Professional Documents
Culture Documents
in various contexts: the CES utility functiofi and the logit discrete choice
model. Examples of both abund in the literature. For the CES. e nixit
and Stiglitz (1977) in monopolistic compefition, Atkinson and Stigk (1980)
in tax incidence, Katsoulacos (1984) in te&uologkai innovation, ~e~prn~~n
and ICrugman (1985) in international trade, etc. For the logit, see Domencich
and McFadden (1975) in transport modal choice, Anas (1982) in residential
location, de Falma et al. (1985) in spatial competition, Train et rl. (1957) in
telephone demand, etc.
These two mod& are especially popular m regional science as
by most of the papers contained in this issue.
The firstauthor thanks the Mgian govemment for financial support under the ‘Projet
d’ktion Cmmke’, contract ?4/8965.
**We haye learned from a referee that some materials p sted in 3 have
developed independently by A. Anas.
In this paper we intend to explore the linkages between the CES and the
logit. Indeed, we find they have similar roots in that the CES can be
interpreted as a s fit discrete choice model, wh% the logit can
interpret& as a representative consumer model. n the former case, we
that the demand system generated from a CES tility is formally
to the solution of a nested logit model in which the second stage is descri
terministic Cobb- utility function. In the latter case, we find
logit demands c associated with a representative consumer
maximizing a deterministic utility of the entropy type.
te e
Consider an economy with n+ 1 perfectly divisible commodities and N
consu income y. Commodities i = 1,. . . , n are the variants
of a uct; odity 0 is the (outside) composite
commodity.
The CES representative consumer supposes that the N consumers can be
represented by a preference ordering given in its simplest form by
f piXi+poXO'F (2)
,. i=l
(A”
.‘A’
xi = j/Pi9 (6)
where j7= y/(1 + a). The expression (6) is in the spirit of the first factor of (4).
The conditional indirect utility is therefore
where p is a positive constant and Ei a random variable with zero mean and
unit variance. If the Ei are independently, kientica!!y Curnbel-distributed, the
the probability for a consumer to choose i is given by the logit, i.e.,
.amj;p
Pr (i) = Pr( K= max vi, = nr eVhpil,rc
/=l....,n c *=1
e - PilP
_“^._
FrjijZXi= n
-=,e
- Pj!C (12)
c
i pjx;-i-poxo=Y
j=l
is satisfied for positive X,,. From (14), it immediately follows that the
site co odity is given by
aXi-NX’Xi
aPj PPO
TX- shows that demand for commodity i rises (falls) with the price of the
uxnposite commodity when pi e s (is less than) the
average price of the variants. In ly, commodity i
more (less) attractive because its relative opportunity cost in terms of the
composite commodity decreases (increas$ GSp. rises.
ax,
c-
1
aY -E’
This is evident from (16).
(21)
As
A
c Xi=1 and
j=l
c Pjxj-
j=l
2
S.P. Andersonet al., Two t-elatedmodelsofheterogeneity
The first term is the income effect [see (2O)J and, as will apparent from
the substitution matrix below, the second term is the substitution e&ct.
We now determine whether or not the demand system (13) and (15) can
from the maximization of a utility function; that is, can the Is
f discrete choice be represented by a model of utihty
~o~ti~uo~ choice? hethe: or not a demand system
a utility function is known as the ‘inte~ab~ty problem’.
if the matrix of substitution is symmetric and ne
~a~~~A(1965, p. 136)]. From the Slutsky equati
terms are
where hi is the compensated demand for commodity i. For tbp t-&t Aems=A
CLII.“E”. Y”II.UI.Y
aFu;err?,using (16~(22), we have the following substitution matrix S:
P2
-- pi,
x1x2 -x2(1-x2) ... c
X2! PO I PO j >
(24)
ituti r
Incqmating (27) into (26) and integmti.ng89 recwer U:
-m otherwise.
eristics model, as
possible only when the fium-her of
to the number of vGants.
To illustrate these ideas for the logic the ter c is interpreted iu the
discrete choice model as the city in consumer tastes.
Under the representative consumer approach p can be interpreted as the
taste for diversity. In a characteristics setting, this ter indicates the
de ofdis ion ofconsumers vis-&is the prod in the characteristics
SpWXt.
Anderson, S.P., k de Ptia and I.-F. T&se, 1?8?, maad far dS%rw&trl productt d&screte
choice models and the address approach, CO discussioc pay ~0. 8702 f
Catholiqcle de Louvain, Louvain-la-Neuve, Belgium).
S.P. Anderson et al., 7%~ related models of heterogeneity