# Capacity and Price Competition in Markets with Congestion Effects

We study oligopolistic competition in service markets where firms offer a service to customers. The service quality of a firm (from a customer's perspective) depends on the level of congestion and the charged price. A firm can set a price for the service offered and additionally decides on the service capacity to mitigate congestion. The total profit of a firm is derived from the gained revenue minus the capacity investment cost. Firms simultaneously set capacities and prices to maximize their profit and customers subsequently choose the services with lowest combined cost (congestion and price). For this model, Johari, Weintraub and Van Roy derived the first existence and uniqueness results of pure Nash equilibria (PNE) assuming mild conditions on congestion functions. Their existence proof relies on Kakutani's fixed-point theorem and a key assumption for the theorem to work is that demand for service is elastic, i.e., there is a smooth inverse demand function determining the volume of customers given the effective customers' costs. In this paper, we consider the case of inelastic demand. This scenario applies to realistic cases where customers are not willing to drop out of the market, e.g., if prices are regulated by reasonable price caps. We investigate existence, uniqueness and quality of PNE for models with inelastic demand and price caps. We show that for linear congestion cost functions, there exists a PNE. This result requires a completely new proof approach compared to previous approaches, since best responses may not exist and thus standard fixed-point arguments are not directly applicable. We show that the game is C-secure (see Reny, and McLennan, Monteiro and Tourky), which leads to the existence of PNE. We furthermore show that the PNE is unique and that the efficiency compared to a social optimum is unbounded in general.

## Authors

• 9 publications
• 1 publication
• ### Efficiency and complexity of price competition among single-product vendors

Motivated by recent progress on pricing in the AI literature, we study m...
02/13/2015 ∙ by Ioannis Caragiannis, et al. ∙ 0

• ### Toll Caps in Privatized Road Networks

We study a nonatomic routing game on a parallel link network in which li...
02/28/2018 ∙ by Tobias Harks, et al. ∙ 0

• ### A survey of queueing systems with strategic timing of arrivals

Consider a population of customers each of which needs to decide indepen...
06/22/2020 ∙ by Moshe Haviv, et al. ∙ 0

• ### On Optimal Pricing of Services in On-demand Platforms

I consider the optimal hourly (or per-unit-time in general) pricing prob...
03/19/2018 ∙ by Vijay Kamble, et al. ∙ 0

• ### Greater search cost reduces prices

The optimal price of each firm falls in the search cost of consumers, in...
04/02/2020 ∙ by Sander Heinsalu, et al. ∙ 0

• ### Customer Price Sensitivities in Competitive Automobile Insurance Markets

Insurers are increasingly adopting more demand-based strategies to incor...
01/21/2021 ∙ by Robert Matthijs Verschuren, et al. ∙ 5

• ### Bargaining for Revenue Shares on Tree Trading Networks

We study trade networks with a tree structure, where a seller with a sin...
04/22/2013 ∙ by Arpita Ghosh, et al. ∙ 0

##### This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

## 1 Introduction

We consider a model for oligopolistic capacity and price competition in service markets exhibiting congestion effects. There is a set of firms offering a service to customers and the overall customer utility depends on the congestion level experienced and the price charged. In addition to setting a price, a firm can invest in installing capacity to reduce congestion. Firms compete against each other in the market by simultaneously choosing capacity and price levels while customers react subsequently choosing the most attractive firms (in terms of congestion- and price levels). This model captures many aspects of realistic oligopolistic markets such as road pricing in traffic networks (see Xiao, Yang and Han [30] and Zhang and Han [32]) and competition among WIFI-providers or cloud computing platforms (see Acemoglu and Ozdaglar [2] and Anselmi et al. [4]).

In a landmark paper, Johari, Weintraub and Van Roy [21] (JWVR for short) studied the fundamental question of existence, uniqueness and worst-case quality of pure Nash equilibria for this model. They derived several existence and uniqueness results under fairly general assumptions on the functional form of the congestion functions. Specifically, they showed that for models with elastic demand, a unique pure Nash equilibrium exists. The existence result is based on the Kakutani fixed-point theorem (see [22] and the generalizations of Fan and Glicksberg [17, 19]) and crucially exploits the assumption that demand is elastic leading to non-emptyness of the best-response correspondences of firms. For the case of inelastic demand

, that is, there is a fixed volume of customers requesting service, much less is known in terms of existence and uniqueness of equilibria. Many works in the transportation science and algorithmic game theory community (see, e.g.,

[7, 12, 13, 14, 31, 18] and [28, 8, 3, 16], respectively) assume inelastic demand and this case is usually considered as fundamental base case. As we show in this paper, in terms of equilibrium existence, the case of inelastic demand is much more complicated compared to the seemingly more general case of elastic demands: the best-response correspondence may be empty in the “inelastic case” putting standard approaches out of reach. Besides this theoretical aspect, the case of inelastic demand is also interesting from a practical point of view as it applies to realistic situations (such as road pricing games), where customers have already made investments on a longer time-scale (like buying a car) so that they are not willing to opt out of the game.

### 1.1 Related Work

JWVR [21] derived several existence and uniqueness results for pure Nash equilibria assuming elastic demand of customers. The elasticity is modeled by a differentiable strictly decreasing inverse demand function. Depending on the generality of the allowed congestion cost functions, further concavity assumptions on the inverse demand function are imposed by JWVR. The existence proof is based on Kakutani’s fixed-point theorem which crucially exploits (together with further convexity and compactness properties) that best-response correspondences of the firms are non-empty.

For the model with inelastic demand – that we consider in this paper – JWVR derive an existence result assuming homogeneous firms, that is, the cost function of every firm is equal. As shown by JWVR, homogeneity implies that there is only one symmetric equilibrium candidate profile. For this specific symmetric strategy profile, they directly prove stability using concavity arguments of the best-reply correspondences. This proof technique is clearly not applicable in the general non-homogeneous case. As already mentioned, a further difficulty stems from the fact that one cannot directly apply standard fixed-point arguments as even a best response of a firm might not exist.

Acemoglu, Bimpikis and Ozdaglar [1] study a capacity and price competition game assuming that the capacities represent “hard” capacities bounding the admissible customer volume for a firm. They observe that pure Nash equilibria do not exist and subsequently study a two-stage model, where in the first stage firms determine capacities and in a second stage they set prices. For this model, they investigate the existence and worst-case efficiency of equilibria (see also Kreps and Scheinkman [23] for earlier work on the two-stage model). Further models in which capacities are determined centrally in order to reduce the travel cost of the resulting Wardrop equilibria have been considered in the (bilevel) optimization community for some time already, see [11, 24]. In a very recent work [29], Schmand, Skopalik and Schröder study existence and inefficiency of equilibria in a network investment game in which the firms invest in edges of a series-parallel graph (but do not directly set prices).

### 1.2 Our Results and Proof Techniques

We study in this paper oligopolistic capacity and price competition assuming that there is a fixed population of customers requesting service. This assumption applies to road pricing games, where customers usually own a car and are not willing to opt out of the game given that prices range in realistically bounded domains. Consequently, our model allows for the possibility of a priori upper bounds with respect to the prices set by the firms. This situation appears naturally, if there are legislative regulations imposing a hard price cap in the market (see Correa et al. [9] and Harks et al. [20] and further references therein). As reported in Correa et al. [9], even different price caps for different firms is current practice in the highway market of Santiago de Chile, where 12 different operators set tolls on different urban highways each with a unique price cap. While firm-specific price caps generalize the model of JWVR, we impose, on the other hand, more restrictive assumptions on the congestion functions, that is, they are assumed to be linear with respect to the volume of customers and inverse linear with respect to the installed capacity.

As our first result, we completely characterize the structure of best-response correspondences of firms; including the possibility of non-existence of best-replies (Theorem 3.6). Our second main result then establishes the existence of equilibria (Theorem 4.6). For the proof we use the concept of -security introduced by McLennan, Monteiro and Tourky [25], which in turn resembles ideas of Reny [27]. A strategy profile is called -secure, if each player has a pure strategy guaranteeing a certain utility value, even if the other players play some perturbed strategy within a (small enough) neighbourhood. Furthermore, for each slightly perturbed strategy profile, there is a player whose perturbed strategy can in some sense be strictly separated from her securing strategies. Intuitively, the concept of securing strategies means that those strategies are robust to other players’ small deviations. The main result of McLennan, Monteiro and Tourky [25] states that a game admits an equilibrium, if every non-equilibrium profile is -secure. It is important to note that the concept of -security does not rely on quasi-concavity of utility functions nor on their continuity. With our characterization of best-response correspondences at hand, we show that the capacity and price game with inelastic demand, linear congestion functions and price caps is -secure and, thus, admits pure Nash equilibria. As our third main result, we show that there is a unique equilibrium (Theorem 5.7). The proof technique follows the general approach as in JWVR [21], however, since our model allows for price caps, some new ideas are required. We finally study the worst case efficiency of the unique equilibrium compared to a natural benchmark, in which we relax the equilibrium conditions of the firms, but not the equilibrium conditions of the customers. We show that the unique equilibrium might be arbitrarily inefficient (Theorem 6.1).

## 2 Model

There is a set , , of firms offering a service to customers. Customers are represented by the continuum (each consumer is assumed to be infinitesimally small and represented by a number in ) and we denote by the standard simplex of assignments of customers to firms.111All results hold for arbitrary intervals by a standard scaling argument. The effective quality of the service of firm offered to a customer depends on two key factors: the level of congestion and the price charged by the firm. The congestion function depends on the volume of customers and the service capacity installed. Clearly grows with the volume of customers  to be served but decreases with the service capacity . If no capacity is installed, i.e. , we assume infinite congestion, and for the case that , we assume that congestion depends linearly on the volume of customers and inverse-linearly on the installed capacity, that is,

 ℓi(xi,zi)={aixizi+bi, for % zi>0,∞, for zi=0,

where and are given parameters for . Note that the case can be interpreted as if firm is just opting out of the market and does not offer the service at all. Each firm  additionally decides on a price which is charged for offering service to its customers, where

is a given price cap. For a capacity vector

with , i.e. the service is offered by some firms, and a price vector , customers choose rationally the most attractive service in terms of the effective costs, that is, congestion and price experienced. This is expressed by the Wardrop equilibrium conditions:

 ci(x,z,p):=ℓi(xi,zi)+pi≤cj(x,z,p):=ℓj(xj,zj)+pj

holds for all with . Note that for given capacities and prices , there is exactly one satisfying the Wardrop equilibrium conditions (see, e.g., [10]). Call this flow the Wardrop flow induced by . In particular, there exists a constant such that holds for each with , and holds for each with . For a Wardrop flow , we call the corresponding constant the (routing) cost of .

If we denote the per-unit cost of installed capacity by , the profit function of a firm can now be represented as

 Πi(z,p)={pixi(z,p)−γizi, for ∑i∈Nzi>0,0, else.

We assume that each firm seeks to maximize her own profit. For each firm , let be the strategy set. A vector consisting of strategies is called a strategy profile, and denotes the set of strategy profiles. We will usually write a strategy profile in the form , where denotes the vector consisting of all capacities for , and is the vector of prices for . The profit of firm for a strategy profile is then defined as . Furthermore, we write for the Wardrop-flow induced by and for the routing cost of . For firm , denote by the vector consisting of strategies . We then write for the strategy profile where firm chooses , and the other firms choose . Moreover, we use the simplified notation and . A strategy profile is a pure Nash equilibrium (PNE) of the thus defined capacity and price competition game, if for each firm :

 Πi(si,s−i)≥Πi(s′i,s−i) for all s′i∈Si.

For given strategies of the other firms, the best-response correspondence of firm is defined as

 BRi(s−i):=argmax{Πi(si,s−i)|si∈Si}.

If is clear from the context, we just write .

We conclude this section with the following fundamental result about the continuity of the profit functions. We will use Theorem 2.1, which completely characterizes the strategy profiles having the property that all profit functions , , are continuous in , several times during the rest of the paper.

###### Theorem 2.1.

Let . Then: All profit functions , , are continuous in if and only if or .

###### Proof.

We start with the strategy profile and show that for each firm , her profit function is continuous in . Let . For , define and let with . If , then . Otherwise,

 |Πi(s′)−Πi(s)|=|xi(s′)p′i−γiz′i|≤xi(s′)p′i+γiz′i≤p′i+γiz′i<δ(1+γi)≤ε

holds, showing that is continuous in .

Now consider with . We again need to show that all profit functions , , are continuous in . Since , we get that . Furthermore, for sufficiently small, holds for all with . Write , where denotes the strategies of the firms in , and denotes the strategies of the firms in .

Now let . We need to show that is continuous in . For all with , firm ’s profit is . Thus it is sufficient to show that is continuous in . The idea of the proof is to show that, for a slightly perturbed strategy profile , the difference between and , as well as the difference between and , is small.

In the following, let with . It is well known ([5]) that is the unique optimal solution of the following optimization problem :

 (Q)min∑j∈N∫xj0(ℓj(t,z′j)+p′j)dt s.t.∑j∈Nxj=1, xj≥0 ∀j∈N.

Furthermore, for . Therefore, the values are the unique optimal solution of

 min∑j∈N+(s′)∫xj0(ajz′jt+bj+p′j)dt s.t.∑j∈N+(s′)xj=1, xj≥0 ∀j∈N+(s′),

which is equivalent to

 max−∑j∈N+(s′)(aj2z′jx2j+(bj+p′j)xj) s.t.∑j∈N+(s′)xj=1, xj≥0 ∀j∈N+(s′).

By Berge’s theorem of the maximum [6], for all there is such that for all with . That is, is continuous in if we only allow changes in , but not in . Furthermore, if denotes the optimal objective function value of , and if only changes in are allowed, also is continuous in , i.e. for all there is such that for all with .

We now distinguish between the two cases that or .

First consider the case , that is, , and let . Note that , thus we need to find such that for all with . To this end, define

 δ=δ(i,ε):=⎧⎨⎩δ3(1), if q(s)+1≤biε,min{δ3(1),aiε22(q(s)+1−biε)}, else,

and let with . In particular, . Furthermore, holds since . If , we immediately get . Thus assume and assume, by contradiction, that . Then, by definition of , we get the following contradiction:

 q(s)+1≥q(s′)≥ai2z′ixi(s′)2+(bi+p′i)xi(s′)≥ai2z′iε2+biε>ai2δε2+biε≥q(s)+1.

Therefore, holds, showing that is continuous in if .

Now consider the case , i.e. . For , we need to find such that for all with . To this end, define

 δ:=min{min{δ(j,ε2n):j∉N+},δ2(ε2)}

and let with . In particular, , thus . Furthermore, since , we get for all . If for all , we get and thus , as desired. Otherwise, there is with . In particular, . We now use a result about the sensitivity of Wardrop equilibria due to Englert et al. [15, Theorem 2]. They show that if firm is deleted from the game, the resulting change in the Wardrop flow can be bounded by the flow that received. More formally, if is the Wardrop flow for the game with firms , and is the Wardrop flow if firm is deleted from the game, then for all . Obviously, changing firm ’s capacity from to has the same effect on the Wardrop flow as deleting firm . Therefore, if we change, one after another, the capacities of all firms having to , we get (note that the flow values for can always be always upper-bounded by ):

 |xi(s′)−xi(s′1,s2)|≤(n−1)ε2n<ε2.

Using this, we now get the desired inequality:

 |xi(s)−xi(s′)|≤|xi(s′)−xi(s′1,s2)|+|xi(s)−xi(s′1,s2)|<ε2+ε2=ε.

Altogether we showed that all profit functions are continuous at if .

To complete the proof, it remains to show that if all profit functions , , are continuous in , then or holds. We show this by contraposititon, thus assume that fulfills and . We need to show that there is a firm such that is not continuous in . To this end, let with . Define the sequence of strategy profiles  by for all , and . Obviously, for . But for the profits, we get

 Πi(sn)=xi(sn)pni−γizni=pi−γin→n→∞pi>0.

Since , this shows that is not continuous in . ∎

## 3 Characterization of Best Responses

The aim of this section is to derive a complete characterization of the best-response correspondences of the firms. We will make use of this characterization in all our main results, i.e. existence, uniqueness and quality of PNE.

Given a firm and fixed strategies for the other firms, we characterize the set of best responses of firm to . To this end, we distinguish between the two cases that (Subsection 3.1) and (Subsection 3.2). Subsection 3.3 then contains the derived complete characterization. In Subsection 3.4, we discuss how our results about the best responses influence the applicability of Kakutani’s fixed point theorem.

### 3.1 The Case z−i=0

In this subsection, assume that the strategies of the other firms fulfill . Under this assumption, firm does not have a best response to :

.

###### Proof.

Whenever firm chooses a strategy with , then holds for the induced Wardrop-flow , thus firm ’s profit is . On the other hand, any strategy with yields a profit of 0. Thus, firm ’s profit depends solely on her own strategy , and can be stated as follows:

 Πi(zi,pi):={pi−γizi, for zi>0% ,0, for zi=0.

Obviously, holds for each , i.e. for and . On the other hand, by for , firm gets a profit of arbitrarily near to , that is, . This shows . ∎

### 3.2 The Case z−i≠0

In this subsection, assume that the strategies of the other firms fulfill . For a strategy of firm , write for firm ’s profit function, for the Wardrop-flow induced by and for the corresponding routing cost.

For , firm ’s profit is . It is clear that each strategy with yields , and thus . On the other hand, each strategy with yields negative profit since

 Πi(zi,pi)=xi(zi,pi)pi−γizi<1⋅Ci−γi⋅Ci/γi=0.

Therefore, each best response fulfills since it yields nonnegative profit. Thus, can be described as the set of optimal solutions of the following optimization problem :

 \hypertargetPi(Pi)max Πi(zi,pi)  % subject to zi∈[0,Ci/γi], pi∈[0,Ci].

has an optimal solution, since the feasible set of is compact and is continuous in (see Theorem 2.1), and thus the theorem of Weierstrass can be applied. Since can be described as the set of optimal solutions of , we get .
Note that is a bilevel optimization problem (since can be described as the optimal solution of a minimization problem ([5])), and these problems are known to be notoriously hard to solve. The characterization of that we derive here has the advantage that it only uses ordinary optimization problems, namely the following two (1-dimensional) optimization problems in the variable ,

 max f1i(K):=¯¯¯xi(K)⋅(K−bi−2√aiγi) s.t. 2√aiγi+bi≤K K≤√aiγi+bi+Ci ¯¯¯xi(K)>0,
 max f2i(K):=¯¯¯xi(K)⋅(Ci−aiγiK−bi−Ci) s.t. √aiγi+bi+Ci0,

where

 ¯¯¯xi(K):=1−∑j∈N(K)(K−bj−pj)zjaj

with and .
Note that is a continuous function which is equal to 1 for , and strictly decreasing for with . Therefore, there is a unique constant with the property . Obviously, if and only if . Furthermore, the function is closely related to Wardrop-flows, as described in the following lemma:

###### Lemma 3.2.
1. [1.]

2. If with , then for .

3. If with , and fulfills and , then and .

###### Proof.

We start with 1, so let such that . By definition of , we get for all with , and for all with . Since

 ℓj(xj(zi,pi),zj)+pj={ajzjxj(zi,pi)+bj+pj, for j∈N+,∞, for j∈N with zj=0,

we get that . Therefore, for each , we get , which is equivalent to . Using yields

 xi(zi,pi)=1−∑j∈N∖{i}xj(zi,pi)=1−∑j∈N(K)(K−bj−pj)zjaj=¯¯¯xi(K),

as desired. Now we show 2, so let with and let be a strategy with and . Consider the vector defined by

 xj:=⎧⎪ ⎪⎨⎪ ⎪⎩¯¯¯xi(K),j=i,(K−bj−pj)zjaj,j∈N+ with bj+pj

It is clear that holds for all with , and . Furthermore, the definition of yields . We now show that fulfills the Wardrop equilibrium conditions. The uniqueness of the Wardrop-flow then implies ; and