Docstoc

Oligopoly and Luce’s Choice Axiom

Document Sample
Oligopoly and Luce’s Choice Axiom Powered By Docstoc
					                 Oligopoly and Luce’s Choice Axiom
                       Simon P. Anderson and André de Palma∗
                                       October 3, 2011


                                             Abstract
          We apply Luce’s choice axiomatic framework to oligopoly pricing of quality differ-
      entiated goods. The demand system is a probabilistic comparison of surpluses across
      products. Zero demands arise naturally, in contrast to the related CES and Mixed Logit
      models. With asymmetric products, high mark-ups and high demands are driven by
      high quality-costs. The oligopoly price equilibrium delivers a simple surplus-split prop-
      erty. We reconcile the model with standard consumer theory by introducing income,
      and hence generate a Representative Consumer formulation, which has a quadratic
      form in a central case. We further introduce a preference representation based on the
      Gabszewicz-Thisse vertical quality formulation.
          Keywords: Choice Axiom, Probabilistic choice models, Oligopoly, Pricing, Repre-
      sentative consumer.
          JEL Codes: D11, D43, L13.
  ∗
    University of Virginia, sa9w@virginia.edu; and Ecole Normale Supérieure de Cachan and Centre
d’économie de la Sorbonne, andre.depalma@ens-cachan.fr. The first author thanks the NSF for support.
We thank Jacques for his perpetual inspiration.




                                                 1
1         Introduction
Luce (1959) postulated a choice probability system which has endured in psychology, eco-
nomics, marketing, and regional science.1 It stems from Luce’s Choice Axiom, and offers
a simple formulation of choices in terms of scale values attributable to alternatives. These
scale values are treated by Luce as exogenous, though their values are typically determined
by the economic decisions of other actors, most notably firms. Our intention in this paper is
to model the endogenous determination of these values by specifying and solving the pricing
game between firms. Indeed, the simplest economic formulation of Luce’s choice probabili-
ties is to use surpluses as scale values, with the “surplus” here measured as a “reservation
price” minus price paid. This is the formulation that we investigate in this paper as a basic
demand system in the context of firms selling differentiated products.
    Luce’s model is based on the Choice Axiom, and highlights the IIA property (first pointed
out by Debreu, 1960, in the famous Red-Bus/Blue-Bus paradox). Both the Logit and CES
models (see Anderson, de Palma, and Thisse, 1988a) embody the IIA property. Several
properties of the formulation we present are shared with those models. However, there are
also substantial differences. For example, Logit and CES models have been criticized because
positive amounts are always consumed regardless of prices (a feature shared with mixed logit
models), and this aspect is not always apparent in data (on, say, import consumption by
sector across countries: see Baldwin and Harrigan, 2007). The current specification implies
choice probabilities are zero when prices get above “reservation values.”
    We apply the demand model that is generated by the Luce approach to the problem of
competing oligopolists selling products that are differentiated by qualities, as manifested in
the “reservation values.” Equilibrium existence is readily demonstrated because strategy sets
are compact, prices being naturally comprised between marginal cost and reservation value,
and profit functions are strictly quasi-concave. The latter property is easily shown using the
monotonicity of a core function within the profit derivative.
    The oligopoly prices generate a simple surplus-split property between firms and con-
sumers. Under asymmetric firm qualities and marginal costs of production, there is a simple
ranking between mark-ups, equilibrium outputs, and firm qualities and costs. The quali-
tative nature of this last relation is shared with the Logit oligopoly model, and this is one
instance where those similar roots are manifested here. Specifically, firms with high mark-
ups are associated to high demands and high profits. What drives these relations is a high
quality-cost margin.
    These relationships sustain once we introduce non-purchase options into the basic Luce
framework. Such “outside” options also enable us to soften some of the stronger properties
of the model.
    Our next contribution is to reconcile the proposed Lucian model with standard consumer
theory. This we do by introducing income into the choice probability system. This enables
us to generate a Representative Consumer formulation for the choice probability system,
which has a simple functional form and is quadratic in a central case.
    1
        See also the perspective 20 years later (Luce, 1977).


                                                         2
   Finally, we propose an alternative formulation for introducing income, based on the
vertical differentiation formulation of Gabszewicz and Thisse (1980). This appealing form
takes as surplus function a product of the product quality and income net of purchase price.


2       The model
2.1     Consumer Choice Model
Coming from a background in mathematical psychology, Luce (1959) viewed the individual
decision process as inherently probabilistic, but in a manner that could be described by
mathematical formulae. In common with the economics formulation of utility theory, Luce
envisioned the use of scale variables that would reflect the attractiveness of alternatives,
although not in the economics fashion of always choosing the option with the greatest scale.
Instead, Luce postulated axioms, most importantly the Choice Axiom, which implied that
the probability an option  would be selected was an increasing function of its (positive)
scale,  , and decreasing in the scales attributed to other alternatives. Luce proved that his
axioms imply that the probability, P , that alternative  is chosen from a choice set  is:2
                                            
                                      P = X              = 1                                    (1)
                                             
                                             ∈

Luce does not specify how the  ’s are determined. Indeed, as argued in the Introduction,
they are often endogenously determined through the actions of other agents, most notably
(for our context) by the firms choosing the prices of the various competing alternatives. With
this application in mind and considering the economic context for the consumer of which
alternative to choose, the simplest formulation is to use the following specification for the
scale values:
                             = max ( −   0)     = 1                       (2)
where  is the price of option , and we understand the individual to be interested in buying
a single unit of the commodity (a motor car, or a chocolate bar with afternoon tea).3 The
    2
     Luce’s (1959) axioms are as follows. Let P () be the probability that the individual selects  from
the subset  , with  ∈  . The most important axiom is the path independence property which states that
P () = P () P () with  ∈  ⊆  . The second axiom is: if  and  are such that P{} () = 0 then
P () = P \{} ( \{} ). Therefore, the analysis can be restricted to the case where P{} () ∈ (0 1).
Under those assumptions, Luce’s theorem (Luce, 1959, p. 23) states that the choice probability of alternative
 in subset  is given by the ratio between a factor which depends on  only, and the sum of all other factors
(the normalization factors):
                                                          ()
                                              P () = X         
                                                            ()
                                                       ∈

with  () a positive real-valued function.
   3                                                                       
     Another simple form that could bear further investigation is  =      .


                                                      3
term  is naturally interpreted as a reservation price insofar as the individual will not buy
 if    , and, since the individual only buys one of the options available under the choice
rule (1), the terms  −  will be interpreted throughout this paper as Lucian surpluses.
However, these are not traditional economic consumer surpluses in the usual sense of the
term because the consumer does not necessarily buy with certainty the alternative with the
greatest surplus under the choice rule (1).
     Note that it is implicitly assumed that income is not binding here, which makes most
sense if the good under consideration is a small fraction of income (like a chocolate bar,
though not, for most of us, a motor car). This case is also more pertinent when income
effects are effectively small. We discuss below how income can be introduced into the model,
and how income effects affect choice probabilities.
     The choice probability system, which is also the demand system faced by the firms,
provides a simple relation between the Lucian surpluses. For the analysis that follows, we
parameterize the basic Luce model (1) by ascribing a power   0 to the scales,  in (2).
This means that the larger is , then the more discriminating is the consumer in terms of
choosing the option with the greatest Lucian surplus.
     Thus, in this paper, we suppose that the choice probabilities, P , follow the form:4

                                             ( )
                                  P = P                          = 1                (3)
                                           =1   ( )

with  ≥ 0 and with  = max ( −   0) ≥ 0. If all the  are zero, then we assume that
P = 0 for all . Note that choices are equally split among options when all the  ’s are equal.
We also find the equal-split result for unequal  ’s when  → 0 (and then P → 1). This
case corresponds to a lack of discriminatory power of the economic variable  −  (see (2)).
At the other extreme, consider  → ∞. If alternative  is dominated (i.e., there is some 
for which    ), then lim→∞ P = 0. If  = max=1  , then

                                              lim P = 1
                                             →∞

where  ∈ {1  } is the number of alternatives which have the (common) largest value of
 .
     These choice probabilities satisfy the well known IIA property (with the blue-bus/red
bus paradox articulated by Debreu, 1960) that the ratio of any two choice probabilities is
independent of the availability (or price) of any third alternative. Note that the denominator
of the expression in (3) is always positive for all  once we introduce an active outside option,
considered below as  = 0 (and so it is active when 0  0).
     The choice probability system obeys two key properties, which are readily verified by
differentiation.

Lemma 1 For all   0 and   0:
  4
      In the standard Luce model above,  = 1.


                                                        4
   The own-price demand derivative is:
                       P       P (1 − P )
                           = −                0,           = 1       (4)
                                   
The cross-price demand derivative is:
                      P     P P
                           =        0,         = 1       6=     (5)
                            
   The first part shows demands slope down; the second part shows that products are
substitutes. We now turn to the firm side of the analysis.

2.2       Firm pricing
There are  firms, and  differentiated products. Firm  = 1   sells product  at constant
marginal cost of production,  , and the “quality” of ’s product,  , is exogenous. We assume
that    for all ; otherwise (as argued below) a firm would be inactive. Firm ’s net
profit when it sets price  and its rivals set prices − , is
                                    (  − ) = ( −  ) P    = 1  
Firms compete in prices, and we seek a Bertrand Nash price equilibrium. Note that the best
response price of firm  satisfies:      . When  =  , Firm  is left with no profit,
while when  =  , the consumer buying good  is left with no surplus and so never buys.
The first-order derivative for pricing, using (4), is:
                      (  − )                     P (1 − P )
                                     = P − ( −  )                 = 1        (6)
                                                           
This is simply the extra profit from the existing consumer base, P , minus the mark-up lost
on the consumers who switch out due to the price rise.
   For   0, we have
                                               µ                         ¶
                (  − )    P (1 − P )       
                               =                             − ( −  )     = 1   (7)
                                             (1 − P )
The factor before the brackets is positive, and hence the sign of  ( − ) depends on the
                                                                             
sign of the term in brackets.
                                      
    The first term in the bracket ( (1−P ) ) is strictly positive for    and zero for  ≥  .
This term is decreasing in  since both  and P are decreasing in  . The second term in
the bracket (minus the mark-up) is also decreasing in  ; it is positive if and only if  ≥  .
                                                                 
There is therefore a unique intersection to the equation (1−P ) = ( −  ) corresponding to
Firm ’s best-response price:
                                                       
                                      =  +                .                                 (8)
                                                   (1 − P )
Because the profit function is strictly increasing for prices below the intersection ((7) is
positive),5 and decreasing for prices above it ((7) is negative), the profit function is strictly
  5                                                                         ( − )
      Firm ’s best-reply price is always above its marginal cost since                   0.

                                                         5
quasi-concave, given 1   [ ]   . This indicates that indeed (8) is the best response.
    The proof of existence follows. Firm ’s strategy is the compact set [   ]. Its profit
function is continuous in  and strictly quasi-concave in  ; hence reaction functions are
continuous and map from the compact and convex set (a hypercube) [1  1 ] ×  × [   ]
into itself: Kakutani’s fixed point theorem then ensures there is a fixed point, which is a
Nash equilibrium.

Proposition 1 There exists a price equilibrium. In equilibrium, all firms for which   
are active, and have strictly positive mark-ups. The other firms are inactive. The equilibrium
prices solve the system
                                   
                                           = ( −  ) ,  = 1 
                                (1 − P )

   We prove uniqueness, below, in a slightly more general case. The next Proposition shows
how mark-ups, Lucian surpluses, outputs, and profits are ranked. To do so, we first label
firms by decreasing quality-cost margins, so that 1 − 1 ≥  ≥  −  .

Proposition 2 (Quality). The equilibrium rankings of the endogenous economic variables
reflect the quality-cost rankings and conversely - in any equilibrium:
    1. Firms’ mark-up ranking: 1 − 1 ≥  ≥  −  ;
    2. Lucian surplus ranking: 1 − 1 ≥  ≥  −  ;
    3. Market size ranking: P1 ≥  ≥ P ;
    4. Profit ranking:  1 ≥  ≥  ;
    where any strict inequality implies a strict inequality, and any equality implies an equality
in the above rankings.
                                                                                                  
                                                                                                    
    Proof. Suppose that  −  ≥  −  for some pair { }. By (8), then (1−P ) ≥ (1−P ) .
         
Since (1−P ) is increasing in  −  (note that P is increasing in  −  ), then  −  ≥  −  .
But then the last statement ( −  ≥  −  ) together with the first ( −  ≥  −  )
implies that  −  ≥  −  . Clearly, if  −  ≥  −  then we cannot have  −    − 
and so  −  ≥  −  implies  −  ≥  −  . The profit rankings follow from the
mark-up and market size rankings. Similar arguments show that  −    −  implies
strict inequality in the other rankings, while  −  =  −  implies equality in the other
rankings.
    Hence a firm delivering a higher quality-cost earns a higher equilibrium mark-up. This
in turn means that it has a higher equilibrium choice probability, and so delivers a higher
net (Lucian) “surplus” to consumers, and earns more profit. In a broader context, insofar as
a free entry equilibrium (with all firms facing the same entry cost) tends to select the firms
with highest profits, Proposition 2 suggests that the highest quality-cost firms would enter
the market.
    For a first appraisal of Proposition 2, suppose that all firms delivered the same quality, ,
but differed by production cost, . Then, firms with lower costs would set larger mark-ups,
but they would also set lower prices. These lower prices ensure higher equilibrium demands.

                                                   6
The fact that a firm has a cost advantage enables it to set a lower price while still setting a
larger mark-up; and its profit is higher on both counts.
    Second, suppose that all firms had the same costs, but differed over qualities. Then, a
higher quality enables a firm to set a higher price, and so get a higher mark-up, but, by
delivering a higher Lucian surplus to consumers, to still get a higher equilibrium demand.
This means that high prices are associated to high demands. The reason is that both are
related through the first-order conditions, and are generated by the economic fundamental
higher quality. Qualitatively similar results have been shown by the authors for the Logit
model (see Anderson and de Palma, 2001).

2.3       Surplus split
The equilibrium is remarkable for the surplus split between consumers and firms; and we
highlight this because it contrasts with the logit and CES models. Let ∆ ≡ ( −  ) be the
quality-cost associated to Firm . We can rewrite the pricing expression (8) as:

                                                       ∆
                                      −  =                    ∆ .                                    (9)
                                                  (1 − P ) + 1

This gives  =  for the monopoly case (P = 1): a monopolist charges the reservation
price of the consumer.6 Clearly      (so that   0).
    An alternative way to write the equilibrium price is in terms of the Lucian surplus:

                                                       (1 − P ) ∆
                                         −  =                                                        (10)
                                                       (1 − P ) + 1
This expression immediately gives rise to the following result (where we recall from Propo-
sition 2 that lower sales are made by firms with lower quality-costs).
                                                                                           ∆
Proposition 3 In equilibrium, the surplus, ∆ , is split between a mark-up              (1−P )+1
                                                                                                     for each
                                         (1−P )∆
firm, and a "Lucian net surplus"              for the consumer. Hence the ratio of Lucian to
                                         (1−P )+1
firm surpluses is  (1 − P ), which is higher for firms with lower quality-costs. Equivalently,
the share of firm surplus in quality-cost is higher for firms with higher quality-cost margins.

    This result indicates that firms with higher quality-costs are more successful in appropri-
ating surplus.

Corollary 1 (i) If  → 0, then all prices tend to reservation values ( →  for all  =
1  ).
     (ii) If  → ∞, then 1 → (1 − 2 ) + 2 while  →  for all   1.
  6
      By contrast, a logit or CES formulation for monopoly with no outside good gives an infinite price.




                                                        7
    Proof. (i) This follows directly from (10).
    (ii) By Proposition 2, P ( P1 ) is bounded strictly below 1 for   1 so that  (1 − P ) →
∞. Hence from (9),  →  for   1. The same argument applies to Firm 1 if 1 −1 = 2 −2 ,
so consider henceforth the case where 1 − 1  2 − 2 . From (10),
                                                     ∆1
                                    1 − 1 =           1                                  (11)
                                                1+   (1−P1 )

Either  (1 − P1 ) goes to zero as  → ∞, in which case 1 → 1 : but this contradicts that
Firm 1 has the highest consumer surplus (Proposition 2: Firm 2, pricing at marginal cost
delivers surplus 2 − 2 , which positive by assumption that 2 is active). Likewise  (1 − P1 )
cannot go to infinity because, from (11) then 1 → 1 : but this contradicts that Firm 1 is
maximizing profits since it can do better setting a positive mark-up and has the leeway to do
so against Firm 2 pricing at its marginal cost. Therefore, the remaining case has  (1 − P1 )
reaching a finite, non-zero limit. It can be shown that
                              P ³  ´
                                 =2 1          X µ  ¶
                                                  
           (1 − P1 ) =         P ³  ´ ≈          1
                            1 + =2 1           =2
                                    ⎛                           ⎞
                            µ ¶           µ ¶           µ ¶         µ          ¶
                               2 ⎜ ⎜1 +     3             ⎟ ⎟≈       2 − 2
                      =                          +  +                            
                               1 ⎝          2            2 ⎠           1 − 1
                                           |        {z         }
                                                      ≈0

This expression is neither zero nor infinite as  → ∞. Hence 1 − 1 must tend to 2 − 2 .
This case yields the value given in the Corollary.
   The first of these results says that if there is no discriminatory power across products,
then all products tend to individual “monopoly” solutions by pricing at reservation values.
The second result corresponds to the case of perfect discriminatory power, in the sense that
consumers choose the product with the highest net quality (Lucian surplus),  −  . The
result is the classic Bertrand one that all prices tend to marginal cost except the highest
quality product, which prices to just beat the keenest competition. This means pricing at
that rival’s marginal cost adjusted by the top firm’s quality advantage.

2.4    Symmetric equilibrium
Suppose now that all “qualities” are equal to , and all costs are equal to , and let ∆ ≡
( − ) be the symmetric quality-cost. Then (9) becomes:
                                                 ∆
                                 −  =    ¡   1
                                                    ¢   ∆
                                              1−  +1
The equilibrium price is decreasing in  (since the demand is more elastic as  increases)
and decreasing in  (more competition drive prices down). When  → 0 (no discriminatory

                                                8
power), we have  → , while when  → ∞ (perfect discriminatory power), we have
 → .
   The alternative way to write the equilibrium price in terms of the Lucian surplus (from
(10)) is:
                                              ¡    1
                                                     ¢
                                              1−  ∆
                                 =  − ¡      1
                                                    ¢     
                                             1−  +1
Comparing this expression with the previous one indicates that the ratio of Lucian to firm
surpluses is (−1) , which is increasing in . This is a classic competition result that con-
                
sumers gain as more firms compete. Note though that the limit is not perfect competition
but rather “monopolistic competition”: as  gets large, the split tends to .


3      Outside option
Suppose now there is also a non-purchase (outside) option with fixed “scale” level 0 ≥ 0;
for example, eating at home rather than eating in the restaurant. The case 0 = 0 will be
seen to concur with the results already found. The choice probabilities then become

                                          ( )
                        P =              P                      = 0 1  
                               (0 ) +     =1 ( )
                                                         



The first order conditions look the same as before (see (6)), except with 0 as an argument in
P . This gives a polynomial system of equations to solve. The proof of equilibrium existence
follows exactly the same lines as before. We here show the uniqueness of the equilibrium.

Proposition 4 There exists a unique price equilibrium for the Lucian model with an outside
option.

    Proof. Existence follows from the same argument as the case without an outside option;
uniqueness is given now by showing that there is a single solution to the equations defining the
(interior) first order conditions. The first-order conditions solve (6), which can be rewritten
as
                                              
                              (1 − P ) =             = 1  
                                           ∆ − 
or
                                            
                             P = 1 −                  = 1                        (12)
                                         (∆ −  )
Summing over all firms, and the outside option yields
                                            X          
                                −1=                            − P0                     (13)
                                           =1
                                                    (∆ −  )


                                                   9
     Observe that the RHS of this expression is increasing in  ,  = 1  . Now, the
existence of an equilibrium guarantees there is at least one solution to the system of equations
(12) above. Suppose there were at least two solutions, denoted with superscripts  and .
W.l.o.g., suppose that    . Then, since the RHS of (13) is monotonic, there must exist
a  such that    . The former relation (   ) implies from (12) that P  P ,                
                                              
             ( )                      ( )                                                              
or          P                         P                . Therefore, this last relation, along with   
    (0 ) + =1 ( )     (0 ) + =1 ( )
                    P         ¡  ¢        P            ¡  ¢
implies that                                  =1       . Using the same argument, the inequality
          
                      =1
                              P           ¡  ¢ P              ¡ ¢
   implies that =1                      =1  , which directly contradicts the earlier
statement. Hence there can only be a single solution to (12).
     The proof above shows a contradiction by assuming there were more than one solution.
A standard uniqueness technique (although with limited applicability) is to show that the
best-replies constitute a contraction mapping. As shown in Appendix 1, this is not the case
here, even though that property holds for the related Logit model (see Anderson, de Palma,
and Thisse, 1992, p. 224).
     The earlier ranking results of Proposition 2 also follow from the same argument as before:
any option with    garners positive sales, otherwise, it has zero sales.
     In the symmetric case, recalling that  =  −  we have:
                                                        µ                ¶
                                                                  
                                        (∆ −  )  1 −                     = 
                                                              0 +  
We know that there is a unique solution to this equation: by Proposition 2, any equilibrium
must be symmetric, and by Proposition 4, there is only one solution.7 It solves:

               ( +  ( − 1))  +1 − ( − 1) ∆  + ( + 1)  0 − ∆0 = 0                          (14)
                                                                              
       As before, the solution from here with 0 = 0 is  =               ∆. The monopolistic
                                                              ( (−1) +)
                                                                  

                                         
competition limit of (14) is: lim  = 1+ ∆, which is independent of the value of 0 . Here
                              →∞
0 disappears because it is effectively dominated by the variety of all the other options. Note
too that the mark-up remains strictly positive in equilibrium, just as for the logit and the
CES.
   The same limit attains when the outside option gets infinitely attractive, i.e., lim  =
                                                                                                   0 →∞
∆
1+
    .  The intuition is that in either case, any individual firm is competing with an aggregate
competition that is infinitely attractive. Indeed, for both the case of monopolistic compe-
tition and an infinitely attractive outside option, the choice probability for Firm  tends to
  (− )
+(− )
           with  very large in both cases. Effectively, the price term in the denominator
is negligible, and we have the monopolistic competition version of the firm’s problem as
max( − )( −  ) , which gives the surplus split result (Proposition 3) as − = .
                                                                                −
 

   7
    The term on the LHS has a power  + 1 and so is the “most convex”: for the other terms on the LHS,
the term in   has a smaller power, so this cuts the other once; the last term is a constant, and the one
before is linear.

                                                      10
4         Introducing income
So far, we have written choice probabilities solely in terms of net “surpluses” (“reservation
prices” minus actual prices). However, to reconcile the model with standard neoclassical
consumer theory, which we should have to do to make any meaningful welfare statements,
means that we need to introduce an explicit role for income in the demand functions. The
simplest way of doing so, and one which we show below is consistent with neoclassical
consumer theory, involves just adding income  to the  terms, so we write

                                        =  +  −  ,       = 1  

Now, redefining ∆ =  +  −  , it is clear that all of the previous analysis up till now goes
through without any changes, as long as the income constraint is not binding. To explicitly
account for this constraint, we therefore write:

                                   = ( +  −  ) I− ≥0 ,     = 1           (15)

where I− ≥0 is an indicator function for whether or not income is breached. Note that 
might be interpreted as total income, or it could (more mildly) be a sub-income devoted to
a sector.8
    The behavior of the demands as a function of income differs as to whether  is positive
or negative. If   0, demands exhibit a jump down to zero at the price  =  at which
income is breached. This is explained further in the next sub-section. On the other hand, if
  0, there is no demand discontinuity, and demand (for product ) tends smoothly to zero
when  approaches  +  . One can interpret  +  as the reservation price in this case, so
that more desirable goods still have larger values of  .
    Generally speaking, we are interested in the behavior of the demands away from the prices
at which income constraints are breached, although the behavior of demands at the critical
values does hold some interest. Indeed, if we are looking at motor-cars, budget constraint
violations seem to be a relevant issue, so that is loosely consistent with the values of   0.
By contrast, for chewing gum, the relevant  might be quite negative, in the simple sense
that the reservation price for any brand is significantly lower than income.
    As an example, suppose an individual is deciding which of several possible motor scooters
to buy. Her monthly RER pass costs 100Euros, and we take that as the value that the motor
scooter has to beat. She retains in her choice set all options costing less that 100Euros, and
then evaluates them according Luce’s choice rule with scale parameters  +  −  . In this
sense, the model above with the indicator function is similar to that of Tversky (1972a,b)
where the first decision is whether or not the item is affordable. Then, also like Tversky’s
model, the decision of which item to buy is probabilistic, though influenced by both the
“quality” scale values, and also prices.
    In the formulation (15) above, as income, , rises, prices are less important determinants
of product choices. This seems a reasonable property of the demand system, corresponding
    8
        In the latter case, it might be nested within a broader model.

                                                        11
loosely (in a neo-classical sense) to decreasing marginal utility of money. However, the scale
values  are also less important in determining choices. Insofar as richer people might tend
to concentrate their purchases on a narrow band of top quality products, the model does not
deliver this intuition. Instead, market shares are monotone in  and the individual tends
to purchase all goods in roughly the same amount as income gets large. The alternative
demand formulation in the next Section delivers instead the property that richer people buy
higher qualities with higher probabilities.

4.1    Inferior and normal goods
The demand system with  = + − allows us to study the effects on demand of changing
incomes. Namely, the demand for any given product can be a non-monotonic function of
income. To see this, as income rises through the product price, demand jumps from a
zero to a positive level. Thenceforth it tends to rise with income, at least until a second,
higher-priced, product becomes viable, and the first one suffers a corresponding jump down.
   This is interesting for a cross-section of individuals of different incomes, and tracking the
purchase patterns of high-quality goods as a function of income: the standard Logit and CES
models have either no income effects (Logit) or unit income elasticity (CES). Indeed, linear
Engel curves are quite standard in consumer models.

4.2    The Representative Consumer
We wish to find the representative consumer preferences consistent with the demand model
where the  are given by (15). Consistency requires that Roy’s Identity holds, and that
the indirect utility function satisfies the appropriate properties. To show the properties are
satisfied, we explicitly introduce (for this Section) the numeraire, good 0, with price 0 .
Consider then the following indirect utility function:
                                            µ             ¶+1
                                      1 X            −           
                         ( ) =              +             + 0                    (16)
                                     + 1 =1         0            0

where  denotes the price vector,  = (0  1   ), and  denotes the income of the repre-
sentative consumer. Then:
                        ( )      Xµ
                                       
                                                      − 
                                                             ¶
                                                                  0
                                  =           +               +  and
                                    =1
                                                        0        0
                                         µ                 ¶
                        ( )                    − 
                                  = −  +                    
                                                   0
   We can then verify that Roy’s identity holds:
                                 ³           ´
                    ()         + −
                                         0              ( )
                     
            = −  () = P ³
                                           ´   =P                       0
                                                                                = P 
                                                   =1 ( ) +
                                       −
                                 + 0  + 0                           0
                               =1


                                               12
Furthermore,  ( ) is clearly homogeneous of degree zero in ( ). It remains to show
that  ( ) is quasi-convex in its arguments. This task is accomplished by showing that
 ( ) has the stronger property of convexity. Indeed, because the sum of convex functions
                                   ³          ´+1
                                         −
is convex, it suffices to show that  + 0           is convex. This is readily done by showing
that the Hessian is positive semi-definite in ( 0   ) for   0.
    We thus have:

Proposition 5 The Lucian demand functions (choice probability system) can be rationalized
by the representative indirect utility function:
                                                µ            ¶+1
                                          1 X           −           
                              ( ) =            +             + 0 
                                         + 1 =1        0            0

        In the symmetric case, we have:
                                                       µ       ¶+1
                                                       −          
                                      (  ) =      +          + 0 
                                                   +1      0          0

This version can be used for surplus analysis of the market outcome.9
    Returning to the asymmetric case, Fajgelbaum, Grossman, and Helpman (2011) have
recently proposed a nested logit model for analyzing trade in differentiated products. That
model, along with the frequently-deployed CES model, implies that demands are never zero
(see Baldwin and Harrigan, 2007, for a strong critique of this property). It would be in-
teresting to see whether the model proposed in this paper, extended perhaps to a nested
version, could provide better empirical verification.


5         A Gabszewicz-Thisse Formulation
Here we provide a parallel analysis using a variant of the Lucian surplus inspired from
the seminal Gabszewicz and Thisse (1980) formulation of vertical product differentiation.
Specifically, instead of (2) consider now the following specification for the scale values:

                                            = max ( ( −  )  0) 

with the choice probabilities again given by (3), i.e.,

                                             ( ( −  ))
                                 P = P                          
                                                                         = 1  
                                           =1 ( ( −  ))

    9
        See also de Palma and Kilani (2011) for ways to compute income effects numerically.




                                                         13
where the sum is again over all active firms. Limit properties are similar to those given
previously, but now note that
                                                      ( )
                                     lim P = P                       
                                     →∞
                                                   =1   ( )
so that higher quality products have higher choice probabilities even for arbitrarily large
income levels.
   Corresponding to Lemma (1), the key derivative properties are now:

                         P       P (1 − P )
                             = −                  0,                 = 1             (17)
                                      
and
                           P       P P
                               =             0,        = 1    6=                 (18)
                                    
Note that the only explicit difference with the previous case is the introduction of the qual-
ities.
    The first-order derivative for pricing is now (for   0):
                                               µ                            ¶
             (  − )    P (1 − P )         
                            =                                   − ( −  )      = 1   (19)
                                             (1 − P ) 
and hence there is a unique best-response price for :
                                   
                                             = ( −  ) ,  = 1                        (20)
                               (1 − P ) 
A similar proof to the earlier one shows that there is a single solution to this system of
equations, which therefore characterizes the unique Nash equilibrium.
    We now turn to the surplus split, where now the Lucian surplus associated to Firm  is
 ( −  ),  = 1  . Let now ∆ ≡ ( −  ) be the income-cost associated to Firm . The
pricing expression (20) can be rewritten as before as:
                                                   ∆
                                  −  =                    ∆ ,                            (21)
                                              (1 − P ) + 1
which gives  =  for the monopoly case (P = 1) independent of quality. Otherwise,
     (so that   0).
    We can also write the equilibrium Lucian surplus as:
                                       ½                 ¾
                                          (1 − P ) ∆
                                   =                                      (22)
                                          (1 − P ) + 1
                                                                                             ∆
   Hence, in equilibrium the income-cost differential, ∆ , is split between a mark-up     (1−P )+1
                   (1−P )∆
for each firm, and (1−P)+1 for the consumer, so the share of this differential going to the
consumer is  (1 − P ), which is higher for firms with lower equilibrium outputs.

                                                 14
     In a symmetric equilibrium we have
                                              −
                               −  =    ¡   1
                                                  ¢    − 
                                            1−  +1
or                                               ¡     ¢
                                                     1
                                                 1−  ∆
                                    −    = ¡   1
                                                      ¢  
                                                1−  +1
which clearly shows the split of income-cost going to firms decreasing as  rises.
   An outside option with fixed “scale” level 0 ≥ 0 is readily introduced, with the choice
probabilities becoming
                                   ( ( −  ))
                      P =         P                            = 1  
                           (0 ) + =1 ( ( −  ))
Existence and uniqueness of the equilibrium are shown in Appendix 2.

Proposition 6 There exists a unique price equilibrium for the Luce-Gabszewicz-Thisse model
with an outside option.

     In the symmetric case, recalling that  =  ( − ) we have:
                                          µ               ¶
                                                   
                              (∆ −  )  1 −              = 
                                               0 +  
There is a unique solution to this equation: by Proposition 6, there is only one equilibrium,
and it solves:

             ( +  ( − 1))  +1 − ( − 1) ∆  + ( + 1)  0 − ∆0 = 0       (23)
which is almost the same as (14), and the corresponding properties are the same, mutatis
mutandis.
   When there is no outside good, the equilibrium price is a convex combination of cost, ,
and income, :
                                                ³       ´
                                                     
                                                 (−1)              
            =−³              ´ ( − ) =  ³            ´ + ³         ´  
                                                                 
                      (−1)
                            +                 (−1)
                                                       +        (−1)
                                                                       +
                                            |       {z    }    |     {z   }
                                                      1                1

Inspection yields the limit properties lim→∞  = ; lim→0  = ; lim→1  = , and
                  
lim→∞  = 1+ + 1+ .
   The representative consumer preferences consistent with this demand model are (cf.
(16)):
                                      1 X 
                                           
                          ( ) =            ( −  )+1 + 0             (24)
                                     + 1 =1

                                                15
which is again a simple sum of quadratics for the central case of  = 1. Roy’s identity is
readily verified:
                        ()
                                          ( −  )
                                                                     ( )
               =   −  ()   =                          =P                          = P 
                                      P 
                                       
                                                                           ( ) + 0
                                       ( −  ) + 0       =1
                                      =1

In the symmetric case the representative consumer’s preferences boil down to:
                                           
                         (  ) =       ( +  − )+1 + 0 
                                         +1
This simple form enables welfare analysis is a straightforward manner.


6    Conclusions
This paper has considered variants on the Independence of Irrelevant Alternatives (IIA)
property of the Luce model, of which the CES and Logit are the mainstream examples (see
also Laurent, 2007a, 2007b). Further work can develop the theme of this paper by considering
other Lucian specifications of the form P = P  ( ) ( ) , with  () an increasing function.
                                               =1 
    The paper brings several of Jacques Thisse’s many contributions to economic theory,
namely his work on the probabilistic discrete choice model of Luce, oligopoly pricing of
differentiated products, formulating representative consumer counterparts to discrete choice
models, and vertical product differentiation. Useful extensions of the current paper could
follow some of Jacques’ other research themes, namely introducing localized competition
elements and allowing for spatial differentiation of consumers.


7    Appendix 1: Contraction Mapping Approach
                                                             
The reaction function is given in implicit form by (8) as (1−P ) − ( −  ) = 0, which can
              
rewrite as (1−P ) +  − ∆ = 0. Recalling Lemma 4 (here rewritten in terms of  ’s), the
own-price demand derivative is (for all   0 and   0):
                         P     P (1 − P )
                             =                0,         = 1  
                                   
while the cross-price demand derivative is:
                       P      P P
                           = −        0,        = 1    6= 
                              
The slope of the best-reply function for Firm ’s choice of  with respect to  is given by
the Implicit Function Theorem as
                                                       ³        ´
                            ¯                            P P
                                                        −  
                            ¯
                         ¯                (1−P )2

                            ¯ =− 1                          ³            ´
                                       + 1 + (1−P )2  P (1−P )
                                                        
                                    (1−P )                     


                                                       16
which simplifies to

                                ¯                         P P
                             ¯
                                ¯                   (1−P )2 
                                     =                                             0
                             ¯             1
                                              (1−P )
                                                       +1      +          P
                                                                        (1−P )
                                                 P P
                                              (1−P ) 
                                      =         1        
                                                
                                                   +1
                                             P            ¯
                                                      ¯
Hence the desired contraction property,         6=  ¯           1, requires proving that
                                                              

                                      P X P       1
                                                      + 1
                                   (1 − P ) 6=   

Note that:
                                                                (                          )
                         P X P             P              X P              P
                                          =                                      −
                      (1 − P ) 6=      (1 − P )                    
                                                                                 
                                                    P        X P                     P2
                                                                                          
                                          =                                       −             
                                                 (1 − P )          
                                                                                    (1 − P )

Note that if  = 1, this reads
                             P X P                  P2 
                                                                  P2
                                                                    
                                                =              −
                          (1 − P ) 6=            (1 − P ) (1 − P )
                                                         P2
                                                          
                                                =               ( − 1) 
                                                      (1 − P )
                                                                                                    P2
which needs to be less than 2. However, P be anywhere between 0 and 1, so (1−P ) can be
                                                                                 


anywhere between 0 and infinity ; so therefore the best reply is not a contraction mapping.


8    Appendix 2: Proof of Proposition 6
Existence follows from similar arguments as used for the other case. Uniqueness is given now
by showing that there is a single solution to the equations defining the (interior) first-order
conditions. Recall that he first-order conditions solve (22), which can be rewritten as
                                                  
                              (1 − P ) =               ,  = 1 
                                               ∆ − 
or
                                             
                           P = 1 −                                     = 1                     (25)
                                       ( ∆ −  )

                                                    17
Note that P is decreasing in  on the reaction function (25). Summing over all firms, and
the outside option yields
                                       X             
                              −1=                            − P0                    (26)
                                      =1
                                               ( ∆ −  )
    Observe that the RHS of this expression is increasing in  ,  = 1  . Now, the
existence of an equilibrium guarantees there is at least one solution to the system of equations
(25) above. Suppose there were at least two solutions, denoted with superscripts  and .
W.l.o.g., suppose that    for a given . Then, since the RHS of (26) is monotonic,
there must exist a  such that    . The former relation (   ) implies from (25)
                                                        
                         (  )                    ( )
that P  P , or ( ) +P 
                                              P              . Therefore, this last relation, along
                     0    =1 ( )    (0 ) + =1 ( )
                                                               
                                P         ¡ ¢ P                 ¡ ¢
with    implies that =1   =1  . Using the same argument, the
                                        P                    P        ¡ ¢
inequality    implies that =1 ( )  =1  , which directly contradict
the earlier statement. Hence there can only be a single solution to (25).


References
 [1] Anderson, Simon P. and André de Palma, 2001. Product Diversity in Asymmetric
     Oligopoly: Is the Quality of Consumer Goods Too Low? Journal of Industrial Eco-
     nomics, 49, 113-135.

 [2] Anderson, Simon P., André de Palma and Jacques-François Thisse, 1988a. The CES
     and the Logit : Two Related Models of Heterogeneity. Regional Science and Urban
     Economics, 18, 155-164.

 [3] Anderson, Simon P., André de Palma, and Jacques-François Thisse, 1988b. A Repre-
     sentative Consumer Theory of the Logit Model. International Economic Review, 29,
     461-466.

 [4] Anderson, Simon P., André de Palma and Jacques-François Thisse, 1992. Discrete choice
     of product differentiation. Cambridge, MA: The MIT Press.

 [5] Baldwin, Richard, and James Harrigan, 2007. Zeros, Quality and Space: Trade The-
     ory and Trade Evidence. NBER Working Papers 13214, National Bureau of Economic
     Review, and American Economic Review, forthcoming.

 [6] Debreu, Gerard, 1960. Review of R. D. Luce, Individual choice behavior: A theoretical
     analysis. American Economic Review, 50, 186-188.

 [7] de Palma, André, and Karim Kilani, 2011. Transition choice probabilities and welfare
     analysis in random utility models. Economic Theory, 46, 427-464.



                                                   18
 [8] Fajgelbaum, Pablo D., Gene M. Grossman, and Elhanan Helpman, 2011, Income Dis-
     tribution, Product Quality, and International Trade, NBER Working Paper 15329.

 [9] Gabszewicz, Jean J. and Jacques-François Thisse, 1980. Entry (and exit) in a differen-
     tiated industry. Journal of Economic Theory, 22(2), 327-338.

[10] Laurent, Reynald, 2007a. Le Modèle "Elimination Par Attributs" de Tversky : une
     application à la différenciation des produits, Revue Economique, 58-3, 545-554.

[11] Laurent, Reynald, 2007b. Choix probabiliste et différenciation par attributs. Ph.D. the-
     sis, Sciences Economiques de l’Université Paris I.

[12] Luce, R. Duncan, 1959. Individual Choice Behavior: A Theoretical Analysis. New York:
     Wiley.

[13] Luce, R. Duncan, 1977. The choice axiom after twenty years. Journal of Mathematical
     Psychology 15 (3): 215—233.

[14] Tversky, Amos 1972a. Choice by elimination. Journal of Mathematical Psychology,
     9:341-367.

[15] Tversky, Amos 1972b. Elimination by aspects: A theory of choice. Psychological Review,
     79:281—299.




                                            19

				
DOCUMENT INFO