Hotelling's Linear City Model

Hotelling's linear city model is the classic model of spatial competition and horizontal product differentiation in industrial organization, game theory, and business and economics theory, introduced by American mathematical statistician and economist Harold Hotelling in his landmark 1929 Economic Journal paper "Stability in Competition". Addressing Joseph Bertrand's (1883) critique of Cournot competition—namely, that under homogeneous goods even a slight price cut by one duopolist captures 100% of the market and drives prices down to marginal cost—Hotelling placed consumers uniformly along a continuous one-dimensional interval [0, 1] ("Main Street" or a spectrum of product attributes such as sweetness, departure times, or ideological positions) where consumers incur transportation or mismatch disutility proportional to distance from a vendor. The model demonstrated that spatial or characteristic differentiation softens discontinuous price switches and gives rise to two competing strategic forces in location-then-price games: the market-share (demand) effect, which pulls firms toward the center of the market, and the strategic (price-competition) effect, which pushes firms apart to relax Bertrand price warfare.

Consumer Utility and Demand Derivation

Consider a linear city (or product characteristic space) represented by the unit interval [0, 1], along which a unit mass of consumers is distributed with uniform density f(x) = 1. Two competing firms, Firm 1 and Firm 2, produce an otherwise physically identical good at constant unit marginal cost c ≥ 0 and locate at positions a and 1 − b along the interval, where a ≥ 0, b ≥ 0, and a + b ≤ 1 (so Firm 1 lies to the left of or at the same point as Firm 2: x₁ = a ≤ 1 − b = x₂).

Each consumer located at coordinate x ∈ [0, 1] demands at most one unit of the good (inelastic unit demand up to a high reservation valuation v ≫ c) and seeks to minimize the full delivered price—the sum of the mill price p_i charged by Firm i and the transportation (or ideal-variety mismatch) cost T(|x − x_i|) required to travel from x to firm location x_i.

In Hotelling's (1929) original specification, transportation cost is a linear function of Euclidean distance with rate t > 0 per unit distance: T(d) = t · d. Assuming neither firm prices the other out of the market, the indifferent consumer x̂ ∈ [a, 1 − b]—who obtains identical net utility v − p₁ − t(x̂ − a) = v − p₂ − t(1 − b − x̂) from purchasing at either store—is located at:

All consumers to the left of x̂ (x ∈ [0, x̂]) purchase from Firm 1, yielding demand D₁(p₁, p₂) = x̂, while all consumers to the right of x̂ (x ∈ [x̂, 1]) purchase from Firm 2, yielding demand D₂(p₁, p₂) = 1 − x̂.

Fixed Prices and the Principle of Minimum Differentiation

When mill prices are fixed exogenously at p₁ = p₂ = p > c (as in regulated retail markets, broadcast media programming, or two-party electoral competition where candidates cannot pay voters directly), the indifferent consumer lies at the midpoint between the two firms: x̂ = (a + 1 − b) / 2. Firm 1's profit π₁ = (p − c) · (a + 1 − b) / 2 is monotonically increasing in its location a for any a < 1 − b, because moving closer to Firm 2 captures additional hinterland customers between a and 1 − b while retaining the entire captive backyard [0, a]. Symmetrically, Firm 2 increases its market share by increasing b (moving left toward Firm 1).

Consequently, the unique pure-strategy Nash equilibrium of the pure-location game occurs when both firms agglomerate at the exact geographical center of the city:

Hotelling termed this clustering tendency the Principle of Minimum Differentiation, observing that rival ice-cream vendors on a beach, competing department stores, and Republican and Democratic political platforms routinely converge toward the median consumer or voter (anticipating the Downsian Median Voter Theorem formulated by Anthony Downs in 1957). Observe that this agglomeration violates Pareto efficiency: a benevolent social planner minimizing aggregate societal transport costs ∫₀¹ t min(|x − x₁|, |x − x₂|) dx would locate the two stores at the first and third quartiles (x₁^soc = 1/4 and x₂^soc = 3/4), cutting total consumer travel distance in half.

The d'Aspremont–Gabszewicz–Thisse Critique and Quadratic Transport Costs

Fifty years after Hotelling's paper, Claude d'Aspremont, Jean Jaskold Gabszewicz, and Jacques-François Thisse (1979) proved a landmark correction: when firms first choose locations (a, 1 − b) and subsequently compete in prices (p₁, p₂), no pure-strategy price equilibrium exists under linear transport costs whenever the two firms locate too close to one another. Specifically, if a consumer at Firm 2's doorstep (x = 1 − b) faces a linear travel cost difference t(1 − a − b) to visit Firm 1, then whenever p₁ − p₂ drops below −t(1 − a − b), every consumer in the interval [1 − b, 1] simultaneously switches to Firm 1, creating a discontinuous upward jump in Firm 1's demand and profit function that destroys the existence of a pure-strategy Bertrand–Nash price equilibrium unless the firms are sufficiently far apart.

To restore smooth, strictly concave profit functions across all location pairs, d'Aspremont, Gabszewicz, and Thisse (1979) replaced linear transport costs with quadratic transportation costs:

Under quadratic transport costs, the indifferent consumer satisfies p₁ + t(x̂ − a)² = p₂ + t(1 − b − x̂)², which—because the x̂² terms cancel on both sides—yields a smooth linear demand boundary valid for all (a, 1 − b):

Solving the second-stage first-order conditions ∂π_i / ∂p_i = 0 yields unique subgame-perfect equilibrium prices and profits:

Substituting these prices into Stage 1 profit π₁(a, b) = (p₁* − c)*x̂** and differentiating with respect to Firm 1's location a reveals that the negative strategic price-competition effect dominates the positive market-share effect everywhere on [0, 1/2]:

Consequently, under quadratic transport costs with endogenous price competition, firms choose the endpoints of the interval (a* = 0 and 1 − b* = 1, so b* = 0), establishing the Principle of Maximum Differentiation: firms differentiate their products as much as physically possible to soften price competition and charge the equilibrium price p₁* = p₂* = c + t.

To analyze free entry of N ≥ 2 symmetric firms without the asymmetric hinterland "backyard" effects of line endpoints, Steven C. Salop (1979) wrapped the unit interval [0, 1] into a circle of circumference 1. When N firms locate equidistantly at spacing 1/N around the circle and each incurs a fixed setup cost F > 0, each firm competes directly only with its two immediate neighbors at x_i ± 1/N. Solving the symmetric Bertrand–Nash price equilibrium yields p* = c + t/N, and imposing zero net profit (π_i = (p* − c)/N − F = t/N² − F = 0) yields the free-entry number of firms N* = √(t/F), which is exactly twice the socially optimal number of firms N^soc = (1/2)√(t/F) because entering firms ignore the business-stealing externality inflicted on incumbents. Furthermore, when firms can observe individual consumer addresses x and deliver goods directly—known as spatial price discrimination (Thisse and Vives, 1988)—firms charge lower delivered prices to contested middle-ground consumers near x̂ = 1/2 than to captive consumers near their own storefronts, intensifying price competition across the entire market.

How does horizontal differentiation in Hotelling's model differ from vertical differentiation?

In horizontal product differentiation (Hotelling's linear city and Salop's circular city), products differ in attributes where consumers disagree on the ideal variety: at identical prices (p₁ = p₂), both Firm 1 and Firm 2 enjoy positive market shares because consumers located near a prefer Good 1 while consumers near 1 − b prefer Good 2. In vertical product differentiation (Jaskold Gabszewicz and Thisse, 1979; Shaked and Sutton, 1982), products differ in objective quality: at identical prices, 100% of consumers agree that the higher-quality product is superior, so the lower-quality firm can survive only by charging a lower price.

Why do political candidates sometimes diverge from the median voter despite Hotelling's Principle of Minimum Differentiation?

While Hotelling's pure-location model predicts convergence to the median voter at x = 1/2 when there are two office-motivated candidates, divergence toward the ideological flanks arises whenever candidates must first win party primary elections, when third-party entry threatens to capture extreme voters, when voters abstain from turning out if their preferred candidate locates too far from their ideal point ("alienation"), or when candidates possess policy-motivated ideological preferences alongside a desire to win office (Wittman, 1983).

1. Hotelling, H. (1929). "Stability in Competition." The Economic Journal, 39(153), 41–57. https://doi.org/10.2307/2224214
2. d'Aspremont, C., Gabszewicz, J. J., & Thisse, J.-F. (1979). "On Hotelling's 'Stability in Competition.'" Econometrica, 47(5), 1145–1150. https://doi.org/10.2307/1911955
3. Salop, S. C. (1979). "Monopolistic Competition with Outside Goods." The Bell Journal of Economics, 10(1), 141–156. https://doi.org/10.2307/3003323
4. Downs, A. (1957). An Economic Theory of Democracy. New York: Harper & Brothers. https://doi.org/10.1086/258056
5. Thisse, J.-F., & Vives, X. (1988). "On the Strategic Choice of Spatial Price Policy." American Economic Review, 78(1), 122–137. https://www.jstor.org/stable/1814702