Competition Between Firms with Multiple Locations in a Hotelling Game Author: Eric Metelka Advisor: Mark Witte Mathematical Methods in the Social Sciences Spring 2007 Metelka 1 Abstract The Hotelling model has been a standard in analyzing linear firm competition for over a decade. It has spawned numerous papers on the extrapolation of its concepts. Yet none of these have ever considered the effect of multiple agents controlling multiple locations. In this paper we explore the classic Hotelling model and some of its implications. Then we reflect on the differences between the classic model and the new model we are building, and present three different cases of how firms can align themselves along our new linear space. A careful analysis will then show how these cases create a general equilibrium of how firms place their locations to control the extremities of the linear space. Finally, we will then explore the implications this new equilibrium has on political theory, competing chains of stores, and transportation infrastructure. Metelka 2 Introduction Competition between firms is a cornerstone of economic analysis. It is the basis on how consumers are able to choose their goods from a variety of producers and likewise how producers determine what and how to produce to meet consumer demand. Any discussion of supply must include a breakdown of the suppliers and how they make their economic decisions to maximize their profit. A number of models have been formalized to this effect. For example, Bertrand and Cournot competition are models on how firms compete against each other in prices and quantities respectively. Similarly Harold Hotelling also produced a model of how firms compete but his model included dimensional space for the first time in the competitive analysis. Two firms compete and their prices and locations are determined endogenously over a uniform space. Just like the Betrand and Cournot models, the simplest way to analyze competition is looking at the effect of two firms competing. From there it is not hard to extrapolate these models to include N firms. However, all these models analyze firms competing individually while we know in reality that owners of firms usually control more than one store. We see chains and franchises in reality: owners of firms controlling more than one location of their store to sell their product. The crux of this paper will be to explain what happens when we apply the Hotelling Model to this competition – that is when there are two players who each control two locations in the simplest case. We will attempt to find the answer to the question of how firms make pricing and location decisions when they are chains. From there the effects of the new model we construct will be extrapolated into different arenas from firm positioning to the model’s effects on political analysis. Metelka 3 Hotelling conceived his model as a reaction to the instability in the Bertrand and Cournot models. He saw that in the Betrand there is an equilibrium, but if one player undercut his price by a minimal amount he would capture all the profit and thus create instability. What Hotelling sought was a stable equilibrium point and thus he added another dimension to the analysis of completion: distance. By creating a linear space over which firms compete, he brought completion from the first to the second dimension. Furthermore, it would not be much harder to extend the theory to the third dimension as well if we used a planar space instead of a line. It is not hard then to construct an equation that solves for the profits of each firm given their location and price. What Hotelling did next was find the stable equilibrium he was looking for that solved the instability problem. The key to the whole problem was finding the point of indifference. “The point of division between the regions served by the two entrepreneurs is determined by the condition that at this place it is a matter of indifference whether one buy from A or from B.” (Hotelling, 46) We can visualize the problem as thus: Where a and b are the distances from each end point and x and y are the differences from points A and B to the indifference point. To find the point of indifference, first he sets the demand equations for each product equal to one another under the constraint that our individual segments a, b, x, and y must equal the length of the whole segment l. Metelka 4 The derivation of Hotelling’s Model can be found in Appendix A. The final profit for both firms are: Hotelling found that profits are directly related to the cost of transportation and where each firm positions itself. Additionally, the greater the value of a for Player 1 and the greater the value of b for Player 2 results in greater profits. (Hotelling, 46, 47, 50) The framework that Hotelling created served as a platform for applications of the model to different areas of study along with the new found implications of the loosening of certain assumptions. Hotelling himself grasped some of the immediate consequences of his model. First, let us note again that profits increase as the cost of travel increases. Thus, firms will not want to encourage easy travel and invest in infrastructure for consumers. Rather they will put up barriers to trade. Hotelling’s model is actually a good argument for protectionist policies. (Hotelling, 50) In another example, Hotelling noted that if instead of determining prices endogenously they were fixed, such as in a socialist regime, firms now locate themselves at the center of the space. Each firm tries to control a greater portion of the space since they now only compete in the amount of the space they control. “As more and more sellers of the same commodity arise, the tendency is not to become distributed in the socially optimum manner but to cluster unduly.” (Hotelling, 53) This clustering effect at the Metelka 5 center is rather striking since it has many real world analogues. Say that instead of the space representing distance, it represented an axis of possible differentiations in the given product. We see that the firms cluster at the center, explaining why when we see two similar goods there may be slight differentiation such as an improvement to attract more buyers, but they are for the most part the same. (Hotelling, 54) Take another analogue – this time in the political landscape. Our space is now a political ideology. Both candidates in a given election will want to focus on the center and most of their platforms will be relatively similar with a few exceptions that still pander to their core base. (Hotelling, 54-55) This is actually a rather dismal conclusion as it leaves the consumer with very little choice. It purports that consumers want the middle option and competing firms will offer them alternatives that are barely different. Obviously while we do see a plethora of undifferentiated products being sold in reality, we so also see products that vary widely. So the model cannot truly account for every scenario. Loosening of further assumptions must be in order to get an even clearer picture of reality. Since we will be exploring a case of two players with two locations, it is only prudent to look at other examples of Hotelling theory with more than two locations. Eaton & Lipsey explored some of these cases. For example, they take the case of three firms and find that there is no pure equilibrium in this game. (Eaton, 47) Additionally they look at the case of what happens in general to equilibrium when a firm is added and define what they call the “principle of local clustering.” This is defined as “When a new firm enters a market, or when an existing firm relocates, there is a strong tendency for that firm to locate as close as possible to another firm.” (Eaton, 46) This is Metelka 6 done because firms can capture more market share by locating close to each other rather than from a distance where it leaves their opponent free to capture a whole region to themselves. It is better to split a local region in half by locating next to one another rather than leave it all to another player. Thus while there is no stable equilibrium in the three firm case, the principle of local clustering helps us define how firms locate when n > 3 and in fact helps explain what we find in the four firm case. The four firm case does indeed have a stable equilibrium and two firms locate at ¼ and ¾ each according to the principle of local clustering. Additionally, all firms receive an equal payoff of ¼ of the profit. (Huck, 232-233) Once we consider more than four firms, no more pure strategy equilibriums can be found. “With five the unique equilibrium configuration is asymmetric and implies unequal payoffs; and with six and more agents the equilibrium configurations cease to be unique.” (Huck, 231-232) Of note in these results is whether our new model will follow the same principles. While it may be doubtful that we receive the same result as the four firm case since all firms are independent of one another, it is likely that the principle of local clustering may still apply. This research that has come before lays a solid foundation of which we can build and understand if our model makes intuitive sense. Model Assumptions Unlike the classic Hotelling model or the four firm case, situating two firms with two different locations presents a more complicated scenario. Like the Hotelling model we have only two players. However, since they each have two locations, the outcome Metelka 7 will be decisively different except for one case which will be explored later. While similar to the four player model, the fact that the two firms have knowledge of where their other location is along the finite space always them access to more information than four individual players would have. The key is how this access to more information is reflected in the model we build. The solution is that the firms compete on profits. This is a fact that the classic Hotelling model and any of its extrapolations can easily agree on. Firms compete on profits and the player that wins is the one that makes the most money. Knowing this fact, we can now fold in the knowledge of their other location into each firm’s profit motive. In our model, instead of competing according to profit maximization at one location, the firms will strive to maximize the joint profit they obtain at both locations and they do so simultaneously by calculating their opponents reaction functions. Now that there is a clear idea about how the firms will compete against each other, it is prudent to explore the underlying assumptions that are inherent to this model. • Firms compete over a closed linear space. This space is one unit in length and its endpoints are 0 and 1. • The demand functions are strictly concave and decreasing. This invalidates the case of any plateaus in demand and also insures that the demand functions of the different players must intersect at some point in out linear space. • All demand functions are quadratic. This is done since they are a more realistic analogue of consumer demand than linear functions. (Tirole, 279) Metelka 8 • No more than one store can operate at any given point. The implication of this is that no more than one location can occupy a given space and there will be no clustering in this model. This is done to make the model more realistic since no more than one store can exist at one physical location. (Eaton, 48) • Locations exist at the points a, a+c, 1-b, and 1-b-d. Additionally, all variables are positive. This is done to impose linear order on the space. Since c must be positive, a+c must be to the right of point a. Likewise, 1-b-d must be to the left of point b. • a+c < 1-b-d to preserve order between a+c and 1-b-d much like in our last assumption. • . This insures that at the minimum a=b=0 and at the maximum a+b=1. • . This states that the maximum value of a+c is 1. • . This states that the maximum value of 1-b-d is 0. • Consumers choose firms based upon prices and transportation costs only (Eaton, 48) • Cost is denoted as co to differentiate it from the point c. This cost refers to both the fixed and variable costs of setting up a store at this location. We assume this costs are similar all over the linear space and thus keep them constant. Metelka 9 With these assumptions in place we now have a realistic model, a defined space which that model exists on, and we know how our demand functions act on that space. We can now start analyzing the model that had been built. The first thing that becomes apparent is how the two players arrange their locations over this space. In fact, there are three distinct combinations of how this can happen. By working through each case it can be determined which one produces the greatest joint profit for both players and that will be the case that would be emulated in real competition. In Case 1, one player controls the extremes of the space while the other controls a central “power core” over which his two locations are placed. This creates an interesting problem since we can explore which is more powerful to control: the center or the extremes. 0 a a+c 1-b-d a 1-b 1 Case 2 presents a scenario similar to the classic Hotelling model in which each player’s two locations exist on the same side. Metelka 10 0 a a+c 1-b-d a 1-b 1 Finally, in Case 3, the players mix their locations by alternating with each other over the space. 0 a a+c 1-b-d a 1-b 1 Case 1 Case 1 starts out much like the traditional Hotelling model. We want to find the indifference point between the two demand curves to find the demand at that point. Once we know demand, we can then maximize profit based on price and costs. In this model, prices are the same at each location. While prices can be determined endogenously, the decision to leave them constant for now is for the purpose of simplification. The interesting aspects of this model is how the players place Metelka 11 themselves in this unique game. The prices charged to maximize profits at these locations can be determined later. First, let us determine the demand at point a. To do this we find the indifference point between D1 and D2 by setting them equal. In our equations, s is consumer surplus (a constant), p is price, t is the cost of transportation for the consumer, and x is the point of indifference. We have found that the point of indifference is half way between a and a+c. We can only use this result to find the demand for D1 since it is bounded on the other end by the endpoint 0. To find D2 we would also need to find the indifference point between D2 and D3. However, since we are looking to maximize joint profit and D2 and D3 are controlled by the same player, it is easier to look at them both as the joint demand D1 + Metelka 12 D3. Just as we found it easy to solve for D1 since it is bounded on one end by an endpoint, it is just as easy to find D4 since it is bounded on one side by the endpoint 1. We do this by equating D3 and D4. Again, we obtain the logical answer that the indifference point is halfway between the two points 1-b and 1-b-d. Now we have found the solutions for D1 and D4 which are controlled by the same player. Maximizing the joint profit we find: Metelka 13 The joint profit D2+D3 must be the leftover demand on over the linear unit space or 1(D1+D4). Since for our purposes, p and co are fixed, the term (p - co) can be treated as a constant and we only need to focuses on the latter half of the profit equation. This still leaves us with a indiscriminant answer since we do not yet know the values of our variables a, b, c, and d. We do know though, that since player i chooses locations a and 1-b, he will choose these to maximize his profits. As such, he chooses the highest value of a and the lowest possible value of b. He would choose to locate a point 1 and would also choose to locate 1-b at point 1 with a value of 0 for b. Due to the ordering of our locations, both a+c and 1-b-d would also have to be at 1 with a value of 0 for both c Metelka 14 and d. We can see this more clearly on a graph that shows the relationship between location of the variables and profit. We then run the same analysis on how player j will choose his locations. Player j chooses the values of c and d. He chooses the lowest possible value for c since it is inversely proportional to profit. However, he wants to maximize the value of d. Due to the fact of what values were chose for a, b, and c, whatever value he chooses for d is negligible since all the locations are located at the endpoint 1 by definition. Metelka 15 Inputting these points we obtain: Player i clearly is in the better position since player j is not even making positive profits. It seems as if controlling the outsides of the space – or adhering to the extremes – is the more powerful situation to hold since player i is essentially putting player j out of business. However, as stated in the assumptions and as logic would dictate, no two Metelka 16 stores can occupy the same location, let alone four stores occupying the same space. Clustering is not allowed in this model by definition. Nevertheless, if the stores cannot occupy the same location, then they will choose to operate the minimum distance ε apart from one another. Since ε is a distance as small as physically possible to locate one store next to another, it is analogous to the case we just explored where the four stores are located at the same space. Therefore, our result is the same: it is better to control the extremes than try to maintain a hold on the center. Case 2 Because of how the players align their locations, there is only one point of indifference that occurs. The locations at the extremes of the space are negligible since the other location by the same player would capture that demand if the extreme location did not exist anyway. I posit that the result is the same as the classic Hotelling model with just two locations competing. Instead of using a+c and 1-b-d to find the different point, since they contain extra variables that are not needed in this case, we will use a and 1-b instead. It greatly simplifies the calculation. Metelka 17 This is a purely logical result. Player i controls the left side and thus as a increases, so does his profit since he controls more of the space. Likewise, Player j controls the right side and his profit increases when 1-b increases. The only aspect that is different from the classic Hotelling model is that we have fixed prices for both players. However, that fact should not be an impediment since when we solve that model the prices come out equal for both players. As such, the result here should be the same and players should position themselves both at ½ The other location each player controls Metelka 18 will be between each endpoint and ½. When we use these location to determine profit, we discover that they split the profit equally between them. Case 3 Case 3 is similar to Case 1. We start with calculating the demand for D1 and D4 the same way. Because both these demand curves are bounded on one side by an endpoint this is easy to calculate, and the result is the same except for in this instance one player controls D1 and the other controls D4. The challenge now is determine the derived demand for D2 and D3. We first consider using the same approach to find the demand as we did to find D1 and D4. The answer for this approach contains several unknowns – many of them quadratic. Instead, it would be easier to shrink the interval our demand curves exist on. Since it is known that that D1 covers from 0 to covers and D4 to 1, we know that the demand for D2 and D3 must exist over the rest of the interval. In effect, we can look at a small piece of the interval whose endpoints are and . Metelka 19 a+c 1-b-d Working over this smaller interval now makes it easier to solve for the demand. Just like in the previous cases, we know that between two locations with constant prices the indifference point will be the exact midpoint between the locations. Since we have a discrete interval over which only these demand curves exist, it is quite easy to find this midpoint. The length of the interval will be denoted as l. Now that we have the point of indifference, it is a simple matter to find the demand over the interval by just taking the distance between the endpoints and the indifference point. Metelka 20 a+c 1-b-d D2 and D3 are equal which is to be expected since all we did is find the distance between the endpoints to the midpoint. Now solving for the joint profit: Metelka 21 Player i controls the position of points a and d. Player j controls the position of points b and c. With this information, players then choose their points to maximize their profits. Player i chooses to maximize a and to minimize d while Player j choose to minimize both b and c. We can once again better visualize this in a graph. Metelka 22 This means that like in Case 1, players choose to cluster around the right endpoint with a value of 1 for point a and a value of 0 for the rest of the points. This outputs: In Case 3 both players make equal profits. Once again, this is not the ideal situation since the players have once again clustered at a single location. But the negligible distance of ε is so small that it does not effect the final profits. The insight that is gleamed from Case 3 is that alternating over the linear space will only serve to divide profits and gives neither player an advantage. It is also more profitable to mix like this than it is to split the continuum as in Case 2. However, it is still not as profitable as controlling the outside of the space in Case 1. General Equilibrium The results that we have found in the previous sections have allowed us to learn about which scenario is the most profitable way to locate firms along a linear space. For Metelka 23 example, Case 2 is strictly dominated by Case 3 since in Case 3 each firm can make twice as much profit as in Case 2. However, Case 1 is still the most profitable if the player maintains control of the extremes of the space. Case 1 gives one player a distinct advantage over the other by allowing him control over the extremes of the linear space. In this section, we will conduct a comparative analysis of which case is the most profitable as we move each location (a, b, c, and d) individually along the continuum while holding all else equal. We will also explore what happens when each player tries to compete to be the most profitable as in Case 1 and how this settles in equilibrium Up until now we have predicted how players will act under the profit maximizing rule. This has led us to find that our solutions have placed us at endpoints of our linear space. Now, let us step back a step back and analyze what happens when players do not act completely rational and do not end up at one of these endpoints. As a player decides where to place his firm, where is each individual location maximized? To see this we plot profit as a function of location of each variable along the unit space. For location a, the leftmost location, profit is maximized over the interval from [0, .35] for Case 3 Player j and from [.35, 1] for Case 1 player i. Metelka 24 It would be irrational to play any other scenario such as being Player j in Case 1 since Player j can make more profit by playing Case 3. We can then work through the graphs for the other locations to come to similar conclusions. In this case, players mix over Case 3 Player j form [0, .65] and Case 1 Player j from [.65, 1]. Metelka 25 Case 3 Player j [0, .65] and Case 1 Player i [.65, 0] are the mixing strategies for c. Playing Case 3 player j dominates any other placement of d. Given these facts, it is possible to not only choose a location from these graphs and know which player and case it is best to be, this information can also be used to undercut your opponent. These graphs give you the best options for reaction functions. Say you know that the other player is placing his first location .30. So his a location (as Metelka 26 long as you place both your firms right of his and you induce him to also place his other location to the right) will make the minimum amount of profit if you make him play Case 3. Since his location is the leftmost, he must be Player i. So to make him worst off you play Case 3 which makes significantly less profit at that location than Case 1. Knowing that you are now Player j in Case 3, you can choose c to be 0, or rather ε remembering our constraints. Now, whatever value Player i chooses for b (likely also ε), you will make maximum profit for any value of d since you are Player j playing Case 3. Just like the scenario that we just walked through, using these plots it is possible to choose strategies based on information and signals from your opponent to choose how to react. The graphs we’ve just examined give good strategies on how to choose locations – especially in the case where a player knows some information about his opponent. But how about the case where we know no information? walkthrough how players will act until they settle to an equilibrium. We can As we know extensively by this point, the most desirable scenario is to be Player i and play Case 1. Therefore, we must assume that both players will try to position their locations on the extremes of the interval in order to be this player. However, by placing both locations at the most extreme points allows the other player to capture the rest of the interval by being an ε closer to the center of both locations. But if the player moves too far in, the other player will take the outside position and grab the desirable outside spot. We must now find the optimal point between being too close to the endpoints of the interval and being too close to the center. In this case, we are dealing with the placement of two points: a and 1-b. The other player will play the same locations but with an ε increment either towards the Metelka 27 outside or the inside of the interval depending on which is more profitable. The player who captures the outside of the interval will control a total distance of a+b. The player that captures the inside will receive 1-a-b. We want to find the point where because this is where the player will be Player i in Case 1 and receive maximum profit by capturing as much of the space as possible. It is trivial to discover that While otherwise it is more desirable for the other player to play the outside. , this does not specify over what interval on our unit space this must be true. Variables a and b can take on any values as long as they sum to slightly less than ½. With this constraint in place, we can now choose the individual values of a and b. Just like in our solution for Case 1, profit is maximized when a is maximized and b is minimized. This means that Player i chooses a to be ½ and b to be 0. Player j chooses to locate an ε to the right of a and an ε to the left of 1-b meaning his choice of c and d are essentially 0. Knowing the values of all our variables we can now solve for profit: Metelka 28 Finally, we have a result for the general equilibrium. Previously we looked at specific cases or ways to maximize based on information. Bringing all this information together allowed us to come to the conclusion of the best case to play. Knowing that both players will try to play the best case allows us to analyze how the game settles in equilibrium and the result is certainly surprising. One player locates himself at the midpoint of the interval and at the far endpoint while his opponent captures the middle of that space. By doing so, the outside player makes positive profits and forces his opponent to take a loss and in the long run, to even possibly shut down. How does one player capture the outside in this case since they are both trying to locate at the same space? This is determined by variables outside our sphere of influence. Perhaps it is more costly for one firm to open at the outside location over the other firm. Perhaps both firms bid on the real estate that is on the outside of the interval and the firm that presents the better deal wins. Or perhaps one firm just gets there first. Either way, one player will come out on top at the expense of the other due to some exogenous variables outside of this model. The consequences of this result will be the focus of the rest of this paper. Metelka 29 Model Implications The political implications of the Hotelling model delve into campaign management and elections. Recall that the regular Hotelling model shows us where two candidates will both attempt to locate at the center of the political spectrum. The analogous concept in our revised model has similar results. In this case the two players are the two political parties: one liberal and one conservative. The two locations each party controls refers to any race in which there is a primary candidate and they also run with a deputy under them. The most relevant example is focusing on a presidential election with both a presidential and vice presidential candidate on the same ticket. Our new model can tell us where we should place both of them along the political spectrum in order to win the election. Before we begin this analysis, it must be noted that in most elections the ideology of the vice presidential candidate does not carry as much weight as the ideology of the president. Nevertheless, in our analysis we will act as if voters care equally about the political views of both the president and vice president on the ticket. We start with the general result we found for our two player two firm case. This says that both candidates choose to locate at ½ and 1 respectively. However, politics is decidedly different from our original model in that there are two ends of the spectrum and people are generally not willing to cross to the other side. A liberal is not likely to vote for a candidate who runs on a conservative ticket. Furthermore, it is just unrealistic and we do not see this occur in reality. Therefore, if both candidates ran at ½ and 1, there would be an outcry from voters. Instead, a liberal candidate would be better off running at 0 and ½ to capture the voters of his ideology. Note that he still maintains Metelka 30 control over ½ the length of the space which was crucial in finding our equilibrium. However, we are no longer playing a Case 1 game and instead have moved to a Case 3. It is a Case 3 because Case 3 dominates Case 2 so that’s what both parties will play. Therefore, the liberals will locate at 0 and ½+ε and the conservatives will locate at 1 and ½-ε. The political spectrum will look like this: liberal conservative The conventional wisdom behind what we observe is that liberals are going to want a liberal candidate – one who ascribes to their views. Likewise, conservatives want to vote for a conservative. That is why there is one candidate located at each extreme for each party. However, the reason candidates also locate around the center, as in the classic Hotelling model, is to capture moderate voters of the opposing political view. The extremist of either party are going to vote for their own candidate regardless of where they are along the political spectrum as long as it still reflects their ideology. Moderates, on the other hand, are much more fickle and often undecided. By locating at the ½ point, it is possible to garner the votes of these voters by not appearing to be too extreme in either direction. The interesting question that we are left with is where do the president and vice president locate? Is the president the one at the extreme or the one at the center? And Metelka 31 how does this play into primary party voting? Let us tackle the last question first. In primary voting, only the party’s registered voters vote for who they want to be candidate. Therefore, candidates have to appeal to the parties ideology and be less moderate and more extreme. This would seem to favor the extremist to be the ones who are president. However, an election is successfully won by not churning out the party base on voting day, but rather capturing those undecided voters at the center from your opponent. New models of politics which take into account this kind of behavioral view move away from the old models of capturing median voters. Specifically, the Erikson-Romero model finds that “…both candidates are motivated to appeal to independent voters, as opposed to campaigning for the support of committed partisans – this because strong partisans can be to some extent ‘taken for granted’…” (Adams, 298) The president, the figure most in the public eye, has to be moderate in order to do this. This also reflects the fact that after primaries, candidates try to make themselves look more moderate. However, to still carry their base, the presidential candidate choose a vice presidential running mate who complements them and capture voters they themselves would not carry. The vice president therefore captures the base and attempts to keep the voters the president might alienate by appearing more moderate. Consequently, according to this model of political theory, it is best for a president to be moderate with an extremist running mate in order to win an election. It is interesting to also note that while the position of the vice president is important, if we think about the fact that the position of the president is more important than the vice president, we still essentially see the same theory that results from the classic Hotelling model. Metelka 32 Presidential candidates locate at the center of the spectrum to capture the undecided moderates in order to win the election. Let us now turn our attention to another type of competition we see with the model we built: competition between chains. This is how firms such as McDonald’s and Burger King compete with each other. We once again start by looking at the general case we solved and see that the firms position themselves at ½ and 1. But now we are going to relax one of the initial assumptions. In this scenario instead of a finite space, these two firms compete over an extended infinite linear space with an infinite number of locations. Let us take a step back to see how we arrive at this scenario. The firms are located at ½ and 1 but what if they want to create a new location as they grow due to profits? They create a new location, also a half length away to the right. Essentially, we are just extending the space to the right half a length and now viewing a new space from ½ to 1½. We continue this ad infinitum. The same principles still apply. The firms still battle each other to control the extremes of the space. This means that whenever one firm places a location, the other will try to capture the outside of it and we will see the two firms both locate at the same point. The principle of local clustering holds in our model. The question now is how far apart these battleground locations of the firms competing are going to be from one another. Due to the lengthening of the linear space, we assumed they were a half length apart from one another. Since we also know that both firms try to control half the space or otherwise they lose the extreme position and market share, this assumption holds. What is a half length though in the real world? It could be half a mile or half a kilometer. Furthermore, different types of firms will have different standards for what a half length is. Department stores are likely Metelka 33 to have longer definitions of a half length than coffee shops. The point is that once a standard is established of how far apart two locations are, this standard distance between any additional locations holds. Now our space looks like this: 0 ½ 1 1½ 2 ½ This explains why we see gas stations across the street from one another or a Coke machine next to a Pepsi machine. Firms are best off when they place their product next to their competitor’s product for the consumer to choose the best one based on the marginal differences in products. By placing their locations anywhere else, they leave more space for their competitor to capture market share. It is also true that if one firm creates a new location to capture what they believe to be some market that was previously not captured, that their competitor can see that and also create a new location to compete for that market. This holds true for entrance into new markets such as new cities and new countries. Furthermore, this once again holds true to the classic Hotelling model where we saw that it is best to locate next to your competitor. The final implication of our model has to do with traveling costs. Our general result reinforces the Hotelling model in that it creates a disincentive to improve transportation infrastructure. As we have seen in this model, profits increase as the cost of travel increases. Firms also have no incentive to see transportation improved so that customers could potentially have access to their competitors goods easier. As access and cost to transportation becomes easier, positioning along the space Metelka 34 becomes less influential. Firms want to control the extremes of the space for a reason – it allows them greater market share. In fact, “buyers’ locations are such that nonminimal (including maximal) transportation costs are incurred in order for competition between sellers to be kept as fierce as possible” (Camacho-Cuena, 91) Additionally, this means that those consumers who reside on or close to the firms profit greatly since they do not incur these great transportation costs. Some of this may be due to consumers smartly positioning themselves next to desirable locations and some may be luckily free riding. (Camacho Cuena, 96) However, this accounts for high rent prices in desirable locations in cities. Due to all this, businesses will not play an active role in governmental policy to increase transportation infrastructure. Corporate lobbies have a major influence on public policy so the disincentive to increase the ease of transportation is somewhat unsettling as it decreases overall societal welfare. Conclusion Using Harold Hotelling’s classic model, we were able to build on his theories and create a better approximation of how firms compete in reality. For after all, while we do see single firms compete against one another, in today’s world of multinational corporations firms must compete in a different form. They must consider that they themselves control multiple locations of their corporation as do their competitors. This led us to consider three cases of how firms could align themselves on a linear unit space which were dubbed Case 1, Case 2, and Case 3. We then found that Case 1, where one player controls the extremes of the space was the most profitable to that Metelka 35 outside player. Knowing that both players could rationalize that Case 1 was the best outcome, we analyzed what happens when both players try to play Case 1 and discover that the firms would end up positioning themselves at the points ½ and 1. We also looked at how we could use the reaction functions of the cases we analyzed to find the best positioning of locations under certain restrictions as well. Using the new general equilibrium we could explore the effects of our model on other institutions. As an analysis of political theory, we find that candidates in a presidential election with a vice presidential running mate should locate at the center of the space while their vice president should cater to the extremes of their political party. For competing chains, the principle of local clustering holds and all firms should endeavor to locate next to their competitor in order to capture the most profit. With regards to transportation infrastructure, we find that there is no incentive for firms to put pressure on governments to improve ease of transportation as profits rise in direct correlation with transportation costs. From all of this we can conclude that while it might not be the most logical thing to locate next to a competitor, it is what we see in reality and the most profitable stratagem. Furthermore, it is what is reflected in reality. In the future, it would be prudent to look at how this model, since it assumes the simultaneous actions of firms, compares to that where firms take terms placing locations against each other over time. It would also be useful to look at cases of more than two firms and more than two locations. This model could also be applied to different spaces besides the linear one which we used which are better approximations of how cities arrange themselves in reality. All in all, this is a great primer to use in our growing world of hyper-chains and mega corporations that compete against one another and perhaps even serves as a Metelka 36 glimpse into the future of how we can expect firms to locate themselves in our ever increasing global world where corporations can spread their arms to any corner of the world. Metelka 37 Appendix A Setting the demand equations for each firm equal we obtain: The length of the total segment is l: By algebra: Firms now maximize profit by finding the optimal price to charge: Metelka 38 (Hotelling, 46, 47, 50) Metelka 39 Works Cited Adams, James. "Multicadidate Equilibrium in American Elections." Public Choice, 103 (2000): 297-325. Camacho-Cuena, Eva, et al. "Buyer-Seller Interaction in Experimental Spatial Markets." Regional Science and Urban Economics, 35 (2005): 89-108. Eaton, B. Curtis and Richard G. Lipsey. "The Principle of Minimum Differentiation: Some New Developments in the Theory of Spatial Competition." The Review of Economic Studies, 42 (1975): 27-49. Hotelling, Harold. "Stability in Competition." The Economic Journal, 39 (1929): 41-57. Huck, Steffen, Wieland Muller and Nicolaas J Vriend. "The East End, the West End, and King's Cross: On Clustering in the Four-Player Hotelling Game." Economic Inquiry (2002): 231-240. Tirole, Jean. The Theory of Industrial Organization. Cambridge: MIT Press, 1988.
© Copyright 2026 Paperzz