Bargains Followed by Bargains: When Switching Costs Make Markets More Competitive∗† Jason Pearcy‡ September 13, 2015 Abstract In markets where consumers have switching costs and firms cannot price discriminate, firms have two conflicting strategies. A firm can either offer a low price to attract new consumers and build future market share or a firm can offer a high price to exploit the partial lock-in of their existing consumers. This paper develops a theory of competition when overlapping generations of consumers have switching costs and firms produce differentiated products. Competition takes place over an infinite horizon with any number of firms. This paper shows that the relationship between the level of switching costs, firms’ discount rate, and the number of firms determines whether firms offer low or high prices. Similar to previous duopoly studies, switching costs are likely to facilitate lower (higher) equilibrium prices when switching costs are small (large) or when a firm’s discount rate is large (small). Unlike previous studies this paper demonstrates that the number of firms also determines whether switching costs are pro- or anti-competitive, and with a sufficiently large (small) number of firms switching costs are pro- (anti-) competitive. JEL Classification: D2, D4, L1. Keywords: Switching Costs, Product Differentiation, Dynamic Competition, Discrete Choice ∗ I thank Daniel Spulber (the Editor), a Co-Editor, two reviewers, Stefano Barbieri, Michael Baye, Keith Finlay, Eric Rasmusen, Jay Shimshack, seminar participants at Tulane University, Montana State University, Purdue University, Quinnipiac University, Indiana University (BEPP), IUPUI, the Spring 2009 Midwest Economic Theory Conference and the 2010 IIOC for helpful comments and suggestions. All remaining errors are my own. Financial support provided by the Committee on Research Summer Fellowship and the Research Enhancement Fund at Tulane University is greatly appreciated. † c Author Posting. Jason Pearcy 2015. This is the authors’ version of the work. It is posted here for personal use, not for redistribution. The definitive version is forthcoming in the Journal of Economics and Management Strategy. ‡ Montana State University; [email protected] 1 1 Introduction When consumers have switching costs, a consumer’s previous purchase partially locks them in to the same future purchase. Firms have two different strategies when their previous customers are partially locked-in. A firm can offer a high price to exploit the partial lock-in of their existing consumer base, but a high price will lead to more of its customers switching and a decrease in future market share. Alternatively a firm can offer a low price to attract new consumers and invest in market share with the hope of offering a high price in the future to more partially locked-in customers. This paper presents conditions which determine whether equilibrium prices will be higher or lower with switching costs and shows that the number of firms is an important factor when considering how switching costs affect competition. The previous literature has approached a firm’s use of pricing strategies with switching costs in different ways.1 Competition usually takes place over two periods or infinitely many periods. Two-period models dampen the exploitation/investment trade-off in any one period by allowing for some inter-temporal substitution between the two different pricing strategies.2 This type of behavior leads to the familiar result of bargains-then-ripoffs (Klemperer, 1995).3 If competition takes place over infinitely many periods instead, firms must balance both investment and exploitation strategies at once in each period. Other approaches, such as dynamic pricing or subscription models, assume that firms can price discriminate to offer a price for its previous customers and a different price for new customers.4 The typical result is a scenario where bargains are followed by ripoffs; where new consumers without switching costs are offered a low price and old customers with switching costs are offered a high price (Farrell & Klemperer, 2007).5 Instead, if firms are unable to price discriminate, because they cannot identify consumers by their past purchases, firms offer only one price to all consumers. In this case, the exploitation and investment strategies conflict with one another and create an explicit trade-off within each period. 1 See Villas-Boas (2015), Farrell and Klemperer (2007) and Klemperer (1995) for surveys of the switching cost literature. 2 The end-game effects of two-period models are discussed in more detail by Cabral (2014) and Rhodes (2014). 3 Firms may have an incentive to lower prices in the second period if new consumers enter the market, but the exploitation strategy still dominates. 4 See Armstrong (2006) and Chen (2005) for a survey of the dynamic pricing literature. 5 Note that this result is the opposite for dynamic pricing models without switching costs, see Chen and Pearcy (2010). 2 This paper develops a switching cost model that exemplifies the trade-offs between the two strategies since firms set one price for all consumers and competition takes place over an infinite number of periods. The main purpose of this paper is to demonstrate under what conditions switching costs lead to lower or higher prices in any steady-state symmetric equilibrium, and under what conditions an increase in switching costs decreases the price. Both of these results depend on the number of firms, the level of consumer switching costs, and firms’ discount factor. One of the contributions of this paper is to show that switchings costs may be pro- or anti-competitive depending on the number of firms in the market, and allowing for any number of firms in the model provides the necessary insight into the main result. The intuition of the main results and the reasoning why both results depend on the number of firms in the market are explained by the following points. For any symmetric equilibrium, the market share of each firm is the inverse of the number of firms. Under the duopoly case, each firm commands half of the market and has a large consumer base partially locked-in to their firm. If instead there are five firms in the market, each firm has only one fifth of consumers with switching costs partially locked-in to their firm. The returns to exploitation are lower when a firm is only able to exploit one fifth of the consumer base versus one half. Therefore it is more likely that the investment effect dominates with many firms and the exploitation effect dominates with a duopoly.6 Recent empirical evidence is consistent with this intuition as well. Viard (2007) examines a duopoly market and finds that a decrease in switching costs leads to a decrease in equilibrium prices. Dubé, Hitsch, and Rossi (2009) examine two markets, one with six products and one with four, and find that as switching costs increase equilibrium prices decrease. The results of Dubé et al. also indicate that the number of firms and products needed to lower the price in a market with switching costs may not be unreasonably large. Existing theoretical studies of dynamic models with switching costs focus on various duopoly outcomes. Earlier studies by Beggs and Klemperer (1992), Farrell and Shapiro (1988), and Padilla (1995) find that switching costs lead to higher prices, while von Weizsäcker shows that prices may be lower with switching costs. More recent dynamic duopoly studies including Arie and Grieco (2014), Cabral (2014), Doganoglu (2010), Dubé et al. (2009), Fabra and Garcı́a (2015), Rhodes (2014), Shin 6 With many firms, prices with switching costs may be lower than prices when all consumers have no switching costs. This result indicates that the investment effect dominates the exploitation effect even in the presence of more intense competition. 3 and Sudhir (2008), and Villas-Boas (2015) have shown that switching costs may be either pro- or anti-competitive depending on the nature of the competition. The growing consensus is that when the level of switching costs is sufficiently low, or when a firm’s discount rate is sufficiently large, switching costs are pro-competitive. Otherwise switching costs are anti-competitive. The findings in this paper support this conclusion and also show that the nature of switching costs depends on another dimension: the number of firms. Other recent papers with switching costs allow for any number of firms. Wilson (2012) considers a model with search and switching costs with any number of firms where competition takes place only over one period. While interesting, the analysis of the static one shot game lacks the exploitation/investment trade-off which is the focus of this paper. Somaini and Einav (2013) focus on antitrust issues and their paper is discussed in Section 3.1. In a considerably different setting, Taylor (2003) finds that with more than two firms competition over switching consumers lowers prices and firm profits. While similar results are obtained here, the number of firms required for switching costs to be pro-competitive is not necessarily two, but is a function of the level of switching costs and firms’ discount rate. The remainder of the paper is organized as follows. Section 2 describes the setup of the model. The symmetric equilibrium price is determined in Section 3. Section 3 also contains a discussion relating the model and equilibrium price to the previous literature. Section 4 analyzes the equilibrium price from Section 3 and contains the main results. Section 5 tests the robustness of the main results by considering a continuous distribution of switching costs and Section 6 concludes. 2 The Model The model developed in this paper combines consumer switching costs with the discrete choice logit model and allows for competition to take place over an infinite horizon. The logit model is typically presented as a static discrete choice product differentiation model and is thoroughly discussed in Train (2003, Chapter 3), Anderson, de Palma, and Thisse (1992) and elsewhere. Consumer switching costs and overlapping generations of consumers are modeled similar to Beggs and Klemperer (1992). The model presented here is discussed in the context of the relevant literature at the end of the next section. 4 There are N ≥ 2 single product firms. Each firm produces a unique variety of the differentiated product and each consumer purchases exactly one differentiated product. Consumers choose the product variety in each period which maximizes their utility. Firms choose their price to maximize the discounted sum of profits over time. Competition takes place over an infinite horizon. 2.1 Consumers i = y − pi − s0 I + εi . The random utility specification of consumer z at time t for product i is Uzt t z zt y is the consumer’s income/baseline valuation of the product which is identical for all consumers. It is assumed that y is sufficiently large so that all consumers purchase, but that y is not infinite. pit is the price of Firm i’s product (i = 1, 2, . . . , N ) at time t. s0z indicates consumer z’s switching cost and I = {0, 1} indicates whether consumer z switches. Consumers also have preferences for the product varieties where εizt is consumer z’s product specific valuation of product i at time t. The following assumptions are made concerning consumer preferences: Assumption 1. Consumers are myopic with preferences described by the random utility function, −ε i , where εi is i.i.d. and follows a Gumbel distribution with density f (ε) = e−ε e−e , and s0 = Uzt zt z s0 ≥ 0. The assumption that εizt is i.i.d. both over time and product varieties along with the preference i is maintained throughout the paper. It is also assumed that consumers are specification of Uzt myopic and choose one product that maximizes their random utility in the current period. In each period every consumer knows only their own realization of brand preferences (or their own preference shock), εizt ∀ i = 1, 2, . . . , N . Alternatively, consumers always know their baseline valuation y and switching cost s0z since these do not change over time. Prices, pit are known to everyone. Consumers have switching costs of s0z = s0 which they incur if a consumer switches products i is an indicator function which takes a value of 1 when consumer after an initial purchase. I in Uzt z switches after an initial purchase and 0 otherwise. I = 1 if the purchase decision in period t − 1 is different than the purchase decision in period t regardless of the entire history of purchases. Assumption 1 forces all consumers to have the same switching cost of s0 although the level of the 5 switching cost may change.7 At the beginning of period t, there are Mt consumers from the previous period. A fraction, α ∈ [0, 1], of these old consumers leave the market forever and the remaining (1−α)Mt consumers remain in addition to ρ new consumers who have just entered the market. Both new and the remaining old consumers choose one product to maximize their current period utility. New consumers do not incur switching costs, but old consumers incur switching costs if they purchase a different product than in the previous period. The number of consumers at the beginning of each period evolves according to Mt+1 = ρ + (1 − α)Mt . Throughout most of the paper M0 = 1 and ρ = α so that there is always a steady state unit mass of consumers and α is the renewal rate of consumers or the fraction of new consumers in each period. Additionally, when M0 = 1 and ρ = α a firm’s share is equivalent to their demand. i . Each consumer chooses a product i in period t to maximize current utility: maxi∈{1,2,...,N } Uzt i = V i + εi . The Let Vti ≡ y − pit − s0 I, so that the current utility can be expressed as Uzt zt t product purchased in period t matters as it determines what product the consumer does not have to pay switching costs for in period t + 1. In period t, consumer z chooses product i if Vti + εizt > Vtj + εjzt ∀ j 6= i. The probability that consumer z chooses product i in period t is8 i eVt i Pzt = PN j Vt j=1 e 2.2 . (1) Firms Each firm serves potentially three different types of consumers simultaneously in every period: new consumers and two types of old consumers. There are ρ new consumers who just entered the market. Since new consumers have not made a previous purchase, they do not incur any switching costs. There are (1−α)Mt old consumers with switching costs and each old consumer is categorized by their purchase history. An old consumer is either a repeat customer and purchases from the same firm as in the previous period, or an old consumer switches and purchases from a different firm. The probability that Firm i attracts each of the three different types of consumers is as follows. 7 8 This assumption is relaxed later on in the paper when a continuous distribution of switching costs is considered. See Train (2003, p. 40 and pp. 78-79) or Anderson et al. (1992, pp. 39-40) for the derivation. 6 i , Firm i attracts an With probability PNi t , Firm i attracts a new consumer, with probability POL t ki Firm old (and loyal) consumer who previously purchased from Firm i and with probability POS t i attracts an old (and switching) consumer who previously purchased from Firm k 6= i. Firm i’s market share in period t − 1 is σti . Firm i’s market demand in each period is i Qit (pt , σ t ) = ρPNi t + (1 − α)Mt σti POL + (1 − α)Mt t N X ki σtk POS t (2) k6=i where each term on the right hand side from left to right is the demand from new consumers, old and loyal consumers, and old and switching consumers. pt is the N × 1 vector of prices in period t and σ t is the N × 1 vector of period t − 1 market shares. In each period, firms choose their price to maximize their discounted sum of profits. The following assumptions apply to the firm’s problem. Assumption 2. Assume all firms have the same discount factor δ ∈ (0, 1) and that a firm’s price in each period is an element of a nonempty, compact, and convex set P. Consider the set of possible prices, P, to have an upper and lower bound, and in each period, a firm chooses a price between the bounds. Bounding a firm’s price from below prevents a firm from paying consumers an infinite sum to switch and bounding from above does not allow a firm to receive an infinite sum from a captive consumer. The period t profit for Firm i is πti (pt , σ t ) = Qit (pt , σ t )(pit −ci ) where ci is the constant marginal cost of Firm i. For simplicity, let ci = 0. Firm i’s formal problem is max pit ∞ X i δ t πti (pt , σ t ) subject to σt+1 Mt+1 = Qit (pt , σ t ). (3) t=0 The solution to Firm i’s dynamic programing problem outlined in equation (3) is a pricing rule pit (σ t ) which is a function of all firms previous shares. Given initial market shares for all firms σ 0 , a set of pricing rules satisfying equation (3) for each firm characterizes a competitive equilibrium of the model. 7 3 Symmetric Equilibrium of the Model In this section, the pure strategy pricing rule is determined for a symmetric Markov perfect equilibrium (MPE) of the model presented in Section 2. While a formal existence argument of a MPE of the model is not presented, the intuition of why such an equilibrium exists is. Also, the pure strategy pricing rule is not necessarily unique and a proof of uniqueness is left for future work. The symmetric MPE determined considers both symmetry across firms within a period and symmetry across time (a steady state). From the outset of the problem, we know that the perperiod symmetric subgame equilibrium of the model will involve all firms having identical market shares of σti = 1/N and each firm charging the same price at an instance in time, pit = pjt ∀i, j. If the size of the market is changing over time, then it is possible that pit 6= pit0 . Steady state conditions of ρ = α along with M0 = 1 ensure that the size of the market does not change over time and a firm’s market share is also equal to their demand. When firms’ prices are symmetric within every period, pit = pjt ∀i, j and Lemmas 1 and 2 describe resulting equilibrium properties of the model.9 These properties are used to simplify the symmetric pricing rule determined in Proposition 1, but only after the first order conditions of the firm’s problem are determined. After the symmetric pricing rule is established, the model is discussed in the context of the previous literature. It is first shown how symmetric prices within each period influence the consumer’s problem and how this effect works it’s way into the firm’s problem. The general probability that any one consumer chooses any one product in a given period is defined by equation (1). The demand segments for each firm are explicitly defined in the following lemma.10 0 Lemma 1. Let s = 1 − e−s . If pit = pjt ∀ i, j: PN = 1 N POL = 1 (1 − s)(N − 1) + 1 POS = (1 − s) (1 − s)(N − 1) + 1 (4) The probability that a new consumer purchases from Firm i, PN , is equally likely for all firms since prices are symmetric within periods and new consumers do not face any switching costs. For old consumers with switching costs equal to s0 , switching costs are incurred if any of the N − 1 9 10 Since the focus is on an MPE, the time subscript is omitted when appropriate to avoid redundancy. Omitted proofs are in the appendix. 8 products not purchased in the previous period are chosen. The difference between POL and POS is that repeat consumers (POL ) do not incur switching costs, while consumers who switch from Firm k to Firm i incur a switching cost of s0 . Notice that while s0 ∈ [0, ∞), s ∈ [0, 1]. Consumer switching costs of zero, s0 = 0, correspond to s = 0 and infinitely high consumer switching costs correspond to s = 1. i M i i Market shares evolve according to equation (2) where σt+1 t+1 = Qt (σt ). Incorporating the steady state conditions of, ρ = α and M0 = 1 results in a simplified version of demand and the evolution of market share. 0 Lemma 2. Let s = 1 − e−s . If pit = pjt ∀ i, j, ρ = α and M0 = 1 Firm i’s demand/share is Qi = (1 − α)(1 − s(1 − σ i )) α + N (1 − s)(N − 1) + 1 (5) and the market share evolves according to the following equation. σti t 1 1 (1 − α)s i = + σ0 − . N N (1 − s)(N − 1) + 1 (6) Lemma 2 leads to an expected result concerning the symmetry of the model: when initial market shares are symmetric, σ0i = 1/N , and prices are symmetric within periods, market shares are symmetric across periods, σ i = 1/N . Although it seems dubious to expect firms’ prices to be symmetric when initial market shares are not symmetric, note that if initial market shares are not symmetric but firms’ prices are symmetric within every period, market shares are converging to σ i = 1/N over time. Convergence occurs because the term on the right hand side of equation (6) is between zero and one as 0 ≤ (1 − α)s ≤ (1 − s)(N − 1) + 1. When initial market shares are symmetric, a pure strategy pricing rule of a MPE dictates that all firms charge the same price, and when all firms charge the same price within a period, market shares remain symmetric (Lemma 2). At the steady state symmetric equilibrium, each firm’s market share is σti = 1/N and each firm announces a price of p∗ = pit (σti = 1/N ) in every period. Individual firm profits are the price times market share, p∗ /N , and industry profits are equivalent to the price, p∗ . The following proposition describes the pure strategy pricing rule of a symmetric MPE and is the main result of this section. 9 Proposition 1. Given that a symmetric MPE exists, σ0i = 1/N for i = 1, . . . , N , ρ = α and M0 = 1, the steady state symmetric pure strategy MPE price of the problem defined by equation (3) is p∗ = N ((1 − s)(N − 1) + 1)((1 − s)(N − 1) + 1 − δ(1 − α)s) (N − 1) (α((1 − s)(N − 1) + 1)2 + (1 − α)(1 − s)N ((1 − s)N + 2s)) (7) 0 where s = 1 − e−s . The steady state symmetric price is a function of four parameters: α, δ, s and N . N ≥ 2 is the number of identical firms in the market, α ∈ [0, 1] indicates the fraction of new consumers in the market, and δ ∈ (0, 1) is the firm’s discount factor. To keep the denominator of equation (7) non-zero, the degenerate case of infinite switching costs with no new consumers (s = 1 and α = 0) is not considered. It is important to note that Proposition 1 is based on the existence of a symmetric MPE, and if a symmetric MPE fails to exist then the following analysis is vacuous. While a formal existence argument is not provided, it is intuitive that such an equilibrium exists. The model presented is symmetric as all firms have the same costs, all profit functions are constructed in a similar manner, and the initial state is assumed to be symmetric. The firm’s payoff function is quasi-concave and continuous – both properties utilized in formal existence arguments. Furthermore, the dynamics of the model are relatively simple by construction as the state space and set of feasible strategies do not change over time, and the set of feasible strategies are not constrained by the current state. A simple analysis of the pricing rule in equation (7) and the choice probabilities of equation (4) provides some intuitive validation. For instance, if switching costs are absent from the model we expect the familiar symmetric logit price of N/(N − 1) and the probability of a repeat purchase i ki . Switching costs do not influence the model when is equally as likely as switching, POL = POS t t either all consumers are new to the market, α = 1, or when all consumers have switching costs equal to zero, s = 0. Either of these cases result in the symmetric logit price of N/(N − 1) and i ki = 1/N . Additionally, when all consumers have infinite switching costs, s = 1, no POL = POS t t ki = 0 and P i consumers switch (POS OLt = 1). t 10 3.1 Discussion & Relation to the Previous Literature The model presented in Section 2 is compared to the earlier models presented by von Weizsäcker (1984) and Beggs and Klemperer (1992) although von Weizsäcker’s equilibrium concept is different. Beggs and Klemperer’s model has new consumers entering the market in each period with all consumers having infinite switching costs (α 6= 0 and s = 1). Alternatively in von Weizsäcker’s model, consumers do not have infinite switching costs and no new consumers enter the market since old consumers live forever (α = 0 and s 6= 1). The model presented in Section 2 could be considered more general than von Weizsäcker (1984) and Beggs and Klemperer (1992) as it accounts for both scenarios and the effect of changing the amount of new consumers without switching costs, α, can be distinguished from the effect of changing the level of switching costs s. The most significant difference between the model in this paper and the relevant previous literature is the use of a logit model to characterize the product differentiation.11 Beggs and Klemperer (1992), Doganoglu (2010), Fabra and Garcı́a (2015), Rhodes (2014), Shin and Sudhir (2008), Somaini and Einav (2013), Villas-Boas (2015), and von Weizsäcker (1984) characterize the product differentiation in their models by utilizing a linear city Hotelling type model where consumers are uniformly distributed over the line and the two firms are located at the opposite ends of the product space. One notable exception is Cabral (2014) who allows for a general form of product differentiation in a duopoly setting. The use of the logit model/Gumbel distribution in this paper has two key advantages. First, the model is able to easily incorporate more than two firms. The Hotelling framework is usually limited to two firms at the opposite end of the product space. One exception in this regard is Somaini and Einav (2013) where they develop an extension to the Hotelling framework allowing for any number of firms.12 Secondly, the logit model is not subject to price under-cutting strategies present in the Hotelling model with linear transportation costs.13 The absence of under-cutting strategies makes the analysis simpler because the parameter space does not need to be restricted as in the aforementioned models. 11 Both Dubé et al. (2009) and Arie and Grieco (2014) examine a switching costs model with a logit model characterizing the product differentiation. The theoretical results of Dubé et al. (2009) are obtained by using numerical methods on a restricted version of the model and Arie and Grieco (2014) only consider the case of no/small switching costs. 12 Other switching cost models allow for more than two firms, but additional firms are regulated to the fringe (see Biglaiser, Crémer, and Dobos (2013)) and/or a pure strategy price equilibrium does not exist (see Farrell and Klemperer (2007, pp. 1985-1986 and footnote 31), Biglaiser et al. (2013) and Rosenthal (1980)). 13 See d’Aspremont, Gabszewicz, and Thisse (1979). 11 While the product differentiation assumptions made in this paper have their advantages, these assumptions create differences in how brand preferences are correlated across products and over time. In Hotelling type models, brand preferences across products are perfectly negatively correlated within periods, but here brand preferences across products are independent within periods. This difference could be viewed as a consequence of allowing more than two products/firms or could be seen as substituting one extreme case (correlation of −1) for another (correlation of 0). Additionally, the model in this paper has brand preferences which are not correlated over time and are independent across periods.14 Most other models also have brand preferences that are not serially correlated over time. Two exceptions are von Weizsäcker’s (1984) model where individual brand preferences change over time with some positive probability of an independent redraw, and Cabral (2014) who considers the serial correlation of preferences as an extension. According to Villas-Boas (2015), if consumer preferences are less stable over time, it is expected that prices will be lower. Besides assumptions regarding the correlation of consumer preferences, it is also assumed in this paper that consumers are myopic. This is not an uncommon assumption in the literature, and Villas-Boas (2015) indicates that prices are lower with myopic consumers in switching cost models. Alternatives include allowing for two generations of overlapping consumers (Doganoglu, 2010; Rhodes, 2014; Somaini & Einav, 2013) or infinitely lived consumers with discounting (Cabral, 2014; Fabra & Garcı́a, 2015). One of the main difficulties with incorporating consumer discounting in the current model is that the renewal rate of consumers is parameterized as α, and in other studies α is fixed. With two generations of overlapping consumers, α = 0.5, and with infinitely lived consumers, α = 0. The benefit of parameterizing α while having myopic consumers is the direct connection made to the sign of ∂p∗ ∂α and whether switching costs are anti- or pro-competitive.15 Another important difference between the current analysis and the previous literature is with the determination of the pricing rule. The focus in this paper is only on the symmetric steady state pricing rule. A downside to this approach is that the current analysis does not address asymmetries in market shares as considered by Cabral (2014), Fabra and Garcı́a (2015), and Rhodes (2014). Cabral (2014) emphasizes that when market shares are more asymmetric, switching costs are anti14 The independence of brand preferences over time greatly simplifies the analysis and is critical in determining an analytical solution to the pricing rule in equation (7). 15 See Corollary 1 and Corollary 4. 12 competitive. Even though the focus here is on symmetric outcomes, it is still found that switching costs can be anti-competitive. Incorporating asymmetries into the analysis is not likely to change the nature of the results, but it should change the pertinent parameter thresholds for which switching costs are either pro- or anti-competitive. The same is true with consumer discounting. When consumers are myopic, it is more likely that switching costs are pro-competitive, but the main point is not that switching costs are either anti- or pro-competitive. The main point of the analysis is to determine subsets of the parameter space for which switching costs are anti- or pro-competitive. Upsides to the current approach include the ability to determine an analytical solution to a closed form pricing rule and the MPE pricing rule not depending on a functional form restriction. Doganoglu (2010), Rhodes (2014), and Somaini and Einav (2013) restrict their analysis to only consider linear pricing rules in a MPE. In the context of the current model, a symmetric steady state linear pricing rule takes the form pit = L1 + L2 σti with the pricing rule being an affine transformation of the market share. In the current analysis, a linear pricing rule is inappropriate when α 6= 1, s 6= 0 and N > 2. 4 Analysis In this section, the symmetric steady state price, profits and consumers’ welfare are analyzed. The comparative statics of the pricing rule are determined with respect to three of the four model parameters: δ, α and s . Then it is determined under what conditions switching costs lead to lower or higher prices when compared to a market without switching costs. The analysis is simplified by rewriting the symmetric equilibrium price in equation (7) as follows. Let X = (1 − s)(N − 1) + 1, Y = X − δ(1 − α)s, and Z = αX 2 + (1 − α)(1 − s)N (X + s). Then the symmetric equilibrium price can be rewritten as p∗ = N XY . (N − 1)Z (8) Since δ, α and s are all bounded between 0 and 1 and N ≥ 2, it follows that X, Y and Z are all greater than or equal to zero. 13 4.1 Comparative Statics The symmetric steady state price of equation (7) is decreasing in δ. Using the simplified price in equation (8), ∂p∗ NX = ∂δ (N − 1)Z ∂Y ∂δ ≤ 0. As δ increases, firms care more about future profits and lower their price sacrificing current profits to invest in future market share. The partial derivative of the equilibrium price with respect to α is ∂p∗ NX = ∂α (N − 1)Z 2 ∂Y ∂Z Z . −Y ∂α ∂α The sign of ∂p∗ /∂α is determined by the sign of (Z∂Y /∂α − Y ∂Z/∂α). Simplification yields ∂Y ∂Z Z −Y = sX(δ(1 − s)N − (1 − δ)s) ∂α ∂α (9) and the sign of ∂p∗ /∂α depends on the relationship between N , s and δ. If δ(1−s)N > (1−δ)s, then an increase in the renewal rate of consumers, α, will raise the equilibrium price. This result occurs when the number of firms (N ) is large, switching costs (s) are small, or when the discount factor (δ) is large. At the beginning of each period, the number of consumers partially locked in to Firm i is (1 − α)/N and the number of consumers partially locked in to other firms is (1 − α)(N − 1)/N . With a large number of firms, there are more consumers who are partially locked in to other firms and this places downwards pressure on each firm’s price as they try to attract consumers who have switching costs. An increase in the renewal rate of consumers α, relieves the downwards pressure on the price as a firm can focus more on consumers without switching costs, and the increase in α results in an increase in the price. Alternatively if δ(1 − s)N < (1 − δ)s, (small N , large s, or small δ) an increase in the fraction of new consumers, α will decrease the equilibrium price. A marginal change in the level of switching cost s influences the equilibrium price where ∂p∗ N = ∂s (N − 1)Z 2 ∂X ∂Y ∂Z YZ + XZ − XY ∂s ∂s ∂s and the sign of ∂p∗ /∂s is determined by term on the far right hand side. The term of interest which 14 determines the sign is ∂X ∂Y ∂Z = (1 − α)N 2(1 − δ)sX + δαs2 − δ(1 − s)2 N 2 YZ + XZ − XY ∂s ∂s ∂s (10) and may be either positive or negative depending on the relationship between the parameters. Examining the bracketed term of equation (10), an increase in switching cost may lead to a decrease in price (equation (10) < 0) if firms are sufficiently patient (large δ), switching costs are low, there are more old consumers (small α), or there are many firms in the market. 4.2 When do Switching Costs Lead to Lower Prices? The effects of marginal changes in the fraction of new consumers and the level of switching costs on the equilibrium price determined in the previous section depend on the parameter specification. The sign of ∂p∗ /∂α depends on the relationship between N , s and δ and the sign of ∂p∗ /∂s depends on the relationship between N , s, δ and α. The interpretation of these results are better explained when one compares the equilibrium price with switching costs to the equilibrium price without switching costs. With no switching costs, s = 0 and the equilibrium price is N/(N − 1). The difference in price created by switching costs is ∆p∗ = p∗ − N N = (XY − Z) N −1 (N − 1)Z (11) and the sign of ∆p∗ is determined by the sign of XY − Z. Expansion of this result yields XY − Z = (1 − α)s((1 − δ)s − N δ(1 − s)) and the sign of XY − Z is determined by the sign of (1 − δ)s − N δ(1 − s). (12) The sign of equation 12 can be either positive or negative so that switching costs can lead to higher or lower prices. Equation 12 provides additional insight to the comparative statics of the model. The sign of ∂p∗ /∂α is determined by the sign of equation (9). When switching costs lead to lower prices, δ(1 − s)N > (1 − δ)s, and an increase in the fraction of new consumers, α, will raise the equilibrium 15 price. An increase in α effectively reduces the importance of switching costs because there will be more new consumers in each period without switching costs. If switching costs lower prices (∆p∗ < 0) and the importance of switching costs is reduced by increasing α, then the effect of an increase in α will be to increase the price. Alternatively, if prices are higher with switching costs, a marginal increase in α causes the equilibrium price to decrease. The following corollary summarizes this relationship. Corollary 1. With homogeneous switching costs, ∆p∗ < 0 if and only if ∂p∗ /∂α > 0. Examining equation (11), switching costs do not influence the price for the two extreme cases of no switching costs, ∆p∗ (s = 0) = 0, or no old consumers, ∆p∗ (α = 1) = 0. With infinite switching costs, ∆p∗ (s = 1) = N (1 − δ)(1 − α) >0 (N − 1)α and switching costs always lead to higher prices. This is the result obtained by Beggs and Klemperer (1992). In their model, they assume that consumers have infinite switching costs, but allow the fraction of new consumers in the market to vary. By changing α when s = 1, prices are always higher with switching costs, but the magnitude of the price difference varies. Avoiding the extreme cases of s = 0, s = 1 and α = 1, switching costs lead to a lower equilibrium price when switching costs are sufficiently low, the discount rate is sufficiently large and there are many firms. Surprisingly the direction of the price difference does not depend on the fraction of new consumers, α, but α does influence the magnitude of the difference. The discount rate and the number of firms influence the model by changing the relative returns to the conflicting investment and exploitation strategies of the firm. As δ increases or N increases, the value to the firm of exploiting its existing consumer base is diminished so firms lower prices to invest in future market share. The main result of the paper provides a more general result with respect to the parameters of the model and is as follows. Corollary 2. Assume s 6= {0, 1} and α 6= 1, then there exists an N 0 , s0 and δ 0 where N0 = (1 − δ)s δ(1 − s) s0 = Nδ N δ + (1 − δ) 16 δ0 s = 1 − δ0 N (1 − s) such that for any N > N 0 , s < s0 , or δ > δ 0 switching costs lead to lower prices (∆p∗ < 0). Corollary 2 is a direct result of equation (12). With respect to N 0 , as long as switching costs are not infinite and as long as some consumers might switch, switching costs lower the equilibrium price if there are a sufficiently large number of firms in the market. The sufficiently large number of firms required to lower prices is increasing in the level of switching costs and decreasing in the discount rate, but irrespective of the level of switching costs and the discount rate there is always some number of firms such that prices are lower with switching costs. Alternatively if the number of firms is sufficiently small, then prices are higher with switching costs. One can also interpret Corollary 2 as indicating that for any number of firms, as long as switching costs are relatively low (s < s0 ) or the discount factor is relatively large (δ > δ 0 ), prices are lower with switching costs. This interpretation gives more general support to previous studies which assume N = 2. The numerical results obtained by Dubé et al. (2009) are consistent with Corollary 2 and this interpretation also supports Cabral’s (2014) finding that switching costs may lower the price. The contribution of Corollary 2 in the context of these studies is to provide analytical results with a single price. Corollary 2 is also more general as the number of firms is not restricted to N = 2 which is important because the number of firms plays a major role in determining whether prices are higher or lower with switching costs. A similar story occurs with the sign of ∂p∗ /∂s which is determined by the sign of equation (10) and results in the following corollary. Corollary 3. Assume s 6= {0, 1} and α 6= 1, then there exists an N 00 , s00 , and δ 00 where N 00 = 1 s 1 − δ + 1 − δ 2 (1 − α) 2 δ(1 − s) s00 = N N δ + (1 − δ) − 1 − δ 2 (1 − α) 1 2 δ(N − 1)2 + 2(N − 1) + δ(1 − α) δ 00 2s((1 − s)(N − 1) + 1) = 00 1−δ (1 − s)2 N 2 − αs2 such that for any N > N 00 , s < s00 , or δ > δ 00 a marginal increase in switching costs lowers the price (∂p∗ /∂s < 0). Comparing the measures from Corollary 3 to Corollary 2 reveals that N 0 < N 00 , s00 < s0 , and 17 δ 0 < δ 00 . For a sufficiently large number of firms (N > N 00 ), sufficiently low switching costs (s < s00 ), or sufficiently large discount rate (δ > δ 00 ), the equilibrium price is lower with switching costs than without and a marginal increase in switching costs leads to a decrease in the equilibrium price.16 Note that N 00 , s00 , and δ 00 are all functions of the other variables and in the case of N 00 , N 00 is increasing in s and decreasing in δ. The price decreases with an increase in switching costs because the overall effect of switching costs is to lower the price. Notice that while the fraction of new consumers, α, does not affect whether prices are higher or lower with switching costs (∆p∗ ), α does partially determine the sign of ∂p∗ /∂s. The following proposition summarizes the main results thus far: Proposition 2. Assume s 6= {0, 1} and α 6= 1, then: 1. For N < N 0 , s > s0 , or for δ < δ 0 , ∆p∗ > 0, ∂p∗ /∂α < 0 and ∂p∗ /∂s > 0. 2. For N 0 < N < N 00 , s00 < s < s0 , or for δ 0 < δ < δ 00 , ∆p∗ < 0, ∂p∗ /∂α > 0 and ∂p∗ /∂s > 0. 3. For N > N 00 , s < s00 , or for δ > δ 00 , ∆p∗ < 0, ∂p∗ /∂α > 0 and ∂p∗ /∂s < 0. The sign of ∆p∗ determines whether prices are higher are lower with switching costs where if ∆p∗ < 0, prices are lower with switching costs. The sign of ∂p∗ /∂α determines whether a marginal increase in the amount of new consumers (who do not have switching costs) leads to an increase or decrease in price. Finally, the sign of ∂p∗ /∂s determines a marginal increase in switching costs affects prices. 4.3 The Duopoly Case While the main contribution of the paper is to examine how the number of firms influences the price in a switching cost model, the analysis is also relevant to other duopoly models with switching costs. In a duopoly setting, N = 2 and the effect of changes in α, δ, and s are examined. From The result that ∂p∗ /∂s < 0 when s < s00 extends Proposition 2 of Doganoglu (2010) who finds that, in the notation of the current model, ∂p∗ /∂s < 0 conditional on s = 0. 16 18 Section 4.1, the relevant comparative statics are ∂p∗ (N = 2) −2s(1 − α)(2 − s) ≤0 = ∂δ 4(1 − s) + αs2 ∂p∗ (N = 2) −2s(2 − s)2 (s − δ(2 − s)) = ∂α (4(1 − s) + αs2 )2 ∂p∗ (N = 2) 4(1 − α)(2s(2 − s) − δ(4(1 − s) + s2 (2 − α))) = ∂s (4(1 − s) + αs2 )2 and from Section 4.2, the price difference is ∆p∗ (N = 2) = 2s(1 − α)(s − δ(2 − s)) . 4(1 − s) + αs2 Figure 1: Duopoly Parameter Space 1 δ Price is Lower with Switching Cost δ= s 2−s 1 3 Price is Higher with Switching Cost 0 0 1 2 1 s As before, a marginal increase in δ leads to a marginal decrease in the price. The sign of both ∆p∗ (N = 2) and ∂p∗ (N =2) ∂α depend on the sign of s − δ(2 − s), where s − δ(2 − s) < 0 implies that prices are lower with switching costs and a marginal increase in α results in a higher price. Figure 1 19 illustrates the parameter space over δ and s where switching costs lead to lower or higher prices. In the duopoly case, switching costs lead to lower prices when the level of switching costs is low or the discount factor is high. An examination of ∂p∗ (N =2) ∂s is complicated as the sign of this comparative static depends on α, δ and s. All things being equal, ∂p∗ (N =2) ∂s < 0 when δ is relatively large, α is relatively small, and s is relatively small. 4.4 Product Differentiation and Switching Costs Also of interest is the level of switching costs (s0 ) compared to the extent of product differentiation in the model (distribution of the εizt ’s). Because a logit model is used to characterize the product differentiation in this paper, one may suspect that the logit model drives certain results rather than switching costs. Implicitly assumed in the model is that the variance of εizt is equal to π 2 /6.17 If instead the variance was π 2 β 2 /6, with β > 0, then s would be expressed as ŝ = 1 − e −s0 β . The level of switching costs is determined by s0 , and β determines the extent of product differentiation. As β approaches zero, the variance of εizt approaches zero and there is no product differentiation because consumers have the same brand preferences for all products. As β approaches zero, ŝ approaches 1 (for a fixed s0 ) and the same effect is achieved by increasing consumer switching costs s0 (holding β constant). Alternatively as β gets large, consumers’ brand preferences are more varied increasing the relative importance of product differentiation. When β gets large, ŝ approaches zero and the same effect is achieved by decreasing consumer switching costs s0 . Examining changes in ŝ, there is a direct trade-off between the importance of switching costs and product differentiation in that when the magnitude of switching costs increases, the effect of product differentiation decreases. A marginal increase in β results in more switching in equilibrium (POL decreases and POS increases) as the marginal increase in β reduces ŝ. Examining the change in the price, let p∗ (s = ŝ) be the price in equation (7) with s replaced by ŝ. The symmetric steady state equilibrium price is p̂ = βp∗ (s = ŝ).18 As β approaches zero, consumers perceive the different products as homogenous, and the price approaches zero (marginal cost) as consumers purchase from the firm with the lowest price. 17 See Train (2003, p. 44) and Anderson et al. (1992, pp. 59-60). Note that π refers to the mathematical constant and not profits. 18 p̂ is obtained using the procedure in the proof of Proposition 1. The major difference is that the partial derivative of a logit probability, P , with respect to p̂ is P (P − 1)/β. 20 For a marginal increase in β, ∂ p̂ ∂p∗ (s = ŝ) ∂ŝ = p∗ (s = ŝ) + β ∂β ∂ŝ ∂β where ∂ŝ ∂β < 0 and the sign of ∂p∗ (s=ŝ) ∂ŝ (13) is determined by the sign of equation (10). As β increases, there are two different effects on p̂ as indicated by the two different terms in equation (13). As β increases, consumers have more diverse brand preferences and firms are able to raise their price by p∗ (s = Ŝ). A change in β also affects p̂ through a change in ŝ. An increase in β decreases ŝ and if ∂p∗ ∂s > 0 the overall effect of the second term in equation (13) is negative. Examining the sign of the combined effect, when β is sufficiently small, ∂ p̂ ∂β > 0 as the second term in equation (13) is negligible, and an increase in the dispersion of brand preferences results in an increase in the price. When β is sufficiently large, ŝ < s00 so that β, numerical simulations indicate that ∂ p̂ ∂β ∂p∗ (s=ŝ) ∂ŝ < 0, and ∂ p̂ ∂β > 0. For intermediate values of < 0 as the second term in equation (13) is negative and dominates the first term. This suggests that increases in the dispersion of brand preferences has a non-monotonic effect on the price. The dispersion of brand preferences also affects whether prices are higher or lower with switching costs. In this case what matters is the ratio of the level of switching costs to the dispersion of brand preferences: s0 /β. Similar to equation (11), the difference in prices created by switching costs is ∆p̂ = β p∗ (s = ŝ) − N N −1 . While ∆p̂ is scaled by β, the sign of ∆p̂ is only influenced by β through ŝ. When the level of switching costs is large compared to the dispersion of brand preferences, s0 /β is large, ŝ is large, and it is more likely that ŝ > s0 so that ∆p̂ > 0. When brand preferences are more dispersed and the level of switching costs are low, s0 /β is small, ŝ is small, and it is more likely that ŝ < s0 so that the price is lower with switching costs. This result corresponds directly to the finding in Shin and Sudhir (2008) except here ŝ is a function of N and δ and Shin and Sudhir’s (2008) version of ŝ is such that ŝ = 1. 21 4.5 Welfare & Policy Implications Missing from the analysis thus far is a discussion of how the model parameters influence firm profits and consumer welfare. Individual firm profits are πi∗ = p∗ /N and total industry profits are equivalent to the equilibrium price. The analysis of Sections 4.1 and 4.2 applies directly to the analysis of profits. If a change in δ, α or s leads to a lower (higher) equilibrium price, then equilibrium profits are lower (higher) as well. Per-period consumer surplus/welfare is measured as the consumer’s expected level of utility. The expected utility for each of the α new consumers is CSN ew = y − p∗ + γ + ln(N ) and the expected utility for each of the 1 − α old consumers is CSOld = y − p∗ + γ + ln((1 − s)(N − 1) + 1) where γ is Euler’s constant. The overall level of per-period consumer surplus is CS = αCSN ew + (1 − α)CSOld . Marginal changes in δ, α and s influence CSN ew and CSOld by way of a change in the equilibrium price where an increase in price decreases welfare. The exception is ∂CSOld /∂s because an increase in switching cost affects old consumers through the price and by the switching that might occur. The change in welfare experienced by old consumers due to a marginal change in the level of switching costs is ∂p∗ N −1 ∂CSOld =− − ∂s ∂s (1 − s)(N − 1) + 1 and the term on the right is the decrease in old consumers’ welfare attributed to switching. From a social efficiency perspective, a change in δ only leads to a transfer between consumers and firms and does not change the overall level of efficiency. An increase in the switching cost, s, decreases social efficiency as switching costs are modeled to be socially wasteful. Related to the switching cost, an increase in the amount of new consumers, α, increases social efficiency because new consumers have no switching costs. With more firms, which have zero costs by assumption, 22 consumers have a larger variety to choose from and overall welfare increases.19 Policy makers have the ability to manipulate s and N . N can be manipulated through programs which discourage or encourage entry and s, for example, has been lowered by the introduction of phone number portability policies. If policy makers are only concerned about overall welfare measures, the model indicates that switching costs should be lowered and/or firm entry/competition should be encouraged. If policy makers are more concerned with consumer welfare and if they can encourage sufficient competition (large enough N ), prices will be lower with switching costs and consumers will have more variety. The downside to encouraging entry/competition is that it results in more consumers switching. If N < N 00 , δ < δ 00 and s > s00 policy makers should decrease the switching costs until s = s00 which would lower the equilibrium price and the cost of switching. If N > N 00 , s < s00 , or if δ > δ 00 policy makers might increase switching costs to lower the price. This clearly benefits new consumers with no switching costs. An increase in switching costs benefits old consumers if ∂CSOld /∂s > 0. Numerical simulations indicate that if α is small, δ is large, s is small and N = 2 it is possible to have ∂CSOld /∂s > 0. Considering the extreme case of α = 0 and s = 0 where ∂p∗ /∂s is minimized and ∂CSOld /∂s is maximized, equation (10) simplifies and ∂p∗ /∂s = −δ/(N − 1). For this extreme case, ∂CSOld /∂s = ∂CS/∂s and δ N −1 ∂CS = − . ∂s N −1 N For more than 2.618 firms, its not possible for ∂CS/∂s > 0 as the upper bound of δ is one. Thus in a duopoly setting with restrictions imposed upon α, δ and s, the welfare of all consumers may increase due to a marginal increase in the switching cost. For more than 2.618 firms, it is not possible to increase the welfare of all consumers with a marginal increase in switching costs. 5 A Continuous Distribution of Switching Costs In the model outlined in Section 2, all old consumers have a switching cost of s0 so that switching costs are homogeneous across consumers and over time. While the results listed in Proposition 2 are true for any level of positive finite switching cost, one might question the validity of the results 19 This result differs from Taylor (2003) where efficiency is maximized with only one firm. 23 since all consumers are forced to have the same level of switching cost. One of the contributions of Biglaiser et al. (2013) is to show that switching cost heterogeneity has important implications for firms’ strategies. This section evaluates the robustness of the homogeneous switching cost assumption and allows for a continuous distribution of switching costs across consumers. Switching cost heterogeneity is modeled similar to the formulation used by Taylor (2003). The specification of s0z defined in Assumption 1 is modified by the following assumption which is maintained throughout this section. Assumption 3. Let s0z = εszt where εszt is i.i.d. and follows an exponential distribution with density g(ε) = e−ε . εszt is independent across consumers and over time, similar to Taylor’s (2003) specification. The exponential distribution is used because it incorporates more realistic switching cost heterogeneity while still allowing for an analytical solution. Allowing for switching cost heterogeneity does not significantly impact the determination of the steady state symmetric pure strategy pricing rule. The biggest change is in the general logit probability, equation (1), which is now i Pzt = Z 0 ∞ i eVzt PN j=1 j eVzt g(εszt )dεszt and is the probability that consumer z chooses product i in period t. The difference between the above equation and equation (1) results from the fact that the expected level of switching cost must now be determined. This difference influences the results of Lemmas 1 and 2. Updating these lemmas to include switching cost heterogeneity results in the following. Lemma 3. If pit = pjt ∀ i, j, PN = 1 N POL = ln(N ) N −1 POS = N − 1 − ln(N ) (N − 1)2 (14) are the demand functions for each consumer type. As before, at the steady state symmetric equilibrium each firm’s market share is σti = 1/N and each firm announces a price of p∗ = pit (σti = 1/N ) in every period. The following proposition 24 describes the pure strategy pricing rule supporting a symmetric MPE with heterogeneous switching costs.20 Proposition 3. Given that a symmetric MPE exists, σ0i = 1/N for i = 1, . . . , N , ρ = α, M0 = 1, and Assumption 3, the steady state symmetric pure strategy MPE price of the problem defined by equation (3) is p∗ = N ((N − 1)2 − δ(1 − α)L) (N − 1)3 − (1 − α)W where L = N ln(N ) − N + 1 and W = N 2 − 1 − 2N ln(N ). The steady state symmetric price is not a function of switching costs anymore since switching costs are now treated as a random variable. The equilibrium price is now only a function of the number of firms, firm’s discount factor and the fraction of new consumers in the market. For N ≥ 2, both L and W are greater than zero. Similar to the previous analysis, when α = 1 there are only new consumers in the model. New consumers do not have any switching costs and the equilibrium price is the standard logit model price of N/(N − 1). As before, the price is decreasing in the discount factor where (1 − α)N L ∂p∗ =− < 0. ∂δ (N − 1)3 − (1 − α)W As firms become more patient, the investment strategy takes precedent and firms offer lower prices to build future market share. A change in the fraction of new consumers affects the equilibrium price in the following way: ∂p∗ N (N − 1)2 (δ(N − 1)L − W ) = . ∂α ((N − 1)3 − (1 − α)W )2 The sign of ∂p∗ ∂α is determined by the sign of δ(N −1)L−W which may be either positive or negative depending upon the relationship between N and δ. Identical to the model with homogeneous switching costs the sign of ∂p∗ ∂α is linked to the sign of the price difference when comparing the price 20 Similar to Proposition 1, Proposition 3 is based on the existence of a symmetric MPE. The intuition regarding existence is identical to that provided for Proposition 1. 25 with switching costs to the price without. The difference in price created by switching costs is ∆p∗ = p∗ − N (1 − α)N (W − δ(N − 1)L) = N −1 (N − 1)((N − 1)3 − (1 − α)W ) (15) and the sign of ∆p∗ is determined by the sign of W − δ(N − 1)L. Switching costs lead to lower prices when W < δ(N − 1)L and this is more likely for large values of δ and large values of N . When W < δ(N − 1)L, the price difference, ∆p∗ is negative and ∂p∗ ∂α > 0. This result leads to the following corollary. Corollary 4. With heterogeneous switching costs satisfying Assumption 3, ∆p∗ < 0 if and only if ∂p∗ /∂α > 0. Note that the price difference does not depend on α as is the case with homogeneous switching costs. In Section 4.2, N 0 was determined such that for N > N 0 switching costs led to lower prices. The analog in this section would be to solve W − δ(N − 1)L = 0 for N , but the analytical solution does not exist. Instead, solve for δ where δ̂ = N 2 − 1 − 2N ln(N ) W = (N − 1)L (N − 1)(N ln(N ) − N + 1) and if δ > δ̂ switching costs lead to lower prices. For the minimum number of firms allowed (two), δ̂(N = 2) = 0.5887. If δ > 0.5887 when there are two firms, equilibrium prices are lower with switching costs. Since it is more likely that equilibrium prices are lower with switching costs as N increases, if δ > 0.5887 equilibrium prices are always lower with switching costs regardless of the number of firms. If instead δ is less than 0.5887, there is some N 0 such that for N > N 0 switching costs lead to lower prices. Figure 2 illustrates the parameter space and indicates when prices are higher or lower with switching costs. Note that the case of N = 10 is only provided as a reference point. The results from this section are collected in the following proposition. Proposition 4. Assume α 6= 1 and consumer switching costs satisfy Assumption 3 then 1. if δ > 3−4 ln(2) 2 ln(2)−1 , switching costs always lead to lower prices (∆p∗ < 0) and 26 2. if δ < 3−4 ln(2) 2 ln(2)−1 , there exists some N 0 such that for N > N 0 ∆p∗ < 0. This section shows that the results contained in Proposition 2 are quite robust. When a continuous distribution of switching costs is incorporated into the model, equilibrium prices are lower with switching costs more than one might expect. 6 Conclusion This paper examines the trade-offs firms face when consumers have switching costs and firms cannot price discriminate. Each firm has two conflicting strategies in which they can offer a low price to invest in future market share or offer a high price to exploit the partial lock-in of their previous customers. The model developed in this paper combines the discrete choice logit model with switching costs in a dynamic setting. Competition takes place over an infinite horizon where overlapping generations of consumers have switching costs and any number of firms produce one variety of a differentiated product. The main results of this paper show under what conditions switching costs lead to lower or higher prices and under what conditions an increase in switching costs decreases the price. These results depend on the relationship between the number of firms, the level of consumer switching costs, and firms’ discount factor. Similar to dynamic duopoly studies in the previous literature, switching costs are pro-competitive when the level of switching costs is sufficiently small or firms’ discount factor is sufficiently large. Otherwise switching costs are anti-competitive. A main result of the paper is to show that with a sufficiently large number of firms switching costs are procompetitive, and with a sufficiently small number of firms switching costs are anti-competitive. The relevant threshold for the number of firms depends on firms’ discount rate and the level of switching costs where this threshold is increasing in the level of switching costs and decreasing in the discount rate. It is also shown that this result is robust to uniform and heterogeneous switching costs. 27 A A.1 Proofs and Derivations Proof of Lemma 1 Proof. Equation (1) defines the probability that consumer z chooses product i in period t. V i is either V i = y − pi or V i = y − pi − s0 depending on the consumer’s previous purchase. For a new consumer with no switching costs, PNi = exp(y − pi ) N X exp(y − pj ) (16) j=1 is the probability that a new consumer purchases from Firm i. An old consumer who purchased from Firm i in the previous period purchases from Firm i again with probability i = POL exp(y − pi ) N X j 0 , (17) i exp(y − p − s ) + exp(y − p ) j6=i There are old and switching consumers who purchased from Firm k 6= i in period t − 1 and switch to Firm i in period t with probability ki POS = exp(y − pi − s0 ) N X . (18) exp(y − pj − s0 ) + exp(y − pk ) j6=k Price symmetry within periods allows exp(y − pi ) terms to be factored out of all the numerator and denominators in the above probability expressions. Factoring out this term for Pni results in Pni = 1/N . The summation of exp(−s0 ) in the denominator of both POL and POS occurs N − 1 times. After substituting in 1 − s for exp(−s0 ), the denominator for both probabilities becomes (1 − s)(N − 1) + 1. The resulting probabilities are those listed in equation (4). A.2 Proof of Lemma 2 Proof. Equation (5) is obtained by substituting the symmetric probabilities from equation (4) into PN equation (2) and observing that k6=i σ k = 1 − σ i . Iteration of equation (5) results in equation (6). Market shares evolve according to equation (5) which can be rewritten as i σt+1 = α (1 − α)(1 − s) (1 − α)sσti + + . N (1 − s)(N − 1) + 1 (1 − s)(N − 1) + 1 Let A= α (1 − α)(1 − s) + N (1 − s)(N − 1) + 1 28 and B= (1 − α)s (1 − s)(N − 1) + 1 i so that σt+1 = A + Bσti . Iteration yields σti A A i Bt = + σ0 − 1−B 1−B where A/(1 − B) = 1/N . A.3 Proof of Proposition 1 Proof. Now suppose that all firms, except for Firm i, use a Markov pricing strategy of g(σ) ∈ P. Below I specify Firm i’s problem conditional on all other players using g(σ), or g for simplicity. When all other firms set their price to g, σ j = (1 − σ i )/(N − 1) and the state space of Firm i’s problem is reduced from σ ∈ [0, 1]N to σ i ∈ [0, 1]. The Bellman equation from the problem described in (3) reduces to i V (σti ) = max π i (pit , σti , g) + δV (σt+1 ) pit where π i (pit , σti , g) = pit Qit (pit , σti , g). Dispensing with the i superscripts and using the given conditions of ρ = α and M0 = 1, the first order condition of the Bellman with respect to pt is ∂Qt ∂V (σt+1 ) ∂Qt Qt + pt +δ = 0. (19) ∂pt ∂pt ∂σt+1 This first order condition reflects the fact that a marginal change in a firm’s price affects it’s current profits, but also changes its future market share affecting the continuation value. The Euler equation is ∂V (σt ) ∂Qt ∂Qt ∂V (σt+1 ) = pt + δ . (20) ∂σt ∂σt ∂σt ∂σt+1 Qt is the demand for Firm i in period t as defined by equation (2). When all other firms use a price of g Qt (pt , σt , g) = αPNt + (1 − α)σt POLt + (1 − α)(1 − σt )POSt where PNt , POLt , and POSt are functions of pt , g, s and N . Partially differentiate firm demand with respect to prices and shares to get ∂Qt = (1 − α)POLt − (1 − α)POSt ∂σt which is substituted back into equation (20), and ∂Qt ∂PNt ∂POLt ∂POSt =α + (1 − α)σt + (1 − α)(1 − σt ) ∂pt ∂pt ∂pt ∂pt 29 (21) (22) which is substituted back into equation (19). Equations (16) through (18) define the logit probabilities from the different types of consumers, and the partial derivatives of the logit probabilities with respect to the price is determined. The probability that a new consumer purchases from Firm i is PNt = exp(y − pt ) exp(−pt ) = N exp(−pt ) + (N − 1) exp(−g) X exp(y − pt ) + exp(y − g) j6=i as defined in equation (16). The partial derivative of PNt with respect to pt is ∂PNt exp(−pt ) (exp(−pt ))2 =− = PNt (PNt − 1) . + ∂pt exp(−pt ) + (N − 1) exp(−g) (exp(−pt ) + (N − 1) exp(−g))2 The other partial derivatives are determined in a similar fashion and are listed below. ∂POLt = POLt (POLt − 1) , ∂pt ∂POSt = POSt (POSt − 1) ∂pt (23) Any symmetric pure strategy MPE is characterized by price symmetry within each period. Now impose g = pt so that prices are symmetric within each period to determine the symmetric pure strategy price. Lemma 1 now applies and PN = 1 N POL = 1 (1 − s)(N − 1) + 1 POS = (1 − s) (1 − s)(N − 1) + 1 where the probabilities are now just functions of N and s. With price symmetry, the partial derivatives in equation (23) are just functions of N and s, and the partial derivatives in equation (21) are just functions of α, N , and s. Since σ0i = 1/N , ρ = α and M0 = 1, if g = pt , Lemma 2 is also applicable. These symmetric steady state conditions result in σ = Q = 1/N , and another condition of the symmetric steady state is that ∂V (σt ) ∂V (σt+1 ) = ∂σt ∂σt+1 for all t because in the steady state the marginal change in firm i’s continuation value over time is independent of the period. With the above conditions, equation (22) can now be expressed as (1 − α) (1 − α)(N − 1) ∂Q = αPN (PN − 1) + POL (POL − 1) + POS (POS − 1) ∂p N N where this partial is a function of α, N , and s. Equation (20) can be rewritten as ∂Q ∂σ p ∂V (σ) = ∂σ 1 − δ ∂Q ∂σ 30 and first order condition, equation (19), simplifies to the following. 1 ∂Q ∂V (σ) ∂Q +δ =0 +p N ∂p ∂p ∂σ After substitution, the first order condition is now a function of p, N , α, δ, and s and solving for p yields the symmetric steady state price listed in equation (7). Note that the method used to determine the price is unconventional. Typically equations (19) and (20), are used to find a Markov pricing rule, pit (σ t ), characterizing firm i’s price in period t as a function of all firms’ market shares in period t. Once the Markov pricing rule is determined, simplifying assumptions can be made to determine the symmetric steady state pricing rule. The approach taken here is to bypass the determination of the Markov pricing rule and just determine the symmetric steady state price. Doing so requires the existence of a Markov pricing strategy resulting in a symmetric steady state MPE in pure strategies for this problem. A.4 Proof of Lemma 3 Proof. When switching costs are drawn from an exponential distribution, the general choice probability is Z ∞ i eVzt i s s Pzt = PN V j exp(−εzt )dεzt . zt 0 j=1 e Similar to the proof of Lemma 1, Vzti is either Vzti = y − pit or Vzti = y − pit − εszt depending on the consumer’s previous purchase. For a new consumer with no switching costs, PNi t = exp(y − pit ) N X exp(y − pjt ) (24) j=1 is the probability that a new consumer purchases from Firm i. Since new consumers do not have switching costs, equation (24) is identical to equation (16). An old consumer who purchased from Firm i in the previous period purchases from Firm i again with probability Z ∞ exp(y − pit ) i POLt = exp(−εszt )dεszt . (25) N X 0 j exp(y − pt − εszt ) + exp(y − pit ) j6=i There are old and switching consumers who purchased from Firm k 6= i in period t − 1 and switch to Firm i in period t with probability Z ∞ exp(y − pit − εszt ) ki POSt = exp(−εszt )dεszt . (26) N X 0 j s k exp(y − pt − εzt ) + exp(y − pt ) j6=k Price symmetry within periods allows exp(y − pit ) terms to be factored out of all the numerator and denominators in the above probability expressions (note that exp(y − p − εszt ) = exp(y − p) exp(−εszt )). Factoring out this term for PNi t results in PN = 1/N . The summation of exp(−εszt ) 31 i i in the denominator of both POL and POS occurs N − 1 times. The choice probabilities for old t t consumers are now Z ∞ 1 POL = exp(−εs )dεs s) + 1 (N − 1) exp(−ε Z0 ∞ exp(−εs ) exp(−εs )dεs . POS = (N − 1) exp(−εs ) + 1 0 s Applying a change of variables where r = e−ε the choice probabilities for old consumers are transformed into Z 1 1 POL = dr 0 (N − 1)r + 1 Z 1 r POS = dr 0 (N − 1)r + 1 and integration yields the choice probabilities listed in equation 14. Firm i’s market demand/share in each period is defined by equation (2). The choice probabilities from equation (14) are substituted into equation (2) to yield Qit = (1 − α)((1 − σti )(N − 1 − ln(N )) + σti ln(N )(N − 1)) α + N (N − 1)2 for the symmetric demand. Rearrange the demand equation to get Qit = α (1 − α)(N − 1 − ln(N )) (1 − α)(N ln(N ) − N + 1) + + σti . 2 N (N − 1) (N − 1)2 The evolution of steady state market share is determined in Lemma 2 for a homogeneous level of switching costs. The technique in the proof to Lemma 2 is applied to the symmetric demand equation above letting A equal the constant term without σti , and letting B equal the multiplicative term on σti . Each firm’s market share evolves according to the following equation: σti 1 (1 − α)(N ln(N ) − N + 1) t 1 i + σ0 − . = N N (N − 1)2 Similar to the results of Lemma 2 if initial market shares are symmetric (σ0i = 1/N ) and prices are symmetric within periods, market shares are symmetric across periods, σti = 1/N . A.5 Proof of Proposition 3 Proof. The proof of Proposition 3 is directly related to the proof of Proposition 1. Suppose that all firms, except for Firm i, use a Markov pricing strategy of g(σ) ∈ P. The first order condition of the Bellman, equation (19), and the Euler equation, equation (20), from Proposition 1 still apply. The partials of firm demand from Proposition 1 are still relevant as well: equation (21) and equation (22). ∂P Similar to Proposition 1, ∂pNt t = PNt (PNt − 1) since switching costs do not influence the choice probability for new consumers. The remaining partial derivatives of choice probabilities, ∂POSt ∂pt , ∂POLt ∂pt and are determined from equation (25) and equation (26). The partial derivative of POLt in 32 equation (25) with respect to a change in pt is i ∂POL t = ∂pit Z ∞ !2 ey−pt N X 0 e −εs s Z dε − 0 s ey−g−ε + ey−pt j6=i ∞ ey−pt N X s e−ε dεs . s ey−g−ε + ey−pt j6=i Now impose price symmetry within periods (g = pt ) to factor out ey−pt terms, and allow for a s change of variables where r = e−ε to get ∂POL = ∂p Z 0 1 1 (N − 1)r + 1 2 Z dr − 0 1 1 1 ln(N ) dr = − . (N − 1)r + 1 N N −1 The remaining partial derivative is determined in a similar fashion and is listed below. ∂POS 2 + N ln(N ) − (N − 1)2 2 ln(N ) = − ∂p N (N − 1)2 (N − 1)3 As in the proof of Proposition 1, since σ0i = 1/N , ρ = α and M0 = 1, if g = pt we have a symmetric steady state of the model. These symmetric steady state conditions result in σ = Q = 1/N , and another condition of the symmetric steady state is that ∂V (σt ) ∂V (σt+1 ) = ∂σt ∂σt+1 for all t because in the steady state the marginal change in firm i’s continuation value over time is independent of the period. After substitution, the first order condition of the Bellman and the Euler equations are now just functions of p, N , α, and δ, and solving for p yields the symmetric steady state price. 33 Figure 2: The Parameter Space with Switching Cost Heterogeneity δ 1 Price is Lower with Switching Cost 0.5887 δ= N 2 −1−2N ln(N ) (N −1)(N ln(N )−N +1) 0.4195 Price is Higher with Switching Cost 0 2 10 34 N References Anderson, S. P., de Palma, A., & Thisse, J.-F. (1992). Discrete choice theory of product differentiation. Cambridge, MA: MIT press. Arie, G., & Grieco, P. L. E. (2014). Who pays for switching costs? Quantitative Marketing and Economics, 12 , 379 – 419. Armstrong, M. (2006). Recent developments in the economics of price discrimination. In Advances in economics and econometrics: Theory and applications: Ninth world congress of the econometrics society (Vol. II, pp. 97 – 141). Cambridge, UK: Cambridge University Press. Beggs, A., & Klemperer, P. (1992). Multi-period competition with switching costs. Econometrica, 60 (3), 651 – 666. Biglaiser, G., Crémer, J., & Dobos, G. (2013). The value of switching costs. The Journal of Economic Theory, 148 (3), 935 – 952. Cabral, L. (2014). Dynamic pricing in customer markets with switching costs. Working Paper . Retrieved from http://luiscabral.org//economics/workingpapers/scostsApril2014 .pdf Chen, Y. (2005). Oligopoly price discrimination by purchase history. In Pros and cons of price discrimination (p. 101-130). Stockholm: Swedish Competition Authority. Chen, Y., & Pearcy, J. (2010). Dynamic pricing: When to entice brand switching and when to reward consumer loyalty. The RAND Journal of Economics, 41 (4), 674 – 685. Doganoglu, T. (2010). Switching costs, experience goods and dynamic price competition. Quantitative Marketing and Economics, 8 (2), 167 – 205. Dubé, J.-P. H., Hitsch, G. J., & Rossi, P. E. (2009). Do switching costs make markets less competitive? Journal of Marketing Research, 46 (4), 435 – 445. d’Aspremont, C., Gabszewicz, J. J., & Thisse, J. F. (1979). On hotelling’s ”stability in competition”. Econometrica: Journal of the Econometric Society, 47 (5), 1145 – 1150. Fabra, N., & Garcı́a, A. (2015). Dynamic price competition with switching costs. Dynamic Games and Applications, May. Farrell, J., & Klemperer, P. (2007). Coordination and lock-in: Competition with switching costs and network effects. In M. Armstrong & R. H. Porter (Eds.), Handbook of industrial organization 35 (Vol. 3, pp. 1967 – 2072). Elsevier. Farrell, J., & Shapiro, C. (1988). Dynamic competition with switching costs. The RAND Journal of Economics, 19 (1), 123 – 137. Klemperer, P. (1995). Competition when consumers have switching costs: An overview with applications to industrial organization, macroeconomics, and international trade. The Review of Economic Studies, 62 (4), 515 – 539. Padilla, A. J. (1995). Revisiting dynamic duopoly with consumer switching costs. Journal of Economic Theory, 67 (2), 520 – 530. Rhodes, A. (2014). Re-examining the effects of switching costs. Economic Theory, 57 , 161 – 194. Rosenthal, R. W. (1980). A model in which an increase in the number of sellers leads to a higher price. Econometrica, 48 (6), 1575 – 1579. Shin, J., & Sudhir, K. (2008). Switching costs and market competitiveness: De-constructing the relationship. Working Paper . Retrieved from http://faculty.som.yale.edu/ksudhir/ papers/Switching%20Cost Shin%20and%20Sudhir%202008.pdf Somaini, P., & Einav, L. (2013). A model of market power in customer markets. The Journal of Industrial Economics, 61 (4), 938 – 986. Taylor, C. R. (2003). Supplier surfing: Competition and consumer behavior in subscription markets. The RAND Journal of Economics, 34 (2), 223 – 246. Train, K. E. (2003). Discrete choice methods with simulation. Cambridge, UK: Cambridge University Press. Viard, V. B. (2007). Do switching costs make markets more or less competitive? the case of 800-number portability. The RAND Journal of Economics, 38 (1), 146 – 163. Villas-Boas, J. M. (2015). A short survey on switching costs and dynamic competition. Working Paper . von Weizsäcker, C. C. (1984). The costs of substitution. Econometrica, 52 (5), 1085 – 1116. Wilson, C. M. (2012). Market frictions: A unified model of search costs and switching costs. European Economic Review , 56 (6), 1070 – 1086. 36
© Copyright 2024 Paperzz