Académique Documents
Professionnel Documents
Culture Documents
aaaaa
2002
1 Introduction
The firm must plan carefully the di¤erent marketing actions which it has to
take to introduce a new product in the market ([7], p. 307). Such coordinated
actions must be performed before the sales start as well as after it. The firm
communication has an important role to play, as one can observe for instance
in the car industry, where an awareness advertising campaign usually precedes
the distribution of a new car model.
Let kðtÞ be the goodwill level at time t, i.e. the product image by the con-
sumers, let wðtÞ be the advertising expenditure intensity and let yðtÞ be the
total discounted advertising expenditure up to time t, at the discount rate
r > 0. After assuming that
a) the goodwill kðtÞ depends on the expenditure rate through a special linear
di¤erential equation,
b) the firm image level at time 0 contributes to determine the initial product
image, and hence the product may have a positive goodwill level kð0Þ ¼
k 0 b 0,
c) a known level kT of goodwill must be reached in order to introduce prof-
itably a given product in the market,
d) the success of the product introduction depends on the time of the launch T
and on the final goodwill kðTÞ,
e) the time of the launch T is constrained to belong to a closed interval T J
½0; þy½, either bounded or unbounded,
we are led to formulate the problem of determining the advertising policy for
the product introduction and the launch time as the following optimal control
problem I ðkT ; TÞ:
yð0Þ ¼ 0; ð5Þ
T A T: ð7Þ
We assume that the objective function S : ½0; þy½ 3 ! R represents the utility
obtained from the introduction of the product and is a continuously di¤er-
entiable function, such that
minimize T; ð9Þ
T b 0: ð11Þ
The minimum time problem It ðk Þ is equivalent to the problem I ðk; ½0; þy½Þ
with the special objective function Sðk; t; yÞ ¼ t.
Let us call sustainable goodwill the parameter
bw a
ks ¼ ; ð12Þ
c
which is in fact the least upper bound of the reachable goodwill values, under
the assumptions of the problem I ðkT ; ½0; þy½Þ and the additional condition
k 0 < ks .
Then we can prove that
New product introduction 59
Theorem 1. A feasible solution for the problem I ðkT ; TÞ exists if and only if
and
8
>
> w; t A ½0; Tmin ðkT Þ;
>
<
^ ðtÞ ¼
w
>
> ckT 1=a
>
: ; t A ½Tmin ðkT Þ; T1 ;
b
We observe that the minimum time Tmin ðkÞ is a 1-1 function, and its in-
verse,
Corollary 1. A feasible solution for the problem I ðkT ; TÞ exists if and only if
either
or
and its inverse, the minimum reachable value of the goodwill at the time t, is
where T > 0 is fixed. The problem Iy ðk; TÞ just introduced is equivalent to the
problem I ðk; TÞ with the special objective function Sðk; t; yÞ ¼ y, in the
sense that a quadruple ðk ðtÞ; y ðtÞ; w ðtÞ; TÞ is an optimal solution of the
latter, where the optimal value of the objective functional (1) is y ðTÞ, if and
only if ðk ðtÞ; w ðtÞÞ is an optimal solution of the former and the optimal
value of the objective functional (17) equals y ðTÞ. We will denote by Y ðk; TÞ
the optimal value function of problem Iy ðk; TÞ and call it the minimum dis-
counted cost function. The minimum discounted cost function Y ðk; tÞ is de-
fined for all ðk; tÞ such that k a k max ðtÞ, because of Corollary 1; moreover,
Y ðk; tÞ ¼ 0 for all k a k min ðtÞ, because the zero control wðtÞ 1 0 is an optimal
solution of the problem Iy ðk; TÞ if and only if k a k min ðTÞ.
determines an admissible solution ðkðtÞ; wðtÞÞ for the problem Iy ðk; TÞ, if and
only if k a ^
k ðTÞ, where
^
k ðTÞ ¼ k 0 ecT þ ^
ky ð1 eððcþarÞ=ð1aÞÞT Þ: ð20Þ
and
^ ð1 aÞcks
ky ¼ : ð21Þ
c þ ar
New product introduction 61
Proof. In fact, the function (19) is an admissible control for problem Iy ðk; TÞ
and ^k ðTÞ is the value, at time T, of the goodwill function which is associated
to the control (19) through the motion equation (2) and the initial condition
given in (18). r
tð^
k ðTÞ; TÞ ¼ T; ð24Þ
Proof. In the special case that k ¼ k max ðTÞ, we notice that there exists a unique
(continuous) feasible solution, which is determined by the control funcion
wðtÞ 1 w. Of course it is the optimal solution and has the form (22) with t ¼ 0,
so that tðk max ðTÞ; TÞ ¼ 0.
Then, let us consider the problem Iy ðk; TÞ with the assumption that
k min ðTÞ < k < k max ðTÞ.
The ‘‘current value’’ Hamiltonian function is H c ðk; w; pÞ ¼ p0 w þ
p½bw a ck and if ðk ðtÞ; w ðtÞÞ is an optimal solution, then it satisfies the
Pontryagin’s Maximum Principle necessary conditions
pðtÞ ¼ pðTÞeðcþrÞðtTÞ
and hence, from (iv), (v) and (i), that a feasible solution to the necessary con-
ditions only exists with p0 ¼ 1, pðTÞ > 0 and k ðTÞ ¼ k. From the condition
(ii) we have that
62 A. Buratto, B. Viscolani
1 abpðTÞ
t¼T ln 1a :
cþr w
Let kt ðtÞ, t b 0, be the solution of equation (2) using the control wt ðtÞ and the
initial condition kt ð0Þ ¼ k 0 , then we obtain that
(
k 0 ect þ ^
ky ect eððaÞ=ð1aÞÞðcþrÞt ðeððcþarÞ=ð1aÞÞt 1Þ; t a t,
kt ðtÞ ¼ ð26Þ
ð^
k ðtÞ ks ÞecðttÞ þ ks ; t > t.
We observe that ðwt ðtÞ; kt ðtÞÞ, t A ½0; T, is an admissible solution of the adver-
tising cost problem which satisfies the transversality condition (v) if and only
if kt ðTÞ ¼ k, i.e. if and only if fðk; t; TÞ ¼ kt ðTÞ k ¼ 0.
The function fðk; t; tÞ is continuously di¤erentiable at all ðk; t; tÞ such that
t 0 t. Moreover, for all ðk; tÞ, the function fðk; t; tÞ is strictly decreasing in t
and has range k min ðtÞ k; k max ðtÞ k; therefore, there exists a function
tðk; tÞ, which is the unique solution to the equation fðk; t; tÞ ¼ 0.
From the definition (20) and the meaning of ^k ðtÞ given in the proof of
Lemma 1, we observe that tð^k ðtÞ; tÞ ¼ t. Moreover, we observe that the implicit
function theorem assumptions hold at all solutions ðk 0 ; t 0 ; t 0 Þ of fðk; t; tÞ ¼ 0
such that t 0 0 t 0 , i.e. such that k 0 0 ^k ðtÞ: hence tðk; tÞ is continuous at all ðk; tÞ
and continuously di¤erentiable for k 0 ^k ðtÞ. Furthermore, tðk; tÞ is strictly
decreasing in k and satisfies the conditions (23)–(25).
The unique solution we have obtained satisfies the assumptions of the
Mangasarian su‰ciency theorem ([10], Th. 4, p. 105). r
Remark. If k min ðtÞ < k a ^k ðtÞ, then we can solve explicitly the equation
fðk; t; tÞ ¼ 0 for the optimal phase function and obtain
1a ^ eððcþarÞ=ð1aÞÞt 1
tðk; tÞ ¼ ln ky : ð27Þ
aðc þ rÞ ke ct k 0
Lemma 2. If k 0 ^
k ðtÞ, then
qtðk; tÞ
< 0; for all t; ð28Þ
qk
New product introduction 63
and
qtðk; tÞ
> 0; for all k > ^k ðtÞ; ð29Þ
qt
moreover, for all k < ks ,
qfðk; t; tÞ
< 0; for all k; t; t;
qt
and
qfðk; t; tÞ
> 0; for all t; t and for k > ^k ðtÞ:
qt
Hence (28) and (29) are consequences of the implicit function theorem as ap-
plied to the equation fðk; t; tÞ ¼ 0.
Now, if k min ðtÞ < k a ^k ðtÞ, then (30) follows from (27). On the other hand,
if ^
k ðtÞ < k < k max ðtÞ, then the limit limt!þy tðk; tÞ exists, because tðk; tÞ is a
monotonically increasing function of t for t > t. If (30) did not hold, i.e. if the
limit were finite, then the equation fðk; t; tÞ ¼ 0 would not hold in the limit as
t ! þy. r
We observe that the discounted advertising cost over the interval ½0; t,
which is associated with the control function wt ðtÞ, defined in the statement of
Theorem 2, is
8
>
> 1a
>
< weððcþrÞ=ð1aÞÞt ðeððcþarÞ=ð1aÞÞt 1Þ; t a t,
c þ ar
zðt; tÞ ¼ ð31Þ
>
> 1a w rt
>
: rt
wðe e ððcþrÞ=ð1aÞÞt rt
Þ þ ðe e Þ; t > t.
c þ ar r
t a tðk; tÞ , k a ^
k ðtÞ: ð32Þ
In the following corollaries we state some general results on the minimum dis-
counted cost function.
64 A. Buratto, B. Viscolani
i) is continuous,
ii) is strictly monotonically increasing in k,
iii) is strictly monotonically decreasing in t, for k > ^k ðtÞ,
iv) has the limit
Proof. The continuity of Y ðk; tÞ follows from the continuity of zðt; tÞ and
tðk; tÞ. The monotonicity properties (ii) and (iii) hold because the discounted
interval advertising cost function zðt; tÞ is strictly decreasing in t, for which
the previous Lemma holds. The limit of Y ðk; tÞ follows from the limit of tðk; tÞ
provided by the previous Lemma. The di¤erentiability statement (v) follows
from the di¤erentiability of zðt; tÞ, for t 0 t, and of tðk; tÞ, for k 0 ^k ðtÞ. r
w
Y ðk max ðtÞ; tÞ ¼ zð0; tÞ ¼ ð1 ert Þ: ð35Þ
r
Proof. From Corollary 2 and Theorem 1 it follows that the function Y ðk; tÞ is
continuously di¤erentiable in k, as k 0 ^k ðtÞ. Then the sensitivity results for the
optimal value function and the adjoint function give qY ðk; tÞ=qk ¼ pðtÞ ([10],
p. 213). We observe that pðtÞ can be written as
ciple necessary conditions, as we have done for the minimum time problem
It ðk Þ and the advertising cost problem Iy ðk; TÞ. Nevertheless, most of the rel-
evant information on the optimal solution of problem I ðkT ; ½T1 ; T2 Þ, which is
obtainable from the Pontryagin conditions, has in fact been used to determine
the solutions to the problems It ðk Þ and Iy ðk; TÞ. Therefore we can exploit the
results of the analysis of Sections 3 and 4 in order to simplify the discussion of
the bounded time problem I ðkT ; ½T1 ; T2 Þ. In fact we argue in the present Sec-
tion that the original optimal control problem can be restated as a non-linear
programming problem and then solved by using the appropriate necessary
conditions and, possibly, some suitable numerical algorithms.
A first equivalence result is provided by the following theorem which ex-
ploits the knowledge of the minimum discounted cost function Y ðk; tÞ.
Theorem 3. The bounded time optimal control problem I ðkT ; ½T1 ; T2 Þ is equiv-
alent to the non-linear programming problem I ðkT ; ½T1 ; T2 Þ:
T A ½T1 ; T2 ; ð38Þ
a) if the problem I ðkT ; ½T1 ; T2 Þ has an optimal solution ðw ðtÞ; k ðtÞ; y ðtÞÞ
with final time T , then ðk ðT Þ; T Þ is an optimal solution to the problem
I ðkT ; ½T1 ; T2 Þ and f ðk ðT Þ; T Þ ¼ Sðk ðT Þ; T ; y ðT ÞÞ;
b) if the problem I ðkT ; ½T1 ; T2 Þ has an optimal solution ðk ; T Þ, then the
optimal solution ðw ðtÞ; k ðtÞÞ to the minimum cost problem Iy ðk ; T Þ de-
termines, through the equation (3) and the initial condition (5), the optimal
solution ðw ðtÞ; k ðtÞ; y ðtÞÞ to the problem I ðkT ; ½T1 ; T2 Þ.
Proof. a) Let ðw ðtÞ; k ðtÞ; y ðtÞÞ be an optimal solution of I ðkT ; ½T1 ; T2 Þ with
final time T , then y ðT Þ ¼ Y ðk ðT Þ; T Þ, because of the definition of
Y ðk; tÞ and of the fact that S is decreasing in y. Moreover, Sðk ðT Þ; T ,
y ðT ÞÞ b Sðk; t; yÞ, for all ðk; t; yÞ which is the final goodwill, time and
discounted advertising cost triplet associated with an admissible solution of
I ðkT ; ½T1 ; T2 Þ; in particular, Sðk ðT Þ; T ; y ðT ÞÞ b Sðk; t; Y ðk; tÞÞ, for all
ðk; tÞ which are feasible solutions to I ðkT ; ½T1 ; T2 Þ.
b) Let ðk ; t Þ, be an optimal solution of I ðkT ; ½T1 ; T2 Þ and let
ðw ðtÞ; k ðtÞÞ be an optimal solution of Iy ðk ; t Þ. The optimal value of the
kT a kmax ðT2 Þ
Proof. The problem I ðkT ; ½T1 ; T2 Þ has an optimal solution, because it has a
continuous objective function f ðk; TÞ and a nonempty and compact feasible
set. Hence, the thesis follows from the equivalence between problems stated by
the Theorem 3. r
t b 0; ð41Þ
T A ½T1 ; T2 ; ð42Þ
a) if wðtÞ 1 0 is an optimal control for the problem I ðkT ; ½T1 ; T2 Þ, then the
problem I~ðkT ; ½T1 ; T2 Þ has not any optimal solution;
b) if the problem I ðkT ; ½T1 ; T2 Þ has an optimal solution ðw ðtÞ; k ðtÞ; y ðtÞÞ
with final time T and phase t , then ðt ; T Þ is an optimal solution to the
problem I~ðkT ; ½T1 ; T2 Þ and kt ðT Þ ¼ k ðT Þ, zðt ; T Þ ¼ y ðT Þ;
a ) if the problem I~ðkT ; ½T1 ; T2 Þ has no optimal solution, then the problem
0
ðw ðtÞ; k ðtÞ; y ðtÞÞ with final time T and final goodwill k ðT Þ, then it must
be an optimal solution of the advertising cost problem Iy ðk ðT Þ; T Þ, so that
it must have the form (22), because of Theorem 2.
The control wðtÞ 1 0 may be optimal for the problem I ðkT ; ½T1 ; T2 Þ, if
kT a k min ðT1 Þ: in such a case the optimal control function has not the form
(22) for any real t and the problem I~ðkT ; ½T1 ; T2 Þ has not any optimal solu-
tion.
If kT > k min ðT1 Þ, then the problem I ðkT ; ½T1 ; T2 Þ has an optimal solution
with control w ðtÞ ¼ wt ðtÞ, for some t b 0: then the problem I~ðkT ; ½T1 ; T2 Þ
has a bounded feasible set and has the optimal solution ðt ; T Þ. If kT a
k min ðT1 Þ, then either the problem I ðkT ; ½T1 ; T2 Þ has the optimal control
w ðtÞ 1 0, and then T ¼ T1 and k ðT1 Þ ¼ k min ðT1 Þ, or I ðkT ; ½T1 ; T2 Þ has the
optimal control is w ðtÞ ¼ wt ðtÞ, for some t b 0: in the first case the prob-
lem I~ðkT ; ½T1 ; T2 Þ has not any optimal solution, whereas in the second case it
has an unbounded feasible set, but has the optimal solution ðt ; T Þ. r
We obtain in particular that the fixed time new product introduction prob-
lem I ðkT ; fTgÞ has an optimal solution, as far as kT a k max ðTÞ. Such a prob-
lem is characterized by an exogenously given launch time and represents e.g.
those cases in which the product has to be introduced in the market on the oc-
casion of an exhibition or other special event. From the Theorems 3 and 4 we
can derive immediately two special representations of the problem I ðkT ; fTgÞ
as a 1-dimensional non-linear programming problem.
Corollary 5. The fixed time new product introduction problem I ðkT ; fTgÞ is the
problem of maximizing the function F ðkÞ ¼ f ðk; TÞ on the bounded interval
½kT ; kmax ðTÞ; it has an optimal control w ðtÞ and
i) either w ðtÞ 1 0,
ii) or w ðtÞ ¼ wt ðtÞ, for some t b 0.
In the latter case t is a maximum point of the function GðtÞ ¼ gðt; TÞ,
ii.1) either on the bounded interval ½0; tðkT ; TÞ, if kmin ðTÞ < kT a kmax ðTÞ,
ii.2) or on the unbounded interval ½0; þy½, if kT a kmin ðTÞ.
We recall that the objective functions f ðk; TÞ and gðt; TÞ of the nonlinear
programming problems I ðkT ; ½T1 ; T2 Þ and I~ðkT ; ½T1 ; T2 Þ are continuous and
so are F ðkÞ and GðtÞ from Corollary 5. The functions f ðk; TÞ and gðt; TÞ are
also continuously di¤erentiable except at the points ð^k ðTÞ; TÞ, the first one,
and ðt; tÞ, the second one. No general monotonicity or concavity properties of
the objective functions hold: therefore we cannot state a priori any su‰ciency
or uniqueness result for the optimal solutions.
The feasible set of probem I ðkT ; ½T1 ; T2 Þ is compact and convex, whereas
the one of problem I~ðkT ; ½T1 ; T2 Þ may be compact and convex as well as not.
As for the I ðkT ; fTgÞ-equivalent 1-dimensional problems, we observe that
the objective functions F ðkÞ and GðtÞ are continuously di¤erentiable except at
the points ^ k ðTÞ, the first one, and T, the second one. In general we cannot
state whether F ðkÞ and GðtÞ are concave or not.
68 A. Buratto, B. Viscolani
6 Conclusion
Acknowledgements. We thank Francesco Casarin, for some useful management comments, and
Paolo Pellizzari, for an illuminating observation concerning the non-linear programming repre-
sentation of the new product introduction problem.
References
1. Buratto A, Viscolani B (1994) An optimal control student problem and a marketing coun-
terpart. Mathematical and Computer Modelling 20:19–33
2. Buratto A, Viscolani B (1996) Errata to: An optimal control student problem and a marketing
counterpart. Mathematical and Computer Modelling 23:123
3. Collesei U (1994) Marketing. CEDAM, Padova
4. Datar S, Jordan C, Kekre S, Rajiv S, Srinivasan K (1997) New product development struc-
tures and time to market. Management Science 43:452–464
5. Feichtinger G, Hartl RF, Sethi SP (1994) Dynamic optimal control models in advertising:
recent developments. Management Science 40:195–226
6. Hendricks KB, Singhal VR (1997) Delays in new product introductions and the market value
of the firm: the consequences of being late to the market. Management Science 43:129–142
7. Kotler P (1997) Marketing Management: Analysis, Planning, Implementation, and Control.
Prentice-Hall, Upper Saddle River
8. Lilien GL, Kotler P, Moorthy KS (1992) Marketing models. Prentice-Hall, Englewood Cli¤s
9. Nerlove M, Arrow KJ (1962) Optimal advertising policy under dynamic conditions. Eco-
nomica 29:129–142
10. Seierstad A, Sydsaeter K (1987) Optimal control theory with economic applications. North-
Holland, Amsterdam