Among the available options: horizontal separation (HS) and vertical separation (VS)

Among the available options: horizontal separation (HS) and vertical separation (VS) Horizontal separation (or systems competition per Pittman 2011) Inter-system competition between vertically integrated rail systems Vertical separation (or components competition per Pittman 2011) Above-the-rail competition by transport operators Below-the-rail infrastructure management by independent (and non-integrated) regulated monopoly network firm 2

Horizontal separation Potential advantages: Economies of scope between transportation operations and infrastructure management, enhancing technical efficiency and productivity (Ivaldi and McCullough, 2008; Growitsch and Wetzel, 2009; Sánchez et al., 2008; Pittman, 2011) Beneficial for attracting private investment capital (Pittman, 2011) Potential drawbacks: Compromises economies of system size (Bitzan, 2003; Pittman, 2011) May leave rail customers with no competition (but even when parallel competition is limited under HS, source competition may still be possible) Vertical separation Potential advantages: Explicitly fosters competition among independent transport providers for the part of the vertical chain that is intrinsically competitive Potential drawbacks: Elevates transaction costs of negotiating and enforcing contracts between the network firm and transport operators (Nash 2008) May result in higher operating costs, either due to foregone economies of scope (Bitzan 2003) or to misalignment of incentives between the network firm and transport operators (Nash et. al. 2014) 3

Research question: How do HS and VS differ in terms of their effects on three metrics of interest to economists and policymakers: network quality, consumer surplus, and social welfare? Research approach: Tease out implications of each structure presuming maximizing behavior by relevant parties and equilibrium outcomes Multi-stage model in which profit-maximizing firms first choose network quality, then choose transport prices (and under VS, regulator chooses access tariff prior to choice of network quality by network firm) Characterize equilibrium behavior and its implications for network quality, consumer surplus, and social welfare: Analytically when possible Numerically over wide swath of parameter space when unambiguous analytical results are not possible Our (hoped-for) contributions: Does industrial organization-style applied theory have anything useful to add in developing intuitions about how well HS and VS perform on key metrics of interest? First paper to comprehensively explores impact of HS and VS on both network quality and consumer, social welfare 4

Social welfare Vertical separation Two competing transport operators, i = 1,2 operate on infrastructure, owned by monopoly network firm Order of moves: Regulator chooses fixed and variable access charges, F, c Horizontal separation Two vertically integrated rail systems, i = 1,2 Each systems owns its own rail infrastructure and operates trains over that infrastructure Order of moves: Network profit Network firm chooses network quality, q Demand functions for transport services Transport operators pay c,f Each system chooses network quality, q i, i = 1,2 Demand functions for transport services System profit Transport profit Each operator chooses transport price, P i, i = 1,2 Equilibrium price, profits, consumer surplus, social welfare consumers pay P i, experience q consumers pay P i, experience q i Each system chooses transport price, P i, i = 1,2 Equilibrium price, profits, consumer surplus, social welfare System profit 5

Demand functions for transport services derived from quadratic net benefit function as in de Rus and Socorro (2014): U k X 1, X 2, q 1, q 2 i 2 P i X i = i 2 a + q i X i 1 2 1+b k X 1 2 + 2b k X 1 X 2 + X 2 2 i 2 P i X i, k = V, H X D 1k (P 1, P 2, q 1, q 2 ) = a 1 b k +q 1 b k q 2 P 1 +b k P 2, X D 2k (P 1, P 2, q 1, q 2 ) = a 1 b k +q 2 b k q 1 P 2 +b k P 1, k = V, H 1 b k Transport operators under VS seen as providing substitute services, with b V [0,1) denoting degree of perceived differentiation: b V 1 offers seen as perfect substitutes, consumer choice depends entirely on price intense price competition b V = 0 offers seen as completely differentiated no price competition; each operator is a little monopoly Systems under HS could be competitors, b H 0,1 or complementors b H ( 1,0) Transport cost function, firm i, structure k: C ik T X ik = γ k X ik, k = V, H, i = 1,2 (transport costs symmetric) 1 b k Network cost functions: HS system: C ih N X ih, q ih 2 = η H X ih + f + λ 2 q ih 2, i = 1,2 (network costs are symmetric) VS network firm: C N V X 1V, X 2V, q 2 V 2 = η V i=1 X iv + 2 1 δ [f + λ q 2 V 2 ], δ 0, 1 2 Network quality is a fixed cost activity f = 0 for simplicity, δ > 0 economies of network size 6

Suppose state-owned national rail system served four urban areas A, B, C, and D and is restructured into two horizontally separated systems: RR 1 (serving ABC) and RR 2 (serving BCD) RR1 RR2 B A Creates partially captive shippers at points A and D: those at A must ship over RR1 and those at D must ship over RR2 D C Presence of partially captive shippers creates possibility that RR 1 and RR 2 are complementors: Traffic originating at A bound for D could ship on RR 1 from A to B and potentially transfer to RR 2 at B for the remainder of journey If RR 2 lowered its rates (holding RR 1 's rates fixed), it would induce a higher quantity of this traffic, which would stimulate demand for RR 1 's service between A and B, while reducing demand for RR 1 's traffic between B and C Depending on overall traffic patterns, it is possible that the cross-price elasticity of overall demand for RR 1 with respect to the price of RR 2 could be negative Implication: under HS the competition intensity parameter b H in our model can be influenced partly by how policy makers choose to restructure the system 7

HS: Step 1: Solve for equilibrium in transport pricing subgame, given arbitrary network qualities q 1H, q 2H Step 2: Solve for (symmetric) equilibrium network quality, q H with firms anticipating impact of quality on pricing subgame Implies: symmetric equilibrium transport price and quantity per firm, P H, X H Consumer surplus CS H found by substituting into representative consumer utility function Social welfare SW H is sum of consumer surplus and profit of each integrated system VS: Step 1: Solve for (symmetric) equilibrium in transport pricing subgame, given arbitrary network quality q V and variable access charge, c Step 2: Solve for profit-maximizing network quality, given arbitrary variable access charge, c, with network firm anticipating effect of network quality on pricing subgame Step 3: Solve for social welfare-maximizing variable access charge, c V, with regulator anticipating impact of access charge on network firm s quality choice and transport firm s equilibrium prices Implies: equilibrium network quality q V and symmetric equilibrium transport price and quantity per firm, P V, X V Consumer surplus CS H found by substituting into representative consumer utility function Social welfare SW H is sum of consumer surplus and profit of each integrated system 8

Horizontal separation Vertical separation Price P H = 1 b H 2 b H a + q H + 1 2 b H [γ H + η H ] P V = 1 b V 2 b V a + q V + 1 2 b V [γ V + c V ] Quantity (per firm) Consumer surplus X H = a + q H [γ H + η H ] X 2 b V = a + q V γ V + c V H 2 b V CS H = U X H, X H, q H, q H 2P H X H = X H 2 CS V = U X V, X V, q V 2P V X V = X V 2 Social welfare SW H = CS H + 2 P H γ H η H X H λ q H 2 = 3 2b H CS H λ q H 2 SW V = CS V + 2 P V γ V c V X V + 2 c V η V X V (1 δ)λ q V 2 = 3 2b V CS V + 2(c V η V )X V (1 δ)λ q V 2 9

Horizontal separation Vertical separation Network quality q H = φ(b H )(a γ H η H ) λ φ(b H ) φ(b H )(a γ H η H ) if b λ φ(b H ) H > 0 and λ > 1 + q if bh > 0 and λ 1 + if b H 0 b H 2 + b H 1 b H φ(b H ) b H 2 + b H 1 b H φ b H, q V = 0 if c V η V 2 c V η V 2(1 δ)λ(2 b V ) if c V η V, c q if c V = c, where: φ(b H ) = (2+b 2 H)(2 b H ) b H, and q is maximum feasible network quality (2 b H )²(2+b H ) Variable access charge c H = η H c V = η V 1 b V a γ V η V if A(b V, λ) 0 η V 1 b V a γ V η V if A b V, λ 0, ψ b V, λ and Ω b 2 V, λ < 1 η V + (ψ(b V, λ)a(b V, λ)(a γ V η V ) ψ b V, λ 2A(b V, λ) c if A b V, λ where: ψ b V, λ = 2 1 δ λ 2 b V > 0 A b V, λ = 2 3 2b V (1 b V )ψ b V, λ 2 3 2b V + ψ b V, λ 0, ψ b V, λ 2 < 1 and Ω b V, λ 1 if A b V, λ ψ b V, λ 2 10

Equilibrium transport prices and quantities differ because network quality under two system differ, i.e., q H q V, due to: Differences in structural incentives : under HS network quality chosen by vertically integrated system to maximize profits; under VS network quality chosen by network firm to maximize network profit Differences in parameter values, i.e., b H may not equal b V, γ H may not equal γ V, η H may not equal η V Equilibrium transport price and quantities also differ because of possible deviation of variable access charge c V from marginal access cost η V Equilibrium network qualities differ due to possible presence of economies of network size, δ (0, 1 2 ] All equilibrium values may differ due to differences in parameter values, i.e., b H may not equal b V, γ H may not equal γ V, η H may not equal η V 11

Isolate structural incentives for network quality under HS and VS: equate all model parameters across the two cases, i.e., set b V = b H = b γ V = γ H = γ η V = η H = η and identify how the difference in the responsibilities for quality choice across the two structures affects incentives to invest in quality Analytical results Computational analysis across wide range of parameter space 12

Appropriability effect Marginal benefit of quality effect: Under HS quality maximizes network profit, but under VS it maximizes just network profit (akin to double marginalization in vertical models) Horizontal coordination on quality effect: Under HS each system chooses own quality noncooperatively, but under VS common network quality chosen by network firm (it s as if two separate networks colluded on quality) $ per unit of quality Marginal benefit of quality & horizontal coordination on quality effects MB H q = 2 2 b μ b [P Hi q, q H MC q H = 2λq γ η], μ b > 1 MC q V = 2(1 δ)λq Appropriability effect: Under HS systems appropriate gains from network quality via system profit margin P H γ η > 0, but under VS systems appropriate gains through network profit margin c V η, which could be smaller or bigger than system profit margin depending on regulator s action MB q V = 2 2 b c η Economies of network size effect: If δ > 0, marginal cost of network quality under VS is less than the marginal cost of increasing network quality on both systems simultaneously under HS q V (c) q H q 13

Marginal social welfare with respect to access charge: dsw(c) dc = 2 i=1 P V (c) [γ V + η V ] dx V dc + U 2(1 δ)q q V (c) dq V V dc Regulator balances two distortions: Downward quantity distortion in downstream market power in transport market P V (c) γ V + η V > 0 (if c η V ) Downward quality distortion when variable access charge = marginal cost: Since dq V dc but dx V > 0, higher access charge reduces downward quality distortion, dc = 1 2 b V 1 (1 δ)λ(2 b V ) U q V 2(1 δ)q V (c) c=η V > 0 1 could be > 0 or < 0, so higher access charge could reduce the quantity distortion or worsen it, creating a quality-quantity tradeoff for regulator 14

No quality-quantity tradeoff c = η V X V X Equilibrium combinations of q, X induced by varying c If dx V dc > 0, then regulator can reduce both distortions by increasing variable access charge above marginal cost η V c = η V Quality-quantity tradeoff X V X If dx V dc < 0, then regulator forced into a trade-off: increasing access charge reduces quality distortion but worsens downstream quantity distortion; Regulator s optimum might occur where q V = 0 If so, it will focus only on correcting the quantity distortion and will set access charge enough below marginal cost so as to induce a downstream transport price that equals system marginal cost γ V + η V 0 q V q 0 = q V q 15

Equilibrium combinations of q, X induced by varying c X X X V X V c = η V c = η V 0 q V Low λ (flat marginal cost of quality curve) δ close to 1 (big economies of 2 network size) b V close to 1 (intense price competition) q 0 = q V High λ (steep marginal cost of quality curve δ close to 0 (no economies of network size) b V close to 0 (weak price competition) q 16

We compute equilibria and compare VS and HS over wide swath of space of possible parameters Set γ V = γ H = γ and η V = η H = η. We vary γ [0,3] and η [0,3] Based on marginal cost of transport services and network operations reflect rail freight conditions as reported in Gargett, Mitchell, and Martin (1999) and Forkenbrock (2001) Marginal transport operating costs γ = 1.53 cents/ton-mile and the marginal network cost and η = 1.28 cents/ton-mile a [6,9] b V [0,0.99] b H [ 0.99,0.99] λ [2,10] To accord with our assumption that λ > φ(b H ) and the range b H [ 0.99,0.99] By trimming λ 10 we avoid outcomes with q V = 0, our analysis stacks the deck to favor higher quality under VS To further stack the deck in favor of higher quality under VS, we fix δ = 1 2 17

We report results to illustrate a thought experiment HS highly likely to give higher network quality for all values of b H Fix focal parameter (e.g., b H ) at a particular level, and then imagine randomly choosing all other parameter values within their designated ranges. Then ask in what proportion of these parameterizations does q H > q V, CS H > CS V, and SW H > SW V? Then vary focal parameter within its range and trace out resulting proportions q CS For b H we get the figure to the right Gives us two insights: Does structure tends to dominate the other on a given performance metric for a wide range of parameterizations? How do the two structures tend to perform as the focal parameter varies? SW HS highly likely to give higher consumer surplus for sufficiently positive values of b H HS highly likely to give higher social welfare for sufficiently positive values of b H but not too close to 1 b H 18

q q q SW SW CS CS b H λ SW CS b V q SW CS γ Plots for α, η are identical Intense price competition among transport operators under VS Make its very likely VS has higher CS and SW and moderately more likely it has higher network quality Variations in γ, α, η do not affect relative performance of VS and HS on q, CS, and SW. Thus, this plot shows general tendency of HS to outperform VS on the three metrics Highly likely HS outperforms VS on network quality Moderately likely VS outperforms HS on consumer surplus and social welfare 19

HS highly likely to dominate VS for positive value of b H but not too close to 1, i.e., moderate, but not cut-throat competition between the systems VS > HS HS > VS b H HS > VS VS > HS b V VS moderately likely to dominate HS on all three dimensions for a given parameterizations as competition intensity is high HS > VS VS > HS λ HS > VS VS > HS γ Plots for α, η are identical Not likely that either system will dominate the other for all parameterizations choice of structures may require trade-offs among worthy objectives But dominance is more likely under HS 20

Positive fixed network costs (independent of quality): Adding f > 0 would tend to accentuate advantages that VS would have due to economies of network size Cost disadvantage under VS: Several studies (e.g., Nash et. al., 2014) find evidence that adoption of VS increases operating costs If VS had a cost disadvantage, then because (for interior equilibria), qualities, quantities, and consumer surpluses are increasing functions of a γ η, it would follow that the already strong tendency for HS to generate higher network quality would become even stronger The set of cases in which HS yields greater consumer surplus and social welfare than VS and exhibits strong dominance also expand Asymmetric transport operators under VS and systems under HS: Does not change results in a meaningful way 21

One size does not fit all: relatively few parameterizations for which either HS or VS dominates on all three metrics: Choice among structures may involve prioritizing incentives HS tends to achieve higher network quality than VS Weaker tendency for VS to dominate on consumer surplus and social welfare Best case for VS to outperform HS on consumer surplus and social welfare: Competition between over the rails transport operators under VS is extremely strong (b V 1) Economies of network size are significant Best case for HS to outperform VS on consumer surplus and social welfare: Moderate but not cut-throat inter-system competition (b H > 0 but not too close to 1) 22

Intensity of competition between vertically integrated systems (under HS) or above the rail transport operators (under VS) captured by b V and b H are critical determinants of which structure delivers the highest consumer surplus and social welfare These parameters may be endogenous to policy decisions Under HS, b H is determined (at least partly) by how the existing network is divided up between competing systems and extent of shipper captivity that results under alternative designs Could also be affected by regulations governing trackage access to competitors Under VS, b V is determined (at least partly) by regulations governing service standards and network access. choice between HS and VS will inevitably reflect both the nature of a country s legacy rail system, as well the capacity of its regulatory institutions to promote competition 23

Best case for q V > 0 is when, dx V dc > 0 so regulator can reduce both the quantity and quality distortions by increasing c Since sign of dx V dc best case for dx V dq V dc dc = sign of dq V 1, dc > 0 is if dq V is big dc is big if the impact of quality on equilibrium total transport demand is big (see diagram to right) $ per unit of quality This bigger is b V, the bigger is d(x 1V D D +X 2V) The bigger is d(x 1V D +X 2V dq V D ) quality goes up as c goes up, the more MC q V dq V MB V q = c η d(x 1V D D +X 2V) dq V MB V q = c η d(x 1V D D +X 2V) dq V The bigger is b V, the more an increase in quality translates into higher equilibrium prices and higher total transport demand q V (c) q V (c ) q 25

Proposition 2: Suppose γ V = γ H = γ, η V = η H = η, and b V = b H = b. Whenever network quality under HS, q H, exceeds network quality under VS, q V, then HS also results in a higher level of consumer surplus than VS. If c V > η, then even if network quality under HS is merely the same as network quality under VS, HS still results in a higher level of consumer surplus than VS Upshot: If competition intensity and cost structures are roughly the same under HS and VS, a quality advantage under HS translates into a consumer surplus advantage for HS as well Proposition 3: Suppose q H < q and q L < q. In the limit, as transport operators become perfect complementors under HS, i.e., b H 1, and perfect substitutes under VS, i.e., b V 1, then provided VS does not have a cost disadvantage relative to HS, i.e., γ V + η V γ H + η H, consumer surplus is strictly greater under VS than under HS Upshot: if competition intensity under VS is high and systems under HS are complementors, then VS can have a consumer surplus advantage, despite the relative ranking of network qualities 26

Legacy structure of Ferrocarriles Nacionales de Mexico (FNM) did not seem well-suited to HS Though some opportunities for dividing up FNM's existing lines to create parallel competition between city-pairs in Mexico, they were not abundant In addition: given the size of Mexico's network (12,000 miles in total, or roughly 60 percent of the size of France's system), new railway networks built to create parallel competition would probably have been less than the 5,000-10,000 mile threshold needed to fully achieve economies of network size (Pittman 2007) Still... it was possible to make HS work in Mexico's legacy system by relying on source competition: Mexico divided FNM into three privately-owned regional lines, each emanating from Mexico City where the presence of an open-access terminal railroad (jointly owned by the three private firms and the government) allowed most shippers in the city to access any one of three railways Because a significant fraction of freight traffic in Mexico was bound for overseas, a shipper in Mexico City could often reach its ultimate destination over any one the regional lines, even though they went in different directions (Pittman, 2004b) Similarly, shipments arriving from Asia could be shipped to one of several ports along Mexico's Pacific coast and then transported to Mexico City by the (single) railroad serving that port 27

Why not VS? Mexico considered VS as a model for freight privatization, but rejected it One concern: quality of the legacy network infrastructure was too poor to safely dispatch multiple operators over the system Another concern (consistent with the insights of our model): VS was expected to induce less investment in the network than would be the case under HS (OECD 1997) In addition, Mexico's institutions for regulating railways were just emerging: Though HS requires regulatory oversight on issues such as safety and enforcement of haulage and trackage rights, VS probably requires even stronger regulatory oversight Why? Recall our analysis suggests that for VS to deliver more consumer surplus and social welfare above the rail competition has to be very intense The regulator needs to assure that transport operators have identical access to the network and offer identical services, subject to clear and enforceable service standards over all possible routes VS would have probably placed more severe demands on Mexico s regulatory capacity than HS 28