📈 economics

Screening with tolls and damages

This paper analyzes how a welfare-maximizing designer should optimally allocate two goods using tolls and damages, demonstrating that while damaging goods is never beneficial when agents differ only in their value for a single good, it becomes a useful screening tool when valuations for both goods are heterogeneous and positively affiliated.

Original authors: Filip Tokarski

Published 2026-07-15

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Filip Tokarski

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world where getting what you want isn't just about having enough money, but about how much you're willing to suffer to get it. This is the playground of mechanism design, a branch of economics that acts like a master architect for rules. Instead of just building bridges or designing apps, these architects design the systems that decide who gets scarce resources—like a spot in a crowded hospital, a ticket to a sold-out concert, or a slice of affordable housing. The big question they ask is: How do we sort people fairly and efficiently when we can't just ask them, "How much do you really want this?" (Because if we did, everyone would lie and say "a lot" to get a better deal).

To solve this, designers use screening instruments. Think of these as the hurdles you have to jump over to get a prize. There are two main types of hurdles. The first is a toll: a cost that feels the same to everyone, regardless of how much they want the prize. Paying a $10 fee or waiting in a line that takes an hour is a toll; it's annoying, but it doesn't hurt a desperate person any more than a casual one. The second type is a damage: a cost that hits harder the more you value the thing. If you're starving, a restriction that forces you to eat only plain toast is a huge "damage." If you're just looking for a snack, it's a minor annoyance. The paper asks a fascinating question: When a designer is trying to help the most people possible, should they use tolls, damages, or a mix of both?

The Great Sorting Game: Tolls vs. Damages

Imagine you are the manager of a very popular, very limited resource. Maybe it's a magical forest with only a few trees that grant wishes, or a city with only a handful of spots in a top-tier school. You have a crowd of people, and everyone wants in, but you can't let everyone. You need a way to sort them out.

In this paper, the author, Filip Tokarski, sets up a game with two kinds of "bad things" you can make people do to get the resource.

Tolls: These are like a flat fee or a long queue. Everyone pays the same price or waits the same amount of time. It's a "one-size-fits-all" annoyance.
Damages: These are like serving a slightly broken toy, making you wait for a specific, inconvenient time, or forcing you to use a service that is lower quality. The catch? The more you need the toy, the more painful the damage is. If you are desperate for a working toy, a broken one is a tragedy. If you just wanted it for fun, it's a minor disappointment.

The paper explores a simple rule first: If everyone is fighting for the exact same thing, you should never use damages.
Imagine a line for a single, perfect concert ticket. If you make the ticket "damaged" (maybe the sound is slightly off) to keep out people who don't really care, you end up hurting the people who do care the most. They are the ones who would have paid the most to get in, but now they are suffering the most from the damage. The paper proves mathematically that you can always do better by just using a toll (like a ticket price or a queue). A toll screens out the people who don't want it enough, but it leaves the super-fans happy and unharmed. Damages are just too cruel to the people you actually want to help.

The Twist: When You Have Two Different Things

But what if you aren't just selling one thing? What if you have two different kinds of resources, like Apples and Oranges? And what if some people love Apples but hate Oranges, while others are the opposite? Or maybe some people love both?

This is where the story gets spicy. The paper finds that when people have different tastes for different things, damages can actually be a good idea, but only under very specific conditions.

The secret sauce is how people's desires are connected.

The "Opposites Attract" Scenario (Negative Affiliation): Imagine a city with two parks. People who live near Park A really want to go there, but they hate Park B because it's far away. People near Park B feel the same. Here, loving one park means you likely hate the other. In this case, the paper says: Stick to tolls. Damages are less likely to be the right choice. Just charge different entry fees or make different queues for each park. If you try to "damage" one park (maybe make Park A have a broken slide), you mess up the sorting. The people who love Park A are the ones who suffer most, and they might just leave the system entirely.
The "All-or-Nothing" Scenario (Positive Affiliation): Now imagine a different world. Maybe it's a housing program. Some families are in huge need of housing; they would love any apartment, whether it's a small studio or a big house. Other families have nice homes already and don't really need anything. Here, if you love one type of house, you probably love the other type too (because you are just generally desperate). In this case, the paper suggests that damages are more likely to be the hero.

Why? Because when everyone who wants the goods loves them all equally, a simple price tag (toll) might not sort them well enough. You might end up with the super-needy families stuck in a long line for the "best" house, while the less-needy families grab the "okay" house just because it was cheaper. By "damaging" the best house (maybe making it slightly smaller or in a worse neighborhood), you force the super-needy families to realize, "Wait, I need any house, so I'll take the damaged one." This frees up the "perfect" house for someone else or just sorts the crowd more efficiently. It's like a bouncer at a club who says, "If you really want in, you have to wear this ugly hat." The people who really need to get in will wear the hat; the people who just want to look cool won't bother.

The Verdict

The paper doesn't just guess; it uses rigorous math to prove these points.

Proven Fact: If there is only one type of good, damages are always a bad idea. Use tolls (fees, queues, paperwork) instead.
Proven Fact: If there are two different goods, damages are more likely to be optimal when values for the two goods are positively related (if you want one, you likely want the other). If your desire for one good means you dislike the other (negative affiliation), tolls alone are likely to be optimal, and damages are less likely to help.
The "Why": The paper shows that damages work by changing how people choose between options, not just by keeping people out. They act like a filter that separates people based on how much they are willing to suffer for a specific type of good.

So, the next time you see a long waitlist for a doctor's appointment or a restriction on a food voucher, ask yourself: Is this a toll or a damage? If everyone is waiting for the same thing, the paper suggests we should probably just charge a fee or manage the queue better, rather than making the service worse. But if we are sorting people between different types of needs, sometimes making things a little "uglier" for the most desperate is actually the most efficient way to make sure everyone gets what they need. It's a counter-intuitive lesson: sometimes, to be kind to the most needy, you have to be a little mean to the system.

Technical Summary of "Screening with Tolls and Damages" by Filip Tokarski

1. Problem Statement

The paper addresses the design problem of a welfare-maximizing agent (the "designer") who must allocate scarce goods to a population of heterogeneous agents. The designer has access to two distinct classes of screening instruments:

Tolls: Costs that are separable from the agent's value for the good (e.g., monetary payments, queues, administrative burdens).
Damages: Costs that are more burdensome to agents with higher values for the good (e.g., quality reductions, delays in allocation, restrictions on use).

The central question is under what conditions a welfare-maximizing designer should utilize damages, as opposed to relying solely on tolls. The analysis covers two settings:

One-dimensional heterogeneity: Agents differ only in their value for a single good (with a common outside option).
Two-dimensional heterogeneity: Agents have heterogeneous values for two distinct, horizontally differentiated goods.

2. Methodology

The paper employs a mechanism design framework with the following specific features:

Deterministic Mechanisms: The analysis restricts attention to deterministic allocation rules, excluding lotteries. This restriction is motivated by practical concerns regarding trust and corruption, as well as analytical tractability.
Boundary Representation: In the two-dimensional setting, the author characterizes incentive-compatible mechanisms using a "boundary" in the type space. This boundary, denoted as $z(a)$ , separates agents who choose Good A from those who choose Good B.
Reduction to One-Dimensional Problems: By fixing the boundary, the multidimensional screening problem is decomposed into two interconnected single-dimensional screening problems. The optimal mechanism is found by first determining the optimal implementation for a given boundary and then optimizing over the shape of the boundary itself using optimal control arguments.
Virtual Valuation and Anti-Hazard Rates: The welfare objective is reformulated in terms of the boundary and indirect utilities. The optimality conditions rely heavily on the monotonicity properties of conditional inverse anti-hazard rates of the value distributions.
Extension to Heterogeneous Costs: The model is extended to allow agents to differ in the cost they incur from a toll (e.g., poorer agents facing higher monetary costs). This is handled by transforming the type space and reweighting the distribution to account for heterogeneous ordeal costs.

3. Key Results

One-Dimensional Heterogeneity

Proposition 1: When agents differ only in their value for one good, damages are strictly suboptimal. Any feasible mechanism using damages is Pareto-dominated by a mechanism using only tolls.
Intuition: Damages disproportionately burden inframarginal agents (those with high values) compared to the marginal agents they are intended to screen out. Tolls, being separable from value, can enforce the same allocation cutoff while leaving higher rents to inframarginal types.

Two-Dimensional Heterogeneity

Theorem 1 (Damages are not optimal): Under conditions where the conditional inverse anti-hazard rates of the value distributions are increasing (a condition compatible with negative affiliation or independence), the optimal mechanism uses only tolls.
- The optimal mechanism is the "market-clearing toll mechanism," where undamaged goods are offered at tolls $c^*_A$ and $c^*_B$ such that supplies are exhausted.
- This result holds regardless of the designer's weight on toll revenue ( $\gamma \in [0,1]$ ).
Theorem 2 (Damages are optimal): Damages can be optimal when the values for the two goods are positively affiliated.
- Specifically, if the covariance between the ratio of inframarginal beneficiaries to marginal switchers and the gain from a damaged option is positive, introducing a damaged option improves welfare.
- Intuitively, when values are positively affiliated, high value for one good predicts high value for the other. Damages allow the designer to sort agents more efficiently by changing the shape of the boundary in the type space, capturing surplus that tolls alone would absorb.
Corollary 1: In the case of fully wasteful tolls ( $\gamma=0$ ) and unit total supply, damages are optimal if the conditional covariance condition holds.

Heterogeneous Toll Costs

Corollaries 2 & 3: When agents have heterogeneous costs for tolls (parameterized by $r$ ), the results hold after transforming the type space. Damages become more attractive when agents with low "transformed values" (value relative to toll cost) have high toll costs. In such cases, a damaged option with a toll discount generates significant welfare gains by alleviating the burden on high-need agents who are otherwise priced out by monetary tolls.

4. Contributions and Significance

The paper makes two primary contributions:

Theoretical Distinction of Screening Instruments: It formally distinguishes between "tolls" and "damages" based on their interaction with agent values. It establishes that while tolls are universally superior for homogeneous goods, damages can be welfare-enhancing for heterogeneous goods when values are positively affiliated. This clarifies the economic logic behind why waitlists (damages) might be optimal in some contexts (e.g., public housing with high correlated need) while fees (tolls) are optimal in others (e.g., geographically dispersed appointments).
Methodological Advancement: The paper provides a tractable approach to multidimensional screening problems by restricting attention to deterministic mechanisms. By representing mechanisms via boundaries in the type space, the author reduces a complex multidimensional problem to a sequence of one-dimensional problems connected by an optimal control problem. This framework allows for the characterization of optimal mechanisms in settings with two goods, a domain where multidimensional screening is typically intractable.

The author notes that the results offer specific design prescriptions: for homogeneous goods, replace damages with tolls; for heterogeneous goods, the choice depends on the statistical dependence of values. If values are negatively related (e.g., proximity preferences), use tolls; if positively related (e.g., overall need), damages may be part of the optimal mechanism. The paper concludes that this methodology could be applied to other settings, such as two-good monopoly pricing or optimal tax problems, provided deterministic mechanisms are economically plausible.