , i = k+l,...,k +r (2)
x£X
where
w is a random parameter with support fi, probability distribution P
and the corresponding expectation E,
X is a given nonempty closed set,
/o : R" x Q -> R1 U {+oo}, / ; : Rn x O — R1 are given functions. This
formulation looks like an ordinary static deterministic program whose objective and
constraints are of the form of expectations. Still it covers a whole spectrum of
stochastic programming models - static, two-stage and multistage ones, models with
probabilistic constraints and those of penalty or recourse type - provided that the
important requirement of nonanticipativity, i.e., the requirement that decisions must
occur before observations, has been properly treated. In the simplest case, X C
Rn and the decision variable x in (1),(2) corresponds mostly to the main decision
that has to be selected before the realizations of the random parameters u> can be
observed. In more complicated cases, x consists of several subvectors, say I 1 , . . ., xT
that correspond to decisions to be taken at the stages 1 , . . . ,T of the decision process
and, as we shall see in Section 2, it may be even convenient to model the decisions
as random functions. The requirement of nonanticipativity can be incorporated
into the definition of functions /,- or it can be formulated as an explicit additional
constraint, see Wets [151] and Rockafellar and Wets [132] for the first developments
of this idea.
A typical assumption is that the probability distribution P is known (decision
making under risk) and independent of decision x, on the other hand, various ap-
proaches have been designed that deal with stochastic programs under incomplete
knowledge about the underlying distribution P (decision making under uncertain-
ty); for a survey of stochastic programming applications under incomplete knowledge
about P see [R6].
The existence of expectations in the constraints (2) for all feasible decisions x is
guaranteed by special assumptions from case to case. Proper treatment of expecta-
tions of extended real functions that enter the objective (1) is not straightforward,
see [155], on the other hand, using this formulation helps to concentrate on influence
of specific constraints, e. g. of the nonanticipativity constraints, or to study problems
in which the choice of solutions is limited by implicit, induced constraints on the
solvability of the system. These constraints are not included explicitly in the system
of constraints of (2), they can be sometimes detected at least partly by suitable
preprocessing techniques, cf. [R16], or treated in course of numerical computations.
Multistage Stochastic Programs: The State-of-tfie-Art and Sclectcd Bibliography 153

Ability to choose decisions that perform well regardless these hidden constraints is
one of strongpoints of stochastic programming.
The prevailing theoretical issue in models with probabilistic constraints turned to
be the convexity property of the resulting deterministic program of type (1),(2) with
indicator functions at the place of some of /,•; cf. Prekopa [R8] for the important
breakthrough.
The theoretical results have been collected in various works, e.g., monographs
[R7], [165] and collection [D]; for more recent results see survey [155] or the new
textbook [85]. The progress in designing efficient algorithms (cf. Part II of [P-W],
[E-W] and Part II of [B-W]) has resulted into special software packages suitable for
solving large stochastic programs that arise in a variety of applications such as power
generation planning, financial modeling or location analysis. The main stumbling
block for algorithms is necessity to compute repeatedly values of multidimensional
integrals (expectations of recourse functions or probabilistic constraints) that enter
the nonlinear program (1),(2). To overcome this problem various approximation
schemes, both stochastic and deterministic ones, were designed; see, e.g., [R18]
and the references therein. In connection with evaluation of their properties and
with the need for proper treatment of uncertainty about the probability distribution
of random parameters, various error bounds have been derived and miscellaneous
results on stability and postoptimality have been achieved; see, e. g., Part I of [P-W],
Part I of [B-W] or survey papers [R4], [R5] and the references therein. A new area
of interest is integer stochastic programming with many open theoretical problems
and various interesting applications; one of the first papers is [126].
The present stage of knowledge and of computer technologies gives a chance to
turn attention to the dynamic multistage stochastic programming problems. This
area was mentioned already in the seminal paper of Dantzig [R3] and in his mono-
graph [27] and the first theoretical results on multistage stochastic programs with
recourse were obtained as generalization of those valid for two-stage stochastic pro-
grams, e.g., [113], [118] - [121], [149], [150]. Deep theoretical results closely connect-
ed with the crucial problem of modeling the multistage nature of the decision process
can be found, e. g., in [32] - [34], [128] - [135]; in these papers, multistage stochastic
programs are treated as optimization problems in infinite dimensional spaces. For
expositions concerning multistage stochastic programs with probabilistic constraints
see, e.g., [47], [124], [162]- [166].
Besides of finance, the most popular areas for applications of multistage stochastic
programs seem to be for the present production planning and management including
electric power generation and transmission, transportation, optimal exploitation of
exhaustible resources and water resources management; see Section 4 for references.
Except for bibliography (Section 4), we shall limit ourselves to multistage stoch-
astic programs with recourse. After discussing briefly the modeling issues (Section
2), we shall turn our attention to scenario based multistage stochastic programs with
recourse and in Section 3, we shall report on the relevant numerical methods.
154 J. DUPACOVA

2. MULTISTAGE MODELS

Consider first the two-stage stochastic programming problems: A decision x 6 X

should be selected before realizations of random parameters can be observed or their
values revealed. After this information becomes available the decision process con-
tinues by the second stage, i.e., by the choice of an auxiliary decision that depends on
the first-stage decision and exploits the already obtained information. The second-
stage decision is interpreted as an updating activity (portfolio revision, adjustment
of the production plan, etc.) that brings about additional, recourse costs. The re-
quirement that the first-stage decision x depends only on the past information and
that it cannot depend on future observations of random parameters corresponds to
the more general nonanticipativity property of the multistage decision processes.
It is important to realize that the stages do not necessarily correspond to time
periods. The first-stage decisions consist of all decisions that have to be selected
before the information is revealed whereas the second-stage decisions arc allowed to
adapt to this information; for a detailed explanation and examples see [49].
The model formulation that reflects the above verbally described decision scheme
can be written in the following way:
Let X\, X2 be nonempty closed sets in Rnt, R"2, respectively, and let ( Q , £ , P )
be a probability space, fu, Vi given functions on R"1, f2i, Vi given functions on
H x Rni x Rn2 that are P measurable for each x1 £ R™1, x2 £ R"2. The problem
is to
minimize fl0(x1)+ I f20(x\x2(w),w)P(Aw) (3)
Jti

over all x1 € R"' that satisfy

xleXx and /i.C* 1 ) < 0, i=l,...,mi (4)

2
and x measurable such that P almost surely (a.s.)

x2(w)eX2 and f2i(x1,x2(w),w) < 0, i = l,...,m2, (5)

an optimization problem in a suitably chosen infinite dimensional space of measur-
able functions.
An alternative formulation of the type (1),(2) reads

minimize E{/o(a; 1 ,w)} subject to the constraints (4) (6)

l
with fo(x ,w) defined as follows:

f0(x\w) := f10(xl) + inf [f20(x\x2,u>)\ f2i(x\x2,w) < 0 Vi, x2 6 X2] , (7)

x 2
The function f20(x , x ,w) that appears in both formulations gives the cost of
the recourse connected with the (not necessarily optimal) second-stage decision x2
1
in case that x is the accepted first-stage decision and w is the subsequently observed
realization of the random parameters. The function /io(* 1 ) corresponds to the costs
Multistage Stochastic Programs: The State-of-the-Art and Selected Bibliography 155

that are independent of the second-stage decisions and it can be also defined as an
expectation.
For the convex case and for x2 in (3), (5) restricted to the class of essential-
ly bounded measurable functions, Rockafellar and Wets [Rll] gave relatively weak
conditions under which the introduced formulations are equivalent. The result ap-
plies, for instance, to X\ and Xi bounded. The first formulation is suitable for
theoretical analysis such as optimality conditions or duality properties for problem
( 3 ) - ( 5 ) . The results depend, inter alia, on the considered space of the measurable
functions x2(u). We refer to the series of papers [R9] - [R12] or to [32].
Special attention has been paid to the class of two-stage linear stochastic pro-
grams, known under the name stochastic linear programs with recourse. Their
generic form that corresponds to the formulation (6), (7) reads

minimize E {c(ui)Tx + Q(x,u)} on the set K\ = {x € R+l : Ax = b} (8)

with the recourse costs Q(x,u) defined for a given x and u as the optimal value of
the auxiliary second-stage program

minimize q(ui)Ty (9)

subject to y £ R^1 that satisfy W(w)y + T(ui)x = h(u).

Notice that only the expectations of the random coefficients c(ui) enter the above
formulation (8) so that fixed costs c can be used without any loss of generality.
According to properties of the recourse matrix W(ui) this stochastic linear pro-
gram (SLP) is classified as
o SLP with fixed recourse if W(ui) — W, a fixed matrix,
o SLP with fixed complete recourse if W(u) = W and if for an arbitrary right
hand side w the system Wy = w has a nonnegative solution,
o SLP with reiativeiy complete recourse if the second-stage problem (9) is a.s.
feasible for an arbitrary x G K\ and w g 0, etc.
Induced constraints concern the case when the second-stage program may happen
to be infeasible for a first-stage decision x £ K\ and a realization of the random pa-
rameter u). From the point of view of the modeled problem, accepting such decision
may lead to disaster (interruption of the production process, bankrupcy, environ-
mental catastroph, etc.) and infinite costs Q(x,u) can be used to reflect this fact;
compare with possibly infinite values of f0(x,u>) in (1). Another idea is to com-
plete the constraints of the deterministic program (8) to avoid accepting first-stage
decisions that may lead to infeasible second-stage program (9); we shall denote by
Ki the set described by induced constraints. As to the structure of the resulting
deterministic equivalent program

minimize E {c(ui)T x + Q(x,u)} on the set K\C\K2 (10)

it is possible to prove that (10) is a convex program provided that W is a fixed

matrix. Additional conditions are needed to guarantee that the objective function
is well defined; for instance, a sufficient condition is the existence of all second-order
156 J. DUPACOVA

moments of the vector of all random parameters. For SLP with fixed recourse the
set K,2 of induced constraints can be written as

K2 = {x :3y>0 such that Wy = h(u) - T(w)x a.s.} (11)

so that (10) is equivalent, compare (3) (5), with

minimize E {c(ui)Tx + q(ui)Ty(u)} (12)

subject to x 6 ICi and y(w) > 0 such that

Wy(ui)+T(ui)x = h(ui) a.s.
Moreover, under additional conditions on the probability distribution P, such as
T(u>) = T or for a discrete distribution P, the set K2 is convex polyhedral so that
it can be generated by a finite number of suitable cuts step by step. For a detailed
survey see [R17]. Characterization of SLP with random recourse is more demanding.
The cases where the above results remain valid are, e. g., SLP with random complete
recourse and problems with discrete distribution P; cf. [R16]. In the latter case,
for instance, one can index by s = 1 , . . . , £ the vectors of all possible realizations of
the random coefficients in q, W, T and h and those of the corresponding second-
stage variables y, assign probabilities ps to these realizations and arrives thus at the
following linear program of a special dual block angular structure

s
minimize c T 3i + V j p j q j y , (13)
J=I

subject to Ax =b
Txx + WlVl =hx
T2x + W2y2 = h2 (14)

Tsx + ... + Wsys =hs

x > 0, y . > 0, s = 1 , . . . , S .
The size of the program (13), (14) can be very large; for instance, consider just
random right hand sides h consisting of m2 independent random components with
probability distributions approximated by alternative ones: it gives mi + 2 m a con-
straints in (14). Usefulness of special numerical techniques is obvious; see [E-W].
An alternative equivalent formulation of (13), (14)
s
minimize cTx + VJp,qTys (13)
s=l

subject to Ax =b
Tsxs + Wsy, = hs (14')
x - xs =0
Multistage Stochastic Programs: The State-of-the-Art and Selected Bibliography 157

x > 0, xs > 0, ys > 0, s = l,...,S

helps to build procedures based on relaxation of the nonanticipativity constraints
x = xs, s = 1, .. ., S. (15)

In the general T-stage stochastic program we think of a stochastic data process

w = {w1, . . . ,w7 }
whose realizations are data trajectories in (fi, E, P) and of a vector decision process
x = {x\...,xT},
a measurable function of w. The whole sequence of decisions and observations can
be, e.g.,
x\w\x2(x\w1),w2,...,wT-\
xT(x\x2,..., xT-\w\. . .,wT-\ = xT(x\w1, . .. ,uT-1)
and wT that contributes to the overall observed costs. The decision process is nonan-
ticipative in that sense that decisions taken in any stage of the process do not depend
on future realizations of random parameters or on future decisions. On the other
hand, in the course of the decision process, the past information is exploited. The
dependence of the decisions solely on the history can be expressed as follows: Denote
Ej_i C E the (T-fleld generated by the observations {w1,. .. ,wt~1} of the part of
the stochastic data process that precedes the stage t. The dependence of the fth
stage decision x1 only on these past observations means that „ ' is E;_i-adaptable
or, in other words, that xt is measurable with respect to Ht-1- In each of stages,
the choice of a decision is limited by constraints that may depend on the previous
decisions and observations. Assumption of nonanticipativity of these constraints in
the sense that no additional constraint can enter later as a consequence of future
observations corresponds to the assumption of relatively complete recourse for two
stage models and assuming nonanticipativity of constraints means that no induced,
hidden constraints can appear.
Once more, two formulations can be used:
Let Xt be given nonempty sets in Rn<, t = 1,., .,T and denote
Xt(w) ={xu = (x\...,xt) : / „ • ( _ " , <~)<0, i = l,...,mt,
x " 6 X± x X2 x ... x Xt}
(17)
the set of the tth stage constraints, t = 2 , . . . ,T and by fo(x,w) the overall cost
connected with the decision process (16).
The T-stage stochastic program is to find
x1€X1 such that /i.O- 1 ) < 0, i =: l , . . . , m 1
x* E*_i -measurable, I s 2 T
(18)
xu G Xt(w),t = 2, ...,T a.s.
1 2 T
that minimizes E {/bO* , E (W), . . . ,x (w),w)} •
158 J- DUPACOVA

Allowing for extended real objective function f0 opens for instance the possibility
to incorporate the constraints xf' E Xt(ui), t — 2 , . . . , T into the objective function
and to study the influence of the nonanticipativity constraints, cf. [58], [132], or to
include, at least theoretically, the prospective induced constraints.
The second formulation is based on a recursive evaluation of the overall objective
function, compare with (7), that spells out the nonanticipativity in an explicit way:

minimize E {f0(x,w)} := /lof* 1 ) + Ewi {^(x1 ,u>1)} (19)

subject to x1 E Xi and / ^ ' ( x 1 ) < 0, i=l,...,nii,
where for t — 2 , . . . , T, (pt-i^x1,. . . ,x'~1,ui1,.. . , O J / _ 1 ) is the optimal value of the

stochastic program

minimize /<o(x') + Ew. {^(x1,..., x ' . w 1 , . . . ,u*~1,u*)} (20)

with respect to x' £ Xt that fulfils

/«(aj1I...,*'-1,a)*,w1>...)w*-i)<0> i= l,...,mt

and ipT,a = 0.

The latter formulation resembles the backward recursion common in stochastic

dynamic programming problems. In these models, the goal is to provide a sequence
of decision rules that can be used in particular stages of the decision process and
in any state of the system, that allow the decision maker to pass from observations
to decisions in an optimal way, e.g., for minimal total expected costs. To get the
decision rules, one applies the principle of optimality that in turn requires evaluation
of the optimal values in all stages in dependence on multidimensional parameters,
e.g., on x1,... ,xl~x ,LJ1 ,... ,u>i~1 in (20). Dimensionality puts serious limitations
even on discrete stochastic dynamic programming problems or stochastic control
problems with discrete time and finite number of states. Moreover, for application
of the backward induction, special structure of the problem (e.g., separability) is
essential.
On the contrary, the main interest in the multistage stochastic programming prob-
lems lies in the first-stage decisions. Even if it is possible to characterize the decision
rules [63], it is not necessary to design a full backward recursion as in dynamic pro-
gramming and, due to large dimensionality of the stochastic programming problems,
such procedure will be hardly tractable. For a qualified discussion of the relation-
ship between the stochastic control problems with discrete time and the multistage
stochastic programs see, e.g., [34],[89],[130], [135], [146], [156]. An important issue
for the both classes of stochastic dynamic decision models is the proper choice of the
horizon. In the context of stochastic programming, there are attempts to decrease
the contribution of costs for recourse activities in distant stages by discounting (see,
e.g., [7]), to decrease the number of stages by aggregation or by exploitation of a
special structure of the solved problem and there are suggestions how to treat the
effects of replacement of an infinite horizon problem by a finite horizon one [60], [72].
Multistage Stochastic Programs: The State-of-the-Art and Selected Bibliography 159

For results concerning the equivalence of the two formulations of multistage stoch-
astic programs with recourse for the decision space i ° ° see, e.g., [132]-[134], Opti-
mality conditions and duality results can be in principle obtained from the standard
results for nonlinear programs of the type (1),(2) that corresponds to the formula-
tion (19). Naturally, optimality conditions and dual problems obtained in this way
involve only the expectation functionals. To get pointwise conditions and dual prob-
lems expressed with respect to all u £ fi one can choose a suitable decision space of
measurable functions in the formulation (18) and exploit methods of abstract opti-
mization. Convex problems allow for application of conjugate duality results; special
care has been devoted to the impact of nonanticipativity constraints and to the cor-
responding dual variables or Lagrange multipliers that can be interpreted as prices
for nonanticipativity (see [34], [R9], [45], [52], [54]) and to the "singular" multipliers
attached to the induced constraints [133], For an extension of results on optimality
conditions to the nonconvex problems of the type (18) treated in LP space see the
recent paper [59].
A crucial problem for multistage stochastic programs is modeling the information
structure; see [155] for a variant of model (1), (2) that corresponds to situations when
the observation process reveals only a partial information.
A popular form of the multistage stochastic Jinear program (MSLP) with recourse
reads
minimize cjx1 + E_. {^(a: 1 ,-. 1 )} (21)

subject to constraints

subject to constraints Aix1 = &i

h < x1 < 1*1,

where the functions p . _ i , i ** 2 , . . .,T, are defined recursively as

^_i(.- < -Sw«- 1 ) = inf[c,(w'-1)T>-« + E u .{v'.(-i t 1 ^)}] (22)

subject to [Link]'-1) x1'1 + M*1'1) xt = bt{ut~l)

lt<x'<ut,

and tpT = 0.
For the sake of simplicity, we denote here by _><_1 the random vector that gen-
erates the coefficients bt, ct and matrices At, Bt in the decision problem of the
<th stage, t s 2 , . . . ,T (compare with the scheme (16) or with (20)), we assume a
Markovian structure of the constraints and of the objective and we suppose that the
corresponding expectations E are well defined. The bounds lt, ut Vi are nonrandom
and for the first stage, known values of all elements of bi, ci, Ai are assumed. The
assumption of fixed recourse means that At are known nonstochastic matrices for all
t. The main decision variable is x1 that corresponds to the first stage. If the random
parameters are stage independent, characteristic properties such as convexity of the
resulting deterministic program can be obtained from the results for two-stage prob-
lems with fixed recourse, see, e.g., [113], [149], [150]; Olsen [118]-[121] allows for
160 J. DUPAČOVÁ

dependence of the random right hand sides. It is clear how to formulate the MSLP
of the form (18) and when to expect (at least intuitively) the equivalence between
the two different formulations that arc used to model the same decision problem;
once more, general statements on equivalence between these two formulations are
not trivial, e. g. [128], [129], [134], The results on equivalence hold true, for instance,
for MSLP with finite discrete distribution P of all coefficients At, Bt, lit, Ct, Vi in
which case (18) reduces into a large linear program of a special structure similar
to (13), (14) that is convenient for decomposition purposes. This observation opens
possibilities of an algorithmic solution.

3. NUMERICAL TECHNIQUES

Various numerical techniques have been developed for multistage stochastic pro­
grams with recourse and with a given discrete multidimensional probability distri­
bution of the random parameters whose atoms, say w.,, s = 1,...,5, are called
scenarios. The origin of this discrete distribution can be very diverse; it can be
obtained as an approximation of a true continuous distribution, based on a sample
information, or related to scenarios provided by an expert. Similarly as for two-stage
problems, the specific assumption of discrete distribution allows for transformation
of multistage stochastic linear programs into large scale linear programs that can be,
in principle, solved by general purpose algorithms adjusted to the special structure of
the solved problem; see, e.g., [70]. However, the size of the resulting linear program
can be prohibitively large; it grows exponentially with the number of scenarios taken
into account and with the number of stages so that the direct approaches based on
standard linear programming software are of limited use. There are various ideas
how to reach a manageable size of the problem: to wave the stochastic character of
the data and to replace the random parameters by some fixed "base" values, e.g.,
by expectations [16]; to wave the possibility of adapting the decisions according to
the past information [81]; to use an appropriate labeling for to avoid ambiguity in
definition of data and of variables, cf. [11], [64], [92]; to aggregate some of periods
into one stage, e. g., [4], [50], [91], [108], [167]; to aggregate scenarios [137], [138], [154]
or both scenarios and periods "[7]; to select "important" scenarios using statistical
techniques [30] or expert's opinion; to decompose the problem into manageable ones
and to use parallel procedures. Regardless of the chosen solution technique the
desired optimal solution depends on the used scenarios and on their probabilities
and its performance under other "out of sample" scenarios should be a subject of
postoptimality, stability or simulation studies.

In primal decomposition methods, the subproblems are related to individual

stages. The original computer code by Birge [9] extends the L-shaped method
[R14] to three stage problems with random right-hand sides. Its generalization
by Gassmann [64], known under name MSLiP, concentrates to increase of efficiency
of the performance and allows for random elements in the matrix of the system of
constraints and for more than three stages. To save labor in implementation devel­
opment, performance evaluation and comparisons of existing and future software for
multistage stochastic programs, a standard input format [11] has been suggested.
Mułtistage Stochastic Pгograms: The State-of-the-Art and Selectєd BiЪłiagrapҺy lбl

The standard form of the T-stage stochastic linear program with fixed recourse
and with a finite number of scenarios (compare (13), (14) or (21), (22)) is

K3 K3 KT
minimize cjx0 + V J pk2cjxk2 + VJ pk3cjxks + ...+ VJ pkrcTxkT
k2 = l k3=K2+l kT=KT-,+l
(23)
subject to constraints

Ai^o = b0
Bk2x0 + A2xk2 =bk2, k2=l,...,K2
Bk3xa(k3) + A3xk3 =6_3, k3= K2+1,...,K3

BkTxa(kT) + ATxkT =bkT, kT=KT-1+l,...,KT

(24)
lt <xkt <ut, kt = Kt-i+ !,..., Kt, t = l,...,T

and it necessarily depends on the used scenarios, i.e., on finitely many sequences
of possible realizations (Bkt ,bkt), t = 2,.. ., T of the random right hand sides and
random transition matrices in the constraints for the stage t and on the path prob-
abilities pkt of subsequences of these realizations pkl>0 Vfc<, ___!„,-___,__+1 Pk, = l> ' =
2 , . . . ,T that identify the discrete distribution P. The probabilities ps of individ-
ual scenarios __>_ are equal to the path probabilities pkr, kT = KT-I + 1 , . . . , KT
and they can be obtained by multiplication of the (conditional) arc probabilities
of the corresponding realizations. In program (23),(24), xkl = x0 denotes the
first-stage decision variable and a(kt) denotes the immediate ancestor of kt, so that
a(fc2) = fc'i, k2 = 1,. .., K2- Induced constraints are not included in in the system
(24) but they can be treated in course of numerical procedure.
The input information that leads to (23), (24) can be represented in the form of
so called scenario tree or event tree. Each path through the tree from its root to one
of its leaves corresponds to one scenario, i.e., to a particular sequence of realizations.
The nodes where the branching occurs correspond to stages and they are indexed
by kt. In our notation, the root is indexed by fci = 0, there are Kt — A'._i nodes
indexed by kt = A . _ i + 1,... ,Kt for the stage t (with K\ = 0); particularly, the
KT — KT-I leaves indexed by kT correspond to scenarios. For each node of the
scenario tree, an entire set of decision variables is introduced in program (23), (24);
for instance, the vector of the first-stage decision variables x0 corresponds to the
root and subvectors xkt of the .th stage decision variables are assigned to the nodes
kt, respectively.
A natural idea is to decompose the large linear program to many relatively small
problems corresponding to individual scenarios. For instance, the problem corre-
sponding to scenario w, = {(_8__,j, 6__-_), J = 2 , . . . , T } in (24), if such scenario
exists, is the following linear program:

Minimize cj x0 + CJXK2 + cJxKl + • • • + C^XRT (25)

162 J- DUPACOVA

subject to constraints A\Xa — "0

BK.2xa + A2xK2 = bKl
BK3KK2 + A3xK3 = bK3

BKTXKT + ATxKr = bKT

(26)
It < xKl < • " ( , t=l,...,T
The x 0 -part of the optimal solution of (25), (26) gives the first-stage solution that
is optimal for the used scenario t<*ASfowleve h this solution need not be cither op-
timal or feasible for other scenarios. Decisions made conditional on knowledge of a
scenario, i.e., based on an assumed realization of ui, do not evidently fulfil require-
ments of the considered decision model such as to get scenario independent optimal
First-stage solution. Therefore, in designing algorithms based on decomposition with
respect to individual scenarios one must take care for nonanticipativity constraints:
At any stage t of the decision process, the final decision x1 can be based only on the
previous decisions and on the already observed segment of data represented in (24)
by one subsequence { ( B ^ , bkj), j= 1, . . . , t — 1}. In other words, decisions accepted
at the stage t that are based on the same history must be equal. In the introduced
form (23), (24) requirement of nonanticipativity has been included directly into the
constraints.
For decomposition methods with respect to scenarios nonanticipativity constraints
can be included in the form of a large system of simple binding linear equations that
spell out explicitly the corresponding requirements (compare (15))

x\us) = z'K) (27)

for those s, j that have the same data up to t.
For a given scenario uis we denote by x(uis) = xT'(uis) the vector of all decision
variables x'(uis), t = 1, . . . ,T in the individual scenario problem such as (25), (26),
ps the path probability of uis, c(uis) the vector of coefficients in the objective function
and X(uis) the corresponding set of feasible decisions described by linear constraints
that are fixed by the assumed values of scenario u>s. The multistage stochastic linear
program based on scenarios u>s, s = 1 , . . . ,S, can be now written as

s
minimize ^^psc(uis)Tx(uis) (28)

5= 1

subject to x(uis) e X(uis),s= [,..., S (29)

and x = Ux (30)
where x contains carefully grouped decision vectors x(uis) Vs and U is the 0 — 1
matrix: of coefficients of the nonanticipativity constraints of the form (27). For
instance, nonanticipativity of the first-stage decisions, i.e., the condition xl(ui\) =
i ' ( u 2 ) = . . . = x^(uis) can be expressed in the form (30) with U equal to the
Multistage Stochastic Programs: The State-of-the-Art and Selected Bibliography 163

following block permutation matrix

/ 0 I ... 0 \

0 0 ... J
\ I 0 ... 0 )

In standard formulations of decomposition algorithms, the nonanticipativity con-

straints (30) correspond to the Master problem. Similarly as for two-stage problems,
Lagrangian dual decomposition methods improve efficiency of scenario decomposi-
tion methods, cf. [34]. For instance, augmented Lagrangian method proposed in
[106] relaxes the nonanticipativity constraints by adding a quadratic penalty term
r||ai — Ux\\2 to the common Lagrangian function and uses suitable diagonal ap-
proximation of the penalty term to recover separability of the resulting objective
function. These ideas can be generalized to multistage stochastic programs with
linear constraints and convex quadratic or convex separable objective function; cf.
[5], [94], [116], [140].
The progressive hedging algorithm designed by Rockafellar and Wets [137], [154] is
another decomposition type procedure based on augmented Lagrangian method, this
time with a projection matrix at the place of U. It aims at progressive construction of
compromising aggregate solutions that sufficiently approximate the optimal solution
of the original stochastic program. It is not limited to linear problems; for its
applications and extensions see, e.g., [80], [109] - [111], [127] or [147].
Further recent algorithmic advances concentrate to exploitation of the structure of
the problem such as separability of the recourse matrices [95] or a network form of the
constraints [108], [112],[114], to parallel computations, e.g., [12],[28],[29],[83],[141],
[153], [167] and to possible reduction of the number of scenarios; cf. the EVPI model
reduction suggested in [33], [34] and included, e.g., into the system MIDAS [35]-[38],
or importance sampling techniques [30], [84].
All numerical techniques for solving multistage stochastic programs are demand-
ing as to the input data (including scenarios) and to the programming efforts. Their
final performance depends on the structure of the solved problem and on the men-
tioned input. In this context, generation of scenarios (discrete approximation of the
underlying probability distribution) is a very important task. In real-life application-
s, it has been solved from case to case and both parametric and nonparametric statis-
tical methods have been applied to historical data. An approach to generation of one-
dimensional scenarios was described in detail in [167] and other subsequent papers.
Generation of vector valued scenarios is more complicated even if the dimension can
be reduced by means of principal components method, factor analysis and regression
analysis or by sampling, see, e. g., [20], [21] - [23], [30], [35], [36], [87], [104], [105], [108].
If a multistage stochastic program is to be applied repeatedly as a part of a decision
support system, additional routines for data processing, scenario generation, simula-
tion and evaluation of results and of their robustness should be developed along the
whole path from theory to implementation and complemented by estimates of error
bounds and by suitable postoptimization techniques. For results concerning error
bounds for multistage stochastic programs see, e.g., [46], [148], postoptimization is
164 J. DUPAČOVÁ

treated in [43], stability results are presented in [56]. One of very important open
problems is a proper design of the information structure, namely, of the interstage
dependence that cannot be overlooked and that complicates scenario generation and
using sampling methods in general.
At this point, interdisciplinary nature of research in multistage stochastic pro­
gramming becomes evident. It is characterized not only by the trade-off between
advanced modeling, available data and algorithmic procedures but also by a strong
interplay between optimization, statistics, numerical methods and computer science.

4. BIBLIOGRAPHY ON MULTISTAGE STOCHASTIC PROGRAMMING

(For full annotation of collections of papers see References)

Classified as
ALG - computation, software, algorithms
APPL - non-financial application
B - approximation and bounds
CON - connection with stochastic control
EX - example
FIN - application in finance
GEN - general reference
P - for probabilistic constraints
S - for scenario generation
STAB - stability, postoptimization
TH - theoretical paper

[1] P. Abel and R. Thiel: Mehrstufige Stochastische Produktionsmodelle - Eine praxis-

orientierte Darstellung mit programmierten Beispielen. Fischer Verlag, Ffm 1981.
APPL/EX
[2] Z. Artstein: Probing for information in two-stage stochastic programming and the
associated consistency. In: Proc. of the 5th Prague Symp. on Asymptotic Statistics
(M. Huskova and P. Mandl, eds.), Physica Verlag, Heilderberg 1994, pp. 21-34. TH
[3] E.M.L. Beale, J. J. H. Forrest and C.J. Taylor: Multi-time -period stochastic pro­
gramming. In: [D], pp. 387-402. APPL
[4] E. M. L. Beale, G. B. Dantzig and R. D. Watson: A first order approach to a class of
multi-time-period stochastic programming problems. Math. Programming Study 27
(1986), 103-117. B/EX/STAB
[5] A. J. Berger, J. M. Mulvey and A. Ruszczynski: Solving stochastic programs with
convex objectives: Extending the DQA algorithm. TR SOR-93-16, Princeton Univ.,
1993. To appear in SIAM J. on Optimization. ALG
[6] J.R. Birge: Aggregation bounds in stochastic linear programming. In: System Mod­
elling and Optimization (A. V. Balakhrishnan and M. Thoma, eds.), Lecture Notes
in Control and Inform. Sci. 59, Springer, Berlin 1984, pp. 672-683. B/APPL
[7] J. R. Birge: Aggregation bounds in stochastic linear programming. Math. Program­
ming 31 (1985), 25-41. ALG/B/APPL
[8] J. R. Birge: Decomposition and partitioning methods for multistage stochastic pro­
grams. Oper. Res. 33(1985), 989-1007. ALG
Muitistage Stochastic Programs: The State-of-the-Art and Selected Bíhiiography 165

[9] J.R. Birge: An L-shaped method computer code for multistage stochastic linear
programs. In: [E-W], pp. 255-266. ALG
[10] J. Birge: Multistage stochastic programming. In: Tutorial lecture presented on 6th
Int. Conf. on Stochastic Programming, Udine 1992. GEN/EX
[11] J. R. Birge et al.: A standard input format for multiperiod stochastic linear programs.
COAL Newsletter 17(1987), 1-19. ALG
[12] J. R. Birge ct al.: A parallel implementation of the nested decomposition algorithm for
multistage stochastic linear programs. TR 94-1, Dept. of Industrial and Operations
Engineering, University of Michigan, Ann Arbor 1994. ALG
[13] J. R. Birge and J. K. Ho: The stochastic dynamic traffic assignment problem. TR 87-
16, Dept. of Industrial and Operations Engineering, Univ. of Michigan, Ann Arbor
1987. ALG
[14] J.R. Birge and J.K. Ho: Optimal flows in stochastic dynamic networks with conges-
tions. Oper. Res. 41 (1993), 203-216. ALG
[15] J. R. Birge and C. Rosa: Modeling investment uncertainty in the costs of global
CO2 emission policy. Submitted for Annals of Oper. Res. "Stochastic Programming:
Algorithms and Models" (Lillehammer 1994). APPL
[16] G. R. Bitran and II. H. Yanasse: Deterministic approximations to stochastic produc-
tion problems. Oper. Res. 32 (1984), 999 1018. B/APPL
[17] N.N. Bordunov: Extremum conditions for multistep control problems with random
convex mappings (in Russian). Kibernetika 1983, 1, 40-45; for translation see Cyber-
netics 19(1983). TH
[18] N. N. Bordunov: Extremum conditions for multistep stochastic programming problem
with smooth constraints. Cybernetics 20 (1984) 863-870; translation from Kiberneti-
ka 1984, 6, 77-82. TH
[19] S.P. Bradley and D. B. Crane: A dynamic model for bond portfolio management.
Management Sci. 19(1972), 139-151. FIN/S
[20] S.P. Bradley and D. B. Crane: Managing a bank bond portfolio over time. In: [D],
pp. 449-471. ITN/S
[21] D. R. Carino et al.: The Russell-Yasuda Kasai Model. Paper presented on 6th Int.
Conf. on Stochastic Programming, Udine 1992. FIN/S
[22] D. R. Carino et al.: The Russell-Yasud a Kasai Model. An asset/liability model for a
Japanese insurance company using multistage stochastic programming. Interfaces 24
(1994), 29-49. FIN/S
[23] D.R. Carino et al.: The Russell-Yasuda Kasai financial planning model. TR 93-
MSC-018, Frank Russell Comp., Washington 1993. FIN/S
[24] R. J. Casimir: The newsboy and the flower-girl. Omega 18 (1990), 395-398. EX/B
[25] K.D. Cocks: Discrete stochastic programming. Management Sci. 15 (1968), 72-79.
EX
[26] H. Dahl et al.: Some financial optimization models: II. Financial engineering. In:
Financial Optimization (S.A. Zenios, ed.), Cambridge Univ. Press 1993, pp. 37-71.
FIN/S
[27] G. B. Danlzig: Linear Programming and Extensions. Princeton Univ. Press, New
Jersey 1963, Chapter 25-3. GEN
[28] G. B. Dantzig: Need to do planning under uncertainty and the possibility of us-
ing parallel processors for this purpose. Ekonom.-Mat. Obzor 23 (1987), 121-135.
ALG
[29] G.B. Dantzig and P. W. Glynn: Parallel processors for planning under uncertainty.
Annals of Oper. Res. 22 (1990), 1-21. ALG
[30] G. B. Dantzig and G. Infanger: Multi-stage stochastic linear programs for portfolio
optimization. Annals of Oper. Res. 45 (1993), 59-76. FIN/ALG/S
166 J. DUPAČOVÁ

[31] X. de Groote, M. C. Noel and Y. Sraeers: Some test problems for stochastic nonlinear
multistage programs. In: [E-W], pp. 515-541. APPL/ALG
[32] M. A. II. Dempster: Introduction to stochastic programming. In: [D], pp. 3-59.
GEN
[33] M. A.H. Dempster: The expected value of perfect information in the optimal evo­
lution of stochastic problems. In: Stochastic Differential Systems (M. Arato et al.,
eds.), Lecture Notes in Inform. Control 36, Springer, Berlin 1981, pp. 25-40. TH/B
[34] [Link]. Dempster: On stochastic programming II: Dynamic problems under risk.
Stochastics 25 (1988), 15-42. TH
[35] M. A. H. Dempster and A. M. Ireland: A financial expert decision support system.
In: Mathematical Models for Decision Support (G. Mitra, ed.), NATO ASI Series
Vol. F48, Springer, Berlin, 1988, p. 415-440. FIN/ALG/S
[36] M. A. II. Dempster and A. M. Ireland: MIDAS: An expert debt management advisory
system. In: Data, Expert Knowledge and Decisions (W. Gaul and M. Schader, eds.),
Springer, Berlin 1988, pp. 116-127. FIN/ALG/S
[37] [Link]. Dempster and A.M. Ireland: Object oriented model integration in fi­
nancial decision support systems. Decision Support Systems 7 (1991), 329-340.
FIN/ALG/S
[38] M.A.H. Dempster and A.M. Ireland: Object-oriented model integration in MIDAS.
In: Proc. of the 22nd Annual Hawaii Int. Conf. on System Sciences, Vol. Ill: Deci­
sion Support and Knowledge Based Systems, IEEE Computer Society Press, 1989,
pp. 612-619. FIN/ALG/S
[39] J. Dupacova: Water resources system modelling using stochastic programming with
recourse. In: [K-P], pp. 121-133. APPL
[40] J. Dupacova: Stochastic optimization models in banking (in Czech). Ekonom.-Mat.
Obzor 27(1991), 201-234. FIN
[41] J. Dupacova: Stochastic programming models in banking. IIASA Laxenburg 1991.
FIN
[42] J. Dupacova: Applications of stochastic programming in finance. In: Atti del XVI
Convegno A. M. A. S. E. S., Treviso 1992, pp. 13-30. FIN
[43] J. Dupacova: Postoptimality for multistage stochastic linear programs. To appear in
Annals of Oper. Res. 51 (1995). STAB/S
[44] E. B. Dynkin: Concave stochastic dynamic programming (in Russian). Math. USSR-
Sb. 57(1972), 490-503. TH
[45] E. B. Dynkin: Optimal programs and stimulating prices in the stochastic models of
economic growth. In: Mathematical Models in Economy (J. Los and M. W. Los, eds.),
North-Holland, Amsterdam -1974, pp. 207-218. TH
[46] N. C. P. Edirisinghe and W. T. Ziemba: Tight bounds for stochastic convex programs.
Oper. Res. 40 (1992) 660-677. B
[47] M. J. Eisner and P. Olsen: Duality for stochastic programming interpreted as L. P.
in Ip-space. SIAM J. Appl. Math. 28 (1975), 779-792. TH
[48] M.J. Eisner and P. Olsen: Duality in probabilistic programming. In: [D], pp. 147-
158. TH
[49] Yu. Ermoliev and R. J.-B. Wets: Stochastic programming, an introduction. In: [E-
W], pp. 1-32. GEN
[50] Yu.M. Ermoliev and A.I. Yastremski: Stochastic Models and Methods in Economic
Planning (in Russian). Nauka, Moscow 1979. TH/APPL
[51] L. F. Escudero et al.: Production planning via scenario modeling. Annals of Oper.
Res. 42 (1993), 311-336. APPL/ALG/S
[52] I. V. Evstigneev: Optimal stochastic programs and their stimulating prices. In: Math­
ematical Models in Economy (J. Los and M. W. Los, eds.), North-Holland, Amster­
dam 1974, pp. 219-252. TH
MЫtistage Stochastic Programs: The State~oï-the~Art and Selected BiЫiography 167

[53] I.V. Evstigneev: Models of economic dynamics taking account of the uncertainty in
the production process (in Russian). Dokl. Akad. Nauk USSR 225(1975), 537-540
translation see Soviet Math. Dokl. 1975. TH
[54] I. V. Evstigneev: Lagrange multipliers for the problems of stochastic programming
In: Warsaw Fall Seminars in Mathematical Economics 1975 (M. W. Los, J. Los and
A. Wieczorek, eds.), Lecture Notes in Econom. and Math. Systems 133, Springer
Berlin 1976, pp. 34-48. TH
[55] I. V. Evstigneev: Measurable selection and dynamic programming. Math. Oper. Res
1 (1976), 267-272. TH/CON
[56] O. Fiedler and W. Romisch: Stability in multistage stochastic programs. Schwer-
punktprogramm der DFG Anwendungsbezogene Optimierung u. Steuerung, Report
426, Berlin 1993. To appear in Annals of Oper. Res. 51 (1995). STAB
[57] U. Fincke and H. W. Hamacher: On the expected value of stochastic linear programs
and (dynamic) network flow problems. EJOR 29 (1987), 307-316. APPL/B
[58] S.D. Flam: Nonanticipativity in stochastic programming. J. Optim. Theory Appl.
46(1985), 23-30. TH
[59] S. D. Flam: Lagrange multipliers in stochastic programming. SIAM J. Control Optim.
50(1992), 1-10. TH
[60] S.D. Flam and R. J.-B. Wets: Finite horizon approximates of infinite horizon stoch­
astic programs. In: Stochastic Optimization (V.I. Arkin, A. Shiraev and R. Wets,
eds.), Lecture Notes in Control and Inform. Sci. 81, Springer, Berlin 1986, pp. 339-
350. TH
[61] K. Frauendorfer: Multistage stochastic programming: Error analysis for the convex
case. Z. Oper. Res. 59(1994), 93-122. B/S
[62] K. Frauendorfer: Stochastic dynamic optimization: Modelling and methodological
aspects. Submitted for Proc. of 16th IFIP Conference 1993. B/CON/S
[63] S. J. Garstka and R. J.-B. Wets: On decision rules in stochastic programming. Math.
Programming 7(1974), 117-143. CON/TH
[64] II. I. Gassmann : MSLiP: A computer code for the multistage stochastic linear pro­
gramming problem. Math. Programming 47(1990), 407-423. ALG
[65] H.I. Gassmann: Optimal harvest of a forest in the presence of uncertainty. Canad.
J. For. Res. 19(1989), 1267-1274. APPL/B
[66] H. I. Gassmann and A. M. Ireland: Input/output standards for the MIDAS optimiza­
tion model. WP-90-2, School of Business Administration, Dalhousie Univ., Halifax
1990. ALG
[67] H.I. Gassmann and A.M. Ireland: Scenario formulation in an algebraic modelling
language. To appear in Annals of Oper. Res. ALG
[68] B. Golub et al.: Stochastic programming models for portfolio optimization with
mortgage-backed securities. To appear in EJOR 1994. FIN/S
[69] J. Gondzio and A. Ruszczyiiski: A sensitivity method for solving multistage stoch­
astic linear programming problems. In: Aspiration Based Decision Support Systems.
Theory, Software and Applications (A. Lewandowski and A. Wierzbicki, eds.), Lec­
ture Notes in Econom. and Math. Systems 331, Springer, Berlin, 1989, pp. 68-79.
ALG
[70] J. Gondzio and A. Ruszczyiiski: Sensitivity method for basis inverse representation
in multistage stochastic linear programming problems. J. Optim. Theory Appl. 74
(1992), 221-242. ALG/EX
[71] R. C. Grinold: A new approach to multi-stage stochastic programs. Math. Program­
ming Study 6 (1976), 19-29. TH
[72] R. C. Grinold: Model building techniques for the correction of end effects in multi­
stage convex programs. Oper. Res. 31 (1983), 407-431. TH
168 J. DUPAČOVÁ

[73] R. C. Grinold: Infinite horizon stochastic programs. SIAM J. Control Optim. 24

(1986), 1246-1260. TH
[74] L. Groenwegen and J. Wessels: Conditions for optimality in multi-stage stochastic
programming. In: [K-P], pp. 41-57. TH
[75] G. Hadley: Nonlinear and Dynamic Programming. Addison-Wesley 1964, Chapters
5, 10 and 11. GEN/CON/EX
[76] R. Hartley: Inequalities for a class of sequential stochastic decision processes. In: [D],
pp. 109-123. TH/B
[77] R. Hartley: Inequalities in completely convex stochastic programming. J. Math. Anal.
Appl. 75 (1980), 373-384. TH/B
[78] W. R. Heilmann : Solving multistage stochastic linear programming problems by non-
stationary stochastic dynamic programming. In: Methods of Oper. Res. 33 (1978),
173-184. CON
[79] W.R. Heilmann: Approximations and error bounds for multistage stochastic linear
programs. Part 2: Approximation by a 'larger' model. Methods of Oper. Res. 31
(1980), 349-356. CON/B
[80] T. Helgason and S. W. Wallace: Approximate scenario solutions in the progressive
hedging algorithm: A numerical study with an application to fisheries management.
Annals of Oper. Res. 31 (1991), 425-444. ALG/APPL
[81] J.L. Higle, J.C. Bean and R. L. Smith: Deterministic equivalence in stochastic infi­
nite horizon problems. Math. Oper. Res. 16(1990), 396-407. TH/EX
[82] M. Holmer et al.: Dynamic models for fixed-income portfolio management under
uncertainty. Report 93-11-01, Decision Sciences Department, The Wharton School,
University of Pennsylvania, Philadelphia 1993. FIN/S
[83] J. M. Hutchinson and S. A. Zenios: Financial simulations on a massively parallel con­
nection machine. Internal. J. Supercomputer Applications 5 (1991), 27-45. FIN/S
[84] G. Infanger: Planning Under Uncertainty. Solving Large Scale Stochastic Linear
Programs. Boyd and Fraser, Danvers 1994, Chapter 4. ALG/FIN/APPL
[85] P. Kail and S.W. Wallace: Stochastic Programming. Wiley, New York 1994, Chap­
ter 2. CON/ALG
[86] R. S. Kaplan and J.V. Soden: On the objective function for the sequential P-model
of chance-constrained programming. Oper. Res. 19(1971), 105-114. TH/P
[87] A. J. King: Multistage stochastic programming with sampling. Submitted for Annals
of Oper. Res. Stochastic Programming: Algorithms and Models (Lillehammer 1994).
S/FIN
[88] W. K. Klein Haneveld: A dual of a dynamic inventory control model: The determin­
istic and stochastic case. In: [K-P], pp. 67-98. TH/APPL
[89] W. K. Klein Haneveld: Duality in Stochastic Linear and Dynamic Programming.
Lecture Notes in Econom. and Math. Systems 214, Springer, Berlin 1986.
TH/CON/APPL
[90] V.V. Kolbin: Stochastic Programming. Reidel publ., Dodrecht 1977, Chapter IV.
GEN
[91] M.I. Kusy and W.T. Ziemba: A bank asset and liability management model. Oper.
Res. 34 (1986), 356-376. FIN
[92] M. Lane and P. Hutchinson: A model for managing a certificate of deposit portfolio
under uncertainty. In: [D], pp. 473-495. FIN/P/S
[93] P.O. Lindberg: A multistage stochastic transportation problem. TRITA-MAT-1971-
12, Dept. of Mathematics, Royal Inst, of Technology, Stockholm 1971. APPL
[94] F. Louveaux: A solution method for multistage stochastic programs with application
to an energy investment problem. Oper. Res. 28 (1980), 889-902. APPL/ALG
[95] F. Louveaux: Multistage stochastic programs with block - separable recourse. Math.
Programming Study 28 (1986), 48-62. APPL/ALG
Multistage Stochaatíc Programa: The State-of-the-Art and Selected Bihliography 169

[96] F. Louveaux and Y. Smeers: A stochastic model for electricity generation. Preprint
1980. APPL/S
[97] F. Louveaux and Y. Smeers: Optimal investments for electricity generation: A stoch-
astic model and a test problem. In: [E-W], pp. 445-453. APPL
[98] F. Mirozakhmedov: A three-stage problem of stochastic programming and the
method of its solution (in Russian). Dokl. Akad. Nauk Tadzhik. SSR 23 (1980),
694-697. APPL
[99] F. Mirozakhmedov: Mathematical Models and Methods for Production Control Tak-
ing Random Factors into Account (in Russian). Nauková Dumka, Kiev 1991.
APPL/ALG
[100] F. Mirozakhmedov and M.V. Mikhalevich: Modelling of optimal cultivated land
distribution by means of multistage stochastic programming (in Russian). Ekonomika
i Matematicheskie Metody J S ( 1 9 8 2 ) , 918-922. APPL
[101] 1). P. Morton: Algorithmic advances in stochastic programming. TR SOL 93-6,
Stanford University 1993. ALG
[102] D. P. Morton: Stoping rules for a class of sampling-based stochastic ptogramming
algorithms. TR NPSOR-94-003, Naval Postgraduate School, Monterey 1994. ALG
[103] J.M. Mulvey: Multi-stage financial planning systems. TR SOR-93-17, Princeton
Univ., 1993. In: Operations Research Models in Quantitative Finance (R. L.
D'Ecclesia and S.A. Zenios, eds.), Physic a-Verlag, Berlin 1994. FIN
[104] J. M. Mulvey: Financial planning via multistage stochastic programs. In: Mathemat-
ical Programming, State of the Art 1994 (J. R. Birge and K. G. Murty, eds.), Univ.
of Michigan 1994, pp. 151-171. FIN
[105] J.M. Mulvey: Generating scenarios for the Towers Perrin investment system. Re-
search Report, Princeton Univ., January 1994. FIN/S
[106] J.M. Mulvey and A. Ruszczyriski: A diagonal quadratic approximation method for
linear multistage stochastic programming problems. In: System Modelling and Op-
timization (P. Kail, ed.), Lecture Notes in Control and Inform. Sci. 180, Springer,
Berlin 1992, pp. 588-597. ALG
[107] J.M. Mulvey and A. Ruszczyriski: A new scenario decomposition method for large-
scale stochastic optimization. TR SOR 91-19, Princeton Univ., 1992. To appear in
Oper. Res. ALG
[108] J. M. Mulvey and H. Vladimírou: Stochastic network optimization models for invest-
ment planning. Annals of Oper. Res. 20 (1989), 187-217. FIN/ALG/S
[109] J. M. Mulvey and H. Vladimírou: Evaluation of a parallel hedging algorithm for stoch-
astic network programming. In: Impact of Recent Computer Advances in Operations
Research (R. Sharda et al., eds.), North-Holland, Amsterdam 1989, pp. 106-119.
ALG
[110] J.M. Mulvey and H. Vladimírou: Solving multistage stochastic networks: An appli-
cation of scenario aggregation. Networks 21 (1991), 619-643. ALG
[111] J.M. Mulvey and H. Vladimírou: Applying the progressive hedging algorithm to
stochastic generalized networks. Annals of Oper. Res. 31 (1991), 399-424. ALG
[112] J. M. Mulvey and H. Vladimírou: Stochastic network programming for financial plan-
ning problems. Management Sci. 38 (1992), 1642-1664. FIN/ALG
[113] K. Murty: Linear programming under uncertainty: A basic property of the optimal
solution. Z. Wahrscheinlichkeitstheorie verw. Geb. 10 (1968), 284-288. TH
[114] S. S. Nielsen and S.A. Zenios: Solving multistage stochastic network programs. RR
92-08-04, Wharton, 1992. To appear in Math. Programming. ALG/S
[115] S. S. Nielsen and S.A. Zenios: A stochastic programming model for funding single
premium deferred annuities. To appear in Math. Programming. FIN
170 J. DUPAČOVÁ

[116] M. C. Noel and Y. Smeers: On the use of the nested decomposition for solving non­
linear multistage stochastic programs. In: Stochastic Programming (F. Archetti et
al., eds.), Springer, Berlin 1985, pp. 235-245. ALG
[117] M.C. Noel and Y. Smeers: Nested decomposition of multistage nonlinear programs
with recourse. Math. Programming 57(1987), 131-152. ALG
[118] P. Olsen: Multistage stochastic programming with recourse: The equivalent deter­
ministic problem. SIAM J. Control Optim. 1J, (1976), 495-517. TH
[119] P. Olsen : When is a multistage stochastic programming problem well defined? SIAM
J. Control Optim. 11, (1976), 518-527. TH
[120] P. Olsen: Multistage stochastic programming with recourse as mathematical pro­
gramming m a Lp space. SIAM J. Control Optim. 14 (1976), 528-537. TH
[121] P. Olsen: Discretization ol multistage stochastic programming problems. Math. Pro­
gramming Study 6(1976), 111-124. TH
[122] M.V.F. Pereira and L.M.V.G. Pinto: Multi-stage stochastic optimization applied
to energy planning. Math. Programming 52 (1991), 359-375. APPL/ALG/S
[123] P. Popela: An object-oriented system for simulation tests in multi-stage stochastic
programming. Lecture presented at IFIP Workshop Stochastic Programming: Algo­
rithms and Models, Lillehammer, 1994. ALG
[124] A. Prekopa: Dynamic type stochastic programming models. In: Math. Methods
of Oper. Res. 1 (A. Prekopa, ed.), Akademiai Kiado, Budapest 1980, pp. 127-145.
TH/P
[125] A. Prekopa and T. Szantai: On multistage stochastic programming (With application
to optimal control of water supply). In: Progress in Oper. Research (Proc. Eger,
Hungary 1974), Colloquia Math. Soc. J. Bolyai 12, North-Holland, Amsterdam 1976,
pp. 733-755. TII/APPL/P
[126] A.H.G. Rinnooy Kan and L. Stougie: Stochastic integer programming. In: [E-W],
pp. 201-213. ALG
[127] S. M. Robinson: Extended scenario analysis. Annals of Oper. Res. 31 (1991), 385-398.
ALG
[128] R. T. Rockafellar: Lagrange multipliers for an N-stage model in stochastic convex
programming. In: Colloque d'Analyse Convexe et ses Applications (J. P. Aubin, ed.),
Lecture Notes in Econom. and Math. Systems 102, Springer, Berlin 1974, pp. 180
187. TH
[129] R. T. Rockafellar: On the equivalence of multistage recourse models in stochastic
optimization. In: Control Theory, Numerical Methods and Computer Systems Mod­
elling (A. Bensoussan and-J. L. Lions, eds.), Lecture Notes in Econom. and Math.
Systems 107, Springer, Berlin 1975, pp. 314-321. TH
[130] R. T. Rockafellar: Cost-to-go in multistage stochastic programming. Lecture present­
ed at the Minisymposium on Stochastic Programming, Berlin, January 22-23, 1994.
CON
[131] R.T. Rockafellar and R. J.-B. Wets: Continuous versus measurable recourse in
N-stage stochastic programming. J. Math. Anal. Appl. 48 (1974), 836-859. TH
[132] R.T. Rockafellar and R. J.-B. Wets: Nonanticipativity and £'-martingales in stoch-
astic optimization problems. Math. Programming Study 6(1976), 170-187. TH
[133] R.T. Rockafellar and R. J.-B. Wets: Measures as Lagrange Multipliers in Multistage
stochastic programming. J. Math. Anal. Appl. 60(1977), 301-313. TH
[134] R.T. Rockafellar and R. J.-B. Wets: The optimal recourse problem in discrete time:
Li-multipliers for inequality constraints. SIAM J. Control i6(1978), 16-36. TH
[135] R.T. Rockafellar and R. J.-B. Wets: Deterministic and stochastic optimization prob-
lems of Bolza type in discrete time. Stochastics 10 (1983), 273-312. TH/CON
Multistage Stochastic Progrsms: The Statc-of-thc-Art and Sehcicd Bibliography 171

[136] R. T. Rockafellar and R. J.-B. Wets: Generalized linear-quadratic problems of deter-

ministic and stochastic optimal control in discrete-time. SIAM J. Control Optim. 28
(1990), 810-822. TH/CON
[137] R. T. Rockafellar and R. J.-B. Wets: Scenarios and policy aggregation in optimization
under uncertainty. Math. Oper. Res. 16 (1991), 119-147. ALG
[138] R. T. Rockafellar and R. J.-B. Wets: A dual strategy for the implementation of the
aggregation principle in decision making under uncertainty. Applied Stoch. Models
and Data Analysis 8 (1992), 245-255. ALG
[139] Ch.H. Rosa: Pathways of economic development in an uncertain environment: A
finite scenario approach to the U. S. region under carbon emission restrictions. WP-
94-41, IIASA Laxenburg, 1994. APPL
[140] A. Ruszczynski: Augmented Lagrangian decomposition for sparse convex optimiza-
tion. WP-92-75, IIASA Laxenburg, 1992. To appear in Math. Oper. Res. ALG
[141] A. Ruszczynski: Parallel decomposition of multistage stochastic programming prob-
lems. Math. Programming 58 (1993), 201-228. ALG
[142] A. Ruszczynski: On convergence of augmented Lagrangian decomposition methods
for multistage stochastic programs. Submitted for Annals of Oper. Res. "Stochastic
Programming: Algorithms and Models" (Lillehammer 1994). ALG
[143] J. F. Shapiro: Stochastic programming models for dedicated portfolio selection. In:
Mathematical Models for Decision Support (B. Mitra, ed.), NATO ASI Series, Vol.
F48, Springer, Berlin 1988, pp. 587-611. FIN/S
[144] M.I. Tacsar: Optimal planning over infinite time interval under random factors.
In: Mathematical Models in Economy (J. Los and M.W. Los, eds.), North-Holland,
Amsterdam 1974, pp. 289-298. TH
[145] T. Tobias: Lagrange multipliers in linear multistage stochastic programming. Izv.
Akad. Nauk Eston. SSR Ser. Fiz.-Math. 28 (1979), 1. TH
[146] P. Varayia and R. Wets: Stochastic dynamic optimization approaches and computa-
tion. In: Mathematical Programming (M. Iri and K. Tanabe, eds.), KTKSci Publ.,
Tokyo 1990, pp. 309-332. CON
[147] S. W. Wallace and T. Helgason: Structural properties of the progressive hedging
algorithm. Annals of Oper. Res. 31 (1991), 445-456. ALG
[148] S. W. Wallace and Tiecheng Yan: Bounding multi-stage stochastic programs from
above. Math. Programming 61 (1993), 111-129. ALG/B/TH
[149] R. J.-B. Wets: Programming under uncertainty: The solution set. SIAM J. Appl.
Math. 14 (1966), 1143-1151. TH
[150] R. J.-B. Wets: Stochastic programs with recourse: A basic theorem for multistage
problems. Z. Wahrscheinlichkeitstheorie verw. Geb. 21 (1972), 201-206. TH
[151] R. J.-B. Wets: On the relation between stochastic and deterministic optimization.
In: Control Theory, Numerical Methods and Computer Systems Modelling (A. Ben-
soussan and J.L. Lions, eds.), Lecture Notes in Econom. and Math. Systems 107,
Springer, Berlin 1975, pp. 350-361. TH/CON
[152] R. J.-B. Wets: Duality for stochastic Bolza problems with applications to a model for
economic growth and a liquidity preference model. In: Mathematical Programming
and its Economic Applications (G. Castellani and P. Mazzoleni, eds.), F. Angeli,
Milano 1981, pp. 405-419. TH/CON/EX
[153] R. J.-B. Wets: On parallel processors design for solving stochastic programs. In:
Proc. 6-th Math. Progr. Symposium, Japanese Math. Progr. Society, Tokyo 1985,
pp. 13-35. ALG
[154] R. J.-B. Wets: The aggregation principle in scenario analysis and stochastic optimiza-
tion. In: Algorithms and Model Formulations in Mathematical Programming (S. W.
Wallace, ed.), Springer, Berlin 1989, pp. 91-113. ALG
 J. DUPAČOVÁ

[155] R. J.-B. Wets, Stochastic programming. In: Handbook on O R and MS, Vol. 1 (G. L.
Nemhauser et al., eds.), North-Holland, Amsterdam 1989, pp. 573-629. GEN
[156] R. J.-B. Wets: Challenges in stochastic programming. WP-94-32, IIASA, Laxenburg
1994. To appear in Math. Programming B. S/CON
[157] S.E. Wright: Primal-dual aggregation and disaggregation for stochastic linear pro­
grams. M a t h . Oper. Res. 19 (1994) 893-908. B
[158] S. E. Wright: Duality and consistent primal - dual approximation for linear - quadrat­
ic stochastic programming problems. Submitted for Math. Programming. TH/B
[159] A. I. Yastremski: Optimality conditions in stochastic programming. Cybernetics 16
(1980), 154-158; translated from Kibernetika (1980), 1, .136-139. TH
[160] A. I. Yastremski: Duality relations and optimality conditions in linear stochastic
programming. Cybernetics 23 (1987) 128-135; translated from Kibernetika (1987),
1, 102-107. TH
[161] D. B. Yudin: Multistage planning under conditions of incomplete information. Engrg.
Cybernet. i 0 ( 1 9 7 2 ) 953-964; translated from Izv. Akad. Nauk SSR Tekhn. Kibernet.
(1972), 6, 17-29. TH
[162] D.V. Yudin: Methods of constructing decision rules for multistage problems of stoch­
astic programming. Soviet. Math. Dokl. H (1973), 824-828; translated from Dokl.
Akad. Nauk SSSR 210 (1973), 779-782. TH
[163] D.V. Yudin: Multistage problems of stochastic programming. Soviet. Math. Dokl.
14 (1973), 787-790; translated from Dokl. Akad. Nauk SSSR 210 (1973), 545-548.
TH/P
[164] D.V. Yudin: Duality in multistage stochastic programming. Engrg. Cybernet. 11
(1973), 906-915; translated from Izv. Akad. Nauk SSSR Tekhn. Kibernet. 6 (1973),
24-31. TH
[165] D.V. Yudin: Mathematical Methods of Control under Uncertainty: Problems and
Methods of Stochastic Programming (in Russian). Soviet Radio, Moscow 1974, Chap­
ters 9-11. (Slovak translation: Alfa, Bratislava 1981.) GEN
[166] D.V. Yudin and E. V. Tsoi: A priori decision rules in multistage problems of stoch­
astic programming (in Russian). Ekonomika i Matematicheskie Metody 9 (1973),
945-961. TH/P
[167] S. A. Zenios: Massively parallel computations for financial planning under uncertain­
ty. In: Very Large Scale Computation in the 21st Century (J. P. Mesirov, ed.), SIAM,
Philadelphia 1991, pp. 273 294. FIN/S
[168] S.A. Zenios and M . S . Shtilman: Constructing optimal samples from a binomial
lattice. J. Inform. Optim. Sri. 14 (1993), 125-147. S/FIN

ACKNOWLEDGEMENT

I would like to thank all who contributed to completion of this bibliography and agreed
with inclusion of their not yet published papers.

(Received December 14, 1993.)

Multistage Stochastic Programs: The State-of-the-Art and Selected Bibliography 173

REFERENCES

A . Collections of p a p e r s
[B-W] J. R. Birge and R. J.-B. Wets (eds.): Stochastic Programming. Annals of Oper. Res.
30 and Si (1991).
[D] M.A.H. Dempster (ed.): Stochastic Programming. Academic Press, London 1980.
[E-W] Yu. Ermoliev and R. J.-B. Wets (eds.): Numerical Techniques for Stochastic Opti-
mization Problems. Springer, Berlin 1988.
[K-P] P. Kali and A. Prekopa (eds.): Recent Results in Stochastic Programming. Lecture
Notes in Econom. and Math. Systems 179, Springer, Berlin 1980.
[P-W] A. Prekopa and R. J.-B. Wets (eds.): Stochastic Programming 84. Math. Program-
ming Study 27 and 28. North-Holland, Amsterdam 1986.

B . A d d i t i o n a l references ( n o t i n c l u d e d in Section 4)
[Rl] E. Beale: On minimizing a convex function subject to linear inequalities. J. Roy.
Statist. Soc. Ser. B 17(1955), 173-184.
[R2] A. Charnes and W. W. Cooper: Chance-constrained programming. Management Sci.
5 (1959), 73-79.
[R3] G. B. Dantzig: Linear programming under uncertainty. Management Sci. 1 (1955),
197-206.
[R4] J. Dupacova: Stochastic programming with incomplete information: A survey of
results on postoptimization and sensitivity analysis. Optimization 18 (1987), 507-
532.
[R5] J. Dupacova: Stability and sensitivity analysis for stochastic programming. Annals
of Oper. Res. 27 (1990), 115-142.
[R6] J. Dupacova: Applications of stochastic programming under incomplete information.
To appear in J. Comput. Appl. Math. 56 (1995), 1-2.
[R7] P. Kail: Stochastic Linear Programming. Springer, Berlin 1976.
[R8] A. Prekopa: Logarithmic concave measures with application to stochastic program-
ming. Acta Sci. Math. 32 (1971), 301-316.
[R9] R. T. Rockafellar and R. J.-B. Wets: Stochastic convex programming: Kuhn-Tucker
conditions. J. Math. Econom. 2 (1975), 349-370.
[RIO] R.T. Rockafellar and R. J.-B. Wets: Stochastic convex programming: Singular mul-
tipliers and extended duality. Pacific J. Math. 62 (1976), 507-522.
[Rll] R.T. Rockafellar and R. J.-B. Wets: Stochastic convex programming: Basic duality.
Pacific J. Math. 62 (1976), 173-195.
[R12] R.T. Rockafellar and R. J.-B. Wets: Stochastic convex programming: Relatively
complete recourse and induced feasibility. SIAM J. Control Optim. H (1976), 574-
589.
[R13] G. Tintner: Stochastic linear programming with applications to agricultural eco-
nomics. In: Proc. of the 2nd Symp. in Linear Programming, Washington 1955,
pp. 197-207. .
[R14] R. Van Slyke and R. J.-B. Wets: L-shaped linear programs with application to op-
timal control and stochastic linear programs. SIAM J. Appl. Math. 17 (1969),
638-663.
[R15] S. W. Wallace and R. J.-B. Wets: Preprocessing in stochastic programming: The case
of linear programs. ORSA J. on Computing 4 (1992), 45-59.
[R16] D. W. Walkup and R. J.-B. Wets: Stochastic programs with recourse. SIAM J. Appl.
Math. .75(1967), 1299-1314.
174 J. DUPAČOVÁ

[R17] R. J.-B. Wets: Stochastic programs with fixed recourse: The equivalent deterministic
program. SIAM Rev, 16 (1974), 309-339.
[R18] R. J.-B. Wets: Stochastic programming: Solution techniques and approximation
schemes. In: Mathematical Programming - The State of the Art Bonn 1982 (J.
Bachem et al., eds.), Springer, Berlin 1983, pp. 566-603.

Prof. RNDr. Jitka Dupačová, DrSc, katedra pravděpodobnosti a matematické statisti­

ky, Matematicko-fyzikálnť fakulta Univerzity Karlovy (Department of Probability and
Mathematical Statistics, Faculty of Mathematics and Physics, Charles University),
Sokolovská 83, 18600 Praha. Czech Republic.