Screening with tolls and damages

This paper demonstrates that while tolls alone suffice when agents differ in value for only one good, a welfare-maximizing designer should optimally employ damages (such as quality reductions or delays) alongside tolls to screen agents when their valuations for both goods are heterogeneous and positively affiliated.

The Gist

Imagine you are the manager of a popular community center. You have limited resources: a few spots for a Gym (Good A) and a few spots for a Library (Good B). You have a huge crowd of people wanting in, but you can't let everyone in. You need a way to decide who gets a spot and who doesn't, while making sure the people who need these services the most are the ones who get them.

The paper by Filip Tokarski explores two different tools you can use to manage this crowd: Tolls and Damages.

The Two Tools

1. Tolls (The "Flat Fee" or "Wait in Line"):
   Think of a toll as a price tag or a line you have to stand in.

   • How it works: Everyone pays the same amount of money, or everyone waits the same amount of time in a queue.
   • The catch: The cost is "separable." Whether you are a fitness junkie who really wants the gym or someone who just wants to try it out, the line takes the same amount of time for both of you. The cost doesn't change based on how much you value the spot.
   • Examples: Paying a $5 fee, waiting in a physical line, or filling out a long form.

2. Damages (The "Quality Cut" or "Delay"):
   Think of damages as a deliberate reduction in the quality of the service or a delay that hurts the most eager people the most.

   • How it works: You give the gym a broken treadmill, or you make the library open only on Tuesdays.
   • The catch: This hurts people differently. A fitness junkie (who values the gym highly) suffers much more from a broken treadmill than a casual walker. The cost is tied directly to how much they want the item.
   • Examples: Offering a slower internet plan, delaying an appointment by weeks, or restricting what you can buy with a food voucher.

The Big Question

The paper asks: When should a manager use "Tolls" and when should they use "Damages" to get the best result for society?

The Two Scenarios

Scenario 1: Everyone wants the same thing (The "One-Item" Case)

Imagine you only have Gym spots. Everyone wants to go to the gym, but some people want it more than others.

• The Paper's Finding: If you only have one type of good, never use Damages.
• The Analogy: Imagine you have a line of people waiting for a free concert ticket.
  • If you use a Toll (a $10 fee), the person who loves music the most is willing to pay $10. The person who doesn't care much won't pay. The line is sorted perfectly.
  • If you use Damage (you make the concert start 2 hours late), the person who loves music the most suffers the most. They lose 2 hours of enjoyment. The person who doesn't care much loses 2 hours of boredom.
  • Why Tolls win: The "Toll" screens people out without hurting the people who do get in. The "Damage" hurts the people who do get in just as much as the people who don't. It's like punishing your best customers to keep out the bad ones. The paper proves that using a toll is always better because it leaves more "happiness" (welfare) on the table.

Scenario 2: People want different things (The "Two-Item" Case)

Now imagine you have both a Gym and a Library. Some people love the Gym but hate the Library; others love the Library but hate the Gym. Some love both; some love neither.

• The Paper's Finding: Here, Damages can sometimes be the best tool.
• The Analogy: Think of sorting people into two different rooms.
  • If you only use Tolls (prices), you draw a straight diagonal line on a graph. People with high values for both go to the expensive room; people with low values for both go nowhere.
  • But sometimes, you want to sort people in a weird way. Maybe you want to send people who love the Gym slightly more than the Library to the Gym, even if they don't love it that much.
  • How Damages help: By "damaging" the Library (making it slower or lower quality), you make it less attractive to the people who really want the Gym. This changes the "sorting" pattern. It's like putting a speed bump on the road to the Library; the people who really want the Gym will drive around it, while the people who are just "okay" with the Library will take the bump.
  • When does this work? It works best when people's desires are linked. If people who love the Gym also tend to love the Library (positive connection), then damaging the Library can help separate the "super fans" from the "casuals" in a way that prices alone can't.
  • When does it fail? If people who love the Gym usually hate the Library (negative connection), then just using prices (Tolls) is better.

The "Market-Clearing" Rule

The paper also identifies a special "Goldilocks" scenario. If the distribution of people's desires follows a specific mathematical pattern (where high values for one item predict low values for the other), the best solution is simply to set prices (Tolls) so that the supply exactly matches the demand. No damage, no weird tricks—just a fair market price that clears the line.

Summary in Plain English

• If you have only one thing to give out: Don't mess it up. Don't lower the quality or delay it. Just charge a fee or make them wait in line. It's fairer and more efficient.
• If you have two different things to give out: Sometimes, making one of them "worse" (damaging it) is actually a smart way to sort people. It helps direct the right people to the right spot, especially if the people who want one thing also tend to want the other.
• The Bottom Line: The best tool depends on how people's desires are connected. If their desires are opposites, use prices. If their desires are linked, sometimes a little bit of "damage" is the most efficient way to sort the crowd.

Technical Summary

Technical Summary of "Screening with Tolls and Damages"

Problem Statement
This paper addresses the design of welfare-maximizing mechanisms for allocating scarce goods when the designer can utilize two distinct classes of screening instruments: tolls and damages.

• Tolls: Instruments whose costs are separable from agents' values for the allocated good. Examples include monetary payments, queues, and administrative burdens.
• Damages: Instruments that are more costly to agents with higher values for the good. Examples include quality reductions, delays in allocation, and restrictions on use.

The central question is under what conditions a welfare-maximizing designer should employ damages (which are inherently wasteful) versus relying solely on tolls. The analysis considers two settings: (1) agents differing only in their value for a single good (with a common-value outside option), and (2) agents with heterogeneous values for two distinct goods.

Methodology
The author restricts attention to deterministic mechanisms, a choice motivated by practical constraints (e.g., concerns over lottery manipulation in public programs) and analytical tractability. This restriction allows the multidimensional screening problem to be decomposed into interconnected single-dimensional problems.

The core methodological contribution is the boundary representation of mechanisms. In the two-good setting, the set of agents choosing good A versus good B is separated by a strictly increasing, continuous boundary $z$ in the type space $[0,1]^2$.

1. Reduction to One Dimension: By defining good-specific indirect utility functions $U_A(a)$ and $U_B(b)$, the mechanism is characterized by the boundary $z$ and the utility profile along it.
2. Two-Stage Optimization: The problem is solved in two stages:
   • First, for a fixed boundary $z$, the optimal implementation (qualities and tolls) is determined to maximize welfare while satisfying incentive compatibility.
   • Second, an optimal control problem is solved to find the welfare-maximizing boundary shape among all implementable ones.
3. Key Technical Conditions: The analysis relies on assumptions regarding the distribution of values, specifically the monotonicity of the inverse anti-hazard rates of the conditional distributions.

Key Results

1. One-Dimensional Heterogeneity (Single Good)
When agents differ only in their value for one good (with a common outside option), the paper proves that damages are never optimal.

• Proposition 1: Any feasible mechanism using damages is weakly Pareto-dominated by a mechanism using only tolls.
• Intuition: Damages impose a higher burden on inframarginal agents (those with high values) than on the marginal agents they are intended to deter. Tolls, being separable from value, burden all agents equally. Therefore, enforcing a cutoff via tolls leaves higher rents to inframarginal agents, increasing total welfare.

2. Two-Dimensional Heterogeneity (Two Goods)
When agents have heterogeneous values for two goods, the optimal mechanism depends on the statistical dependence between these values.

• Theorem 1 (Damages are Suboptimal): If the conditional inverse anti-hazard rates of the value distribution are increasing (a condition satisfied by negative affiliation or independence), the optimal mechanism is the market-clearing toll mechanism. In this mechanism, goods are undamaged, and tolls are set such that supply is exhausted.
• Theorem 2 (Damages are Optimal): The paper provides sufficient conditions under which damages are optimal. This occurs when values for the two goods are positively affiliated (strongly positively correlated).
  • Mechanism: Under positive affiliation, a damaged option allows the designer to "sort" agents more efficiently. A damaged option with a toll discount can attract agents with strong relative preferences for one good without inducing the massive "toll-correction" effects (price increases required to clear the market) that would occur if only tolls were used.
  • Condition: Damages are optimal if the covariance between the inverse anti-hazard rate and the gain from the damaged option is positive. This typically holds when high values for one good predict high values for the other.

3. Heterogeneous Toll Costs
The paper extends the model to allow agents to differ in the cost they incur to complete a toll (e.g., poorer agents finding monetary payments more burdensome). By transforming the type space, the main results hold: the optimality of damages still hinges on the correlation structure of the transformed values.

Significance and Contributions

• Distinction of Instruments: The paper formally distinguishes between "tolls" and "damages" based on how their costs scale with agent value, a distinction often blurred in literature on self-targeting and rationing.
• Sorting vs. Exclusion: While previous literature focused on screening devices as tools for exclusion (determining who gets the good), this paper highlights a second role: sorting agents across heterogeneous goods. Damages can be welfare-improving not because they exclude low-value agents, but because they direct high-value agents toward the specific good they value most relative to the alternative, preserving surplus that would otherwise be consumed by tolls.
• Market Design Implications:
  • For homogeneous goods, the paper advocates replacing delays and quality reductions with fees or queues.
  • For heterogeneous goods, the choice of instrument depends on the correlation of preferences. If preferences are negatively related (e.g., location-specific appointments), tolls suffice. If preferences are positively related (e.g., housing in a specific cluster), differentiated waitlists (damages) combined with tolls may be optimal.
• Technical Tractability: The paper demonstrates that restricting to deterministic mechanisms renders a multidimensional screening problem tractable, offering a framework applicable to other settings such as monopoly pricing with unit-demand consumers or optimal tax design.

Limitations and Scope
The author notes that the restriction to deterministic mechanisms precludes the use of lotteries, though it is argued that randomization with re-entry is often equivalent to the deterministic options in this specific context. The results are derived for a welfarist designer; while the logic extends to profit-maximizing sellers, the specific conditions for optimality may differ. The paper does not propose specific experimental validations but grounds its theoretical findings in existing empirical observations of congestion pricing, healthcare referrals, and social program targeting.