Large-Market Discipline in Combinatorial Double Auctions: No Recombination, Bundle Selection, and Complementarities

arXiv:2608.06134 · econ.GN, cs.GT, q-fin.EC · Submitted 2026-08-06 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Large-Market Discipline in Combinatorial Double Auctions".

Warren: Here's the gist of the paper. The study investigates double auctions for markets dealing with goods valuable in bundles, such as data or AI assets, focusing on a key friction: No Assembly, where platforms cannot legally or technically combine components from different sellers into a single bundle.

Ruth: The authors formulate this as a combinatorial buyer’s-bid double auction under this constraint.

Warren: Under explicit stability and price-influence conditions—maintained generally and for two goods derived from local price-taking and a feedback bound—each bundle submarket inherits the large-market discipline of single-good double auctions: bid shading vanishes, and clearing prices concentrate on competitive levels tracking the common value (price discovery).

Ruth: The key incentive step—that bidding on a bundle creates no first-order strategic distortion beyond single-good logic—is proven for two goods, and this condition is maintained for larger item sets.

Title and authors: Ruth: So we're talking about this paper today, "Large-Market Discipline in Combinatorial Double Auctions: No Recombination, Bundle Selection, and Complementarities." It looks at markets where things are valuable in bundles, like data or AI assets. The main sticking point they focus on is the "No Assembly" problem.

Warren: Right. So platforms can't legally or technically put different pieces from different sellers together into one bundle. That's the friction they are looking at when they model this as a combinatorial buyer's-bid double auction under that constraint.

Iris: And what they do is look at how this structure behaves under some conditions, specifically explicit stability and price-influence conditions, and see if it inherits the large-market discipline from single-good double auctions. They are checking if bid shading disappears and prices start tracking the common value, which is kind of like price discovery.

Ruth: Exactly. The core idea they prove is that for two goods, bidding on a bundle doesn't create any first-order strategic distortion beyond what you would see in a single-good auction. And they say this condition holds even when you have more than two goods, which is pretty important for real markets.

Warren: That sounds promising because it suggests that the fundamental incentives for traders don't change much just because the items are bundled together, as long as assembly isn't allowed. But I wonder if that holds up in practice when you have many items?

Iris: That's a fair question. The paper uses a submarket reduction technique to show this behavior, conditional on the market sorting traders into thick, regular bundle submarkets. They argue that as a submarket gets thicker, any single trader’s chance of setting the price shrinks, which reduces the incentive for shading.

Ruth: So it's not about some magical mechanism preventing misreporting; it's about the structure of the market itself leading to less incentive to deviate from truthfulness as things get more crowded. It keeps attention on where inefficiency comes from rather than just trying to patch reporting errors.

Warren: I read that they use a specific auction mechanism, a combinatorial double auction with XOR bidding language, and their allocation rule picks the welfare-maximizing trade with the highest total clearing prices. That's how they test this theory in simulation.

Iris: They also define some losses to measure how much stuff is going wrong. They call the structural loss, which is about the difference between what a truthful planner could get versus what happens, and the allocational loss, which measures strategic inefficiency from misreporting.

Title and authors: Ruth: And they find that in moderately thick markets, both those losses are small. This suggests that even with this No Assembly friction present, things can still run pretty efficiently if the market is large enough and the goods are genuine complements.

Warren: Wait, they also have a few ideas for improvement themselves. They suggest designing the system so that "bid shading vanishes" as market thickness increases, meaning buyers stop reporting below their true values because "the gain from unilateral deviation shrinks."

Iris: That points to a design solution, trying to make the system itself enforce truthfulness by making it less profitable to cheat when the market is dense. They also suggest ensuring competitive price discovery where clearing prices track the common value.

Ruth: And structurally, they propose that if you want to address No Assembly directly, the platform design needs to ensure that "a buyer's bundle S must be supplied by a single seller offering exactly S," so trades only happen for pre-assembled bundles.

Warren: That sounds like a very specific constraint on how the auction must operate, moving away from just letting any components combine freely. It makes the structural problem explicit instead of hoping it disappears naturally.

Iris: And they use simulation evidence to show that the structural loss from No Assembly actually exceeds the allocational loss in every scenario they test, and this difference shrinks as synergy between goods increases. This suggests complementarity helps reduce both types of inefficiency simultaneously.

Ruth: So, their conclusion is that a decentralized bundle auction can be efficient and informative once the market is moderately thick and the goods are true complements, without needing a central authority to put things together. The binding friction isn't strategic misreporting; it's this inability to assemble.

Warren: That’s a big picture shift if you think about how AI assets or data might be traded in reality, where physical assembly is often impossible anyway. It means the market structure itself can solve some of the coordination problems.

Iris: Before we wrap up, I just want to reiterate that the paper's main finding relies heavily on those explicit stability and price-influence conditions being met for two goods, and they note that proving this for larger item sets remains a maintained condition.

Ruth: And Warren, from an economic standpoint, it seems like a solid result because it moves the focus away from trying to fix individual bidders' incentives toward understanding how the market structure itself handles bundles.

Warren: Yeah, but I’m still thinking about what happens when you scale this up to thousands of items. The paper suggests that while the friction shrinks at a rate close to one over n as market size grows, that still means something for very large markets.

Title and authors: Iris: It does mean that even in those larger markets, the structural loss bound is conjectured to be C times n, and numerical verification supports this. The No-Assembly price gap they suggest also satisfies a bound of C one times e to negative c n plus C two over n.

Ruth: It sounds like they’ve done a lot of heavy lifting showing that the large-market virtues of single-good auctions can survive this structural constraint in bundle settings. It's a strong argument for how decentralized markets can function when components are hard to combine.

Warren: So, to bring it all together, the paper "Large-Market Discipline in Combinatorial Double Auctions: No Recombination, Bundle Selection, and Complementarities" suggests that for these bundled goods markets, efficiency and informative pricing are achieved when the market is thick enough and when those goods are genuine complements.

Iris: And they point out that the binding friction at realistic sizes is not strategic misreporting but this inability to assemble, which complementarity helps mitigate as synergy increases.

Ruth: That's the main implication for me: we need to stop looking only at individual reporting errors and start focusing on how bundling and market density interact with structural limitations like No Assembly.

Warren: And I think for trading strategies, it means that if you can design a market where thickness naturally reduces the incentive to shade, then you might be able to deploy those strategies without worrying about bidders intentionally misreporting their values.

Iris: Before we move on, I just want to say that while the results are strong under their assumptions, the paper's limitation is that it focuses on two goods initially and proving that condition for larger item sets remains a maintained condition.

Ruth: And Warren, if you had to summarize what this means for how we think about markets dealing with complex digital assets right now, what’s your take?

Warren: I see it as an indicator that market design around bundles needs to account for physical or technical constraints like No Assembly from the start. It forces us to build in structural rules rather than relying solely on behavioral assumptions about perfect rationality.

Iris: I just hope future work can explore those larger item sets more thoroughly because that would really cement the idea that this discipline survives even when there are many different things being bundled together.

Ruth: We’ll keep an eye on those follow-up papers, but for now, this study on "Large-Market Discipline in Combinatorial Double Auctions: No Recombination, Bundle Selection, and Complementarities" shows how structural friction can be manageable through market thickness and complementarity.

The paper's summary: Ruth: So, we're talking about this paper today on combinatorial double auctions for bundled goods like data or AI assets, and they focus on why platforms can't legally combine different sellers into one bundle, which they call No Assembly.

Warren: Right. So the main point is that even without assembly, if the market gets thick enough and the items are genuine complements—meaning they help each other out—the system naturally settles down in a way that mimics single-item auctions.

Iris: What they’re showing is that under certain conditions, like explicit stability, this means bid shading vanishes and prices start tracking what everyone actually values, which is price discovery.

Ruth: They prove this holds for two items, but they maintain it for bigger sets of items too. The key finding is that bidding on a bundle doesn't create any extra strategic distortion beyond what you’d see in a single-item auction.

Warren: That sounds like good news because it means the fundamental incentives don't change much just because you have more things bundled together, as long as assembly isn't allowed. But they use simulations to show that No Assembly, not shading, is the main friction slowing things down in moderately thick markets.

Iris: They use those simulations to decompose the losses into structural loss and allocational loss. The structural loss comes from the fact that you can't assemble anything, while the allocational loss is about people misreporting their values.

Ruth: And they find that in moderately thick markets, both of those losses are actually pretty small. This suggests that if your market is big enough and your goods complement each other, things can still run efficiently without needing a central authority to put everything together.

Warren: The paper gives some design ideas too. They suggest making the system so that a buyer's bundle must be supplied by exactly one seller offering that specific bundle, which makes sure trades only happen for pre-assembled bundles.

Iris: That’s a way to tackle the No Assembly problem directly by setting a hard rule on how trades can occur, rather than just hoping the market structure fixes itself.

Ruth: The paper also shows that structural loss actually beats allocational loss in every test case they ran. And when you increase synergy between goods, both types of losses go down.

Warren: So, to wrap up the big picture, they’re saying that a decentralized bundle auction can achieve efficient allocation and good prices if the market is moderately thick and the goods are real complements. The binding friction isn't people lying about their values; it's just that you can't physically assemble things.

Iris: And they point out that this structural problem gets smaller as the market size grows, specifically shrinking at a rate close to one over n as the number of items increases.

Ruth: That means even in really big markets, there’s a predictable way for efficiency to be restored through thickness and complementarity. It shifts our focus away from just trying to fix individual reporting errors toward designing market structures that handle these bundle constraints naturally.

The paper's improvements: Tom: So we're talking about how they suggest fixing the model for these bundle auctions, moving beyond just describing what’s happening right now.

Ruth: They suggest two main things to make this system better: first, design it so bid shading naturally vanishes as the market gets thicker. This means buyers stop reporting values below their true cost because the incentive to cheat just shrinks with crowding.

Warren: If bid shading goes away because of thickness, that changes how traders behave in a real auction environment. It means you don't have to worry as much about bidders intentionally misreporting their private information if the market is dense enough.

Iris: They also suggest making the structure enforce assembly rules more strictly so that a buyer’s bundle must be supplied by exactly one seller offering that exact bundle. This makes the No Assembly constraint explicit in how trades happen.

Ruth: That structural fix, combined with the thickness effect, is supposed to lead to competitive price discovery where clearing prices track what everyone actually values, which is good for outsiders who might not know the true values.

Warren: I like that idea of enforcing a strict supply rule. It moves the solution from being purely behavioral—people stop lying—to being structural—the mechanism physically prevents the lie.

Iris: And they use those welfare metrics to measure this, looking at the gap between what a central planner could get if assembly were allowed versus what this auction gets under truthful bidding. That gives us a concrete benchmark for how much efficiency we’re losing.

Ruth: It sounds like they’re pushing for market design that builds in these self-correcting properties instead of relying only on behavioral assumptions about perfect rationality.

Warren: So, the implication is that if you design the platform correctly—with those strict supply rules and a mechanism sensitive to market thickness—you can achieve efficient allocation without needing a central authority to manage all those pieces for everyone.

Iris: They flag that while these results are strong under their current assumptions, they have to keep in mind that proving this discipline holds up when you have many different types of items bundled together remains an open challenge.

Ruth: That's a fair caveat. The paper proves it for two goods, and they hope the core idea survives when you scale up to thousands of different assets being combined.

Warren: So, the next thing we look at is whether this structural solution actually works when you introduce more complex things, like those many items they mentioned.

Conclusion: Tom: So we’re wrapping up on "Large-Market Discipline in Combinatorial Double Auctions: No Recombination, Bundle Selection, and Complementarities" by Ruth and Warren today.

Ruth: We looked at how the market structure itself—specifically the No Assembly rule—can actually lead to efficiency when goods are complements and the market gets thick.

Warren: It basically means that instead of just blaming individual traders for misreporting their values, we can look at how bundling and market density interact to create a more stable system.

Iris: I just want to stress that this whole argument relies on those explicit stability conditions holding true, especially when you move beyond just two different types of goods into larger sets.

Ruth: Exactly. The authors are showing us that if the structure is right, we don't need to rely on perfect individual behavior for the whole system to function well.

Warren: For a trader looking at this, it suggests that if you can design a market where thickness naturally reduces the incentive to shade, you might be able to deploy those strategies without worrying as much about bidders intentionally misreporting their private information.

Iris: It’s important to remember that the authors flagged something specific: they haven't fully proven this discipline holds up for any arbitrary number of items yet.

Ruth: True. The work is strong, but it opens the door for future research into scaling this idea to much more complex scenarios where you have a huge variety of bundled assets.

Warren: Moving forward, I’m curious to see if other papers explore how these structural rules might interact with things like dynamic mechanism design or those complex information aggregation models we’ve been discussing.

Iris: Yeah, that makes sense. We need to check if this discipline survives when you start mixing in hidden actions or complicated endogeneity issues into the bundle structure itself.

Ruth: Well, for now, the paper shows us a clear path: design markets with structural rules that account for physical limits like No Assembly to get better outcomes without needing external coordination.

Warren: That’s what we’re heading toward—designing the market itself to handle complexity rather than just trying to fix the people in it.

Iris: We'll be keeping an eye on those follow-up papers as they tackle those larger item sets and see if this discipline holds firm under more demanding conditions.

Konstantinos E. Zachariadis, Yongxin Yang

School of Economics and Finance, Queen Mary University of London · School of Electronic Engineering and Computer Science, Queen Mary University of London

econ.GN, cs.GT, q-fin.EC

Submitted: 2026-08-06

Updated: 2026-11-08

Comments: 107 pages, including a 55-page online appendix

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 92/100

The gist: The study investigates double auctions for markets dealing with goods valuable in bundles, such as data or AI assets, focusing on a key friction: No Assembly…

Key concepts

No Assembly
This is the constraint where platforms cannot legally or technically combine items from different sellers into a single bundle. It forces buyers to bid on pre-formed bundles rather than assembling them themselves, creating a key friction point in bundle markets.
Bid Shading Vanishing
This means that as the market becomes very large (thick), the incentive for buyers to lie by bidding lower than their true value disappears. This happens because the potential gain from strategically underbidding on a bundle shrinks as more traders participate.
Submarket Reduction
The analysis simplifies complex bundle markets by assuming that when a market is thick, individual traders behave like they are in a simpler, single-good auction. This reduction helps prove that strategic behavior becomes less distorted in larger markets.

Terminology

Summary

The study investigates double auctions for markets dealing with goods valuable in bundles, such as data or AI assets, focusing on a key friction: No Assembly, where platforms cannot legally or technically combine components from different sellers into a single bundle. p. 0

Where the paper says it

Zachariadis∗ Yongxin Yang† August 18, 2026 Abstract We study double auctions for markets in which goods are valuable in bundles, such as data, model weights, and fine-tuned AI assets. A key friction in such markets is No Assembly: a platform may be unable, for legal or technical reasons, to combine components supplied by different sellers into a single bundle.

The authors formulate this as a combinatorial buyer’s-bid double auction under this constraint. p. 0
Where the paper says it

A key friction in such markets is No Assembly: a platform may be unable, for legal or technical reasons, to combine components supplied by different sellers into a single bundle. We formulate a combinatorial buyer’s-bid double auction under this constraint.

Under explicit stability and price-influence conditions—maintained generally and for two goods derived from local price-taking and a feedback bound—each bundle submarket inherits the large-market discipline of single-good double auctions: bid shading vanishes, and clearing prices concentrate on competitive levels tracking the common value (price discovery). p. 0
Where the paper says it

We formulate a combinatorial buyer’s-bid double auction under this constraint. Under explicit stability and price-influence conditions— maintained in general, and for two goods derived from local price-taking and a feedback bound—each bundle submarket inherits the large-market discipline of single-good double auctions: bid shading vanishes, and clearing prices concentrate on competitive levels and track the common value (price discovery).

The key incentive step—that bidding on a bundle creates no first-order strategic distortion beyond single-good logic—is proven for two goods, and this condition is maintained for larger item sets. p. 0
Where the paper says it

Under explicit stability and price-influence conditions— maintained in general, and for two goods derived from local price-taking and a feedback bound—each bundle submarket inherits the large-market discipline of single-good double auctions: bid shading vanishes, and clearing prices concentrate on competitive levels and track the common value (price discovery). The key incentive step—that bidding on a bundle creates no first-order strategic distortion beyond the single-good logic—is proved for two goods; for larger item sets it remains a maintained condition.

Multi-agent reinforcement learning simulations decompose the welfare loss, indicating that No Assembly, rather than strategic shading, is the binding finite-market friction. p. 0
Where the paper says it

The key incentive step—that bidding on a bundle creates no first-order strategic distortion beyond the single-good logic—is proved for two goods; for larger item sets it remains a maintained condition. Multi-agent reinforcementlearning simulations decompose the welfare loss and indicate that No Assembly, not strategic shading, is the binding finite-market friction, with both losses small in moderately thick markets and declining with complementarity amongst goods.

The simulation evidence suggests that structural costs are the larger of the two losses in moderately thick markets, and both costs diminish with market growth and complementarity between goods. p. 0
Where the paper says it

The key incentive step—that bidding on a bundle creates no first-order strategic distortion beyond the single-good logic—is proved for two goods; for larger item sets it remains a maintained condition. Multi-agent reinforcementlearning simulations decompose the welfare loss and indicate that No Assembly, not strategic shading, is the binding finite-market friction, with both losses small in moderately thick markets and declining with complementarity amongst goods.

The model features buyers and sellers trading a fixed set of items, where each trader wants or can supply at most one bundle. p. 1

Where the paper says it

Complementarities help: stronger synergies lead highvalue traders to specialise in supplying full bundles, which thickens the bundle market and lowers both sources of welfare loss—the structural cost of No Assembly and the strategic cost of shading. The model has buyers and sellers—not necessarily equal in number—trading a fixed set of items, with each trader wanting, or able to supply, at most one bundle.

A trader’s value or cost for a bundle is composed of a common component shared by everyone plus an idiosyncratic component personal to the trader, with synergies increasing the full bundle's value above the sum of its parts. The analysis assumes correlated private values (CPV), where every trader knows their own value exactly, and values are linked across traders only through this common component. p. 2
Where the paper says it

every trader knows his own value exactly, and values are linked across traders only through the common component. This keeps attention on the structural and strategic sources of inefficiency and sets aside the winner’s curse, which we take up as an extension.

The central analytical device is submarket reduction: conditional on market sorting traders into thick, regular bundle submarkets, each behaves in the large-market limit like a single-good buyer’s bid double auction run on the selected traders. p. 2

Where the paper says it

Our central device is a submarket reduction. Conditional on the market sorting traders into thick, regular bundle submarkets, each submarket behaves in the large-market limit like a single-good buyer’s bid double auction run on the traders who selected into it, and the established single-good results then apply submarket by submarket.

The intuition for reduced strategic behavior is that as a submarket thickens, any single trader’s chance of setting the price shrinks, reducing the gain from shading. p. 2
Where the paper says it

Conditional on the market sorting traders into thick, regular bundle submarkets, each submarket behaves in the large-market limit like a single-good buyer’s bid double auction run on the traders who selected into it, and the established single-good results then apply submarket by submarket. The intuition for why strategic behaviour fades is that, as a submarket thickens, any single trader’s chance of being the one who sets the price shrinks, and with it the gain from shading—from reporting below one’s value in the hope of a better price.

The mechanism used is a combinatorial double auction, utilizing XOR bidding language [Nisan, 2006], which is fully expressive over bundles. p. 6

Where the paper says it

Buyers submit bids bm(S) ∈ R and sellers submit asks an(S) ∈ R for the bundles they wish to trade; in the language of combinatorial auctions, this format, in which a price is quoted for every bundle and each trader is awarded at most one, is the XOR bidding language [Nisan, 2006]. It is fully expressive over bundles, at the cost of a message space that grows exponentially with the number of items, which is one reason the analysis focuses on a small item set.

The allocation rule employs the Buyer’s Bid selection rule: clearing prices are read off the buyers’ side of the market—similar to single-item cases where price is set by a bid—and among welfare-maximizing allocations, it selects the one with the highest total clearing prices. p. 6
Where the paper says it

We adopt the Buyer’s Bid selection rule: clearing prices are read off the buyers’ side of the market—in the single-item case the price is an order statistic set by a bid, never by an ask (§3. 2)—and, among welfare-maximising allocations, the rule picks the one with the highest total clearing prices, a selection that favours the price-taking seller side.

The structural loss from No Assembly is defined as ELstructural 1 - E[WTruthful]/E[WOptimal], while the allocational efficiency loss, or strategic loss, is ELallocational 1 - E[WRealized]/E[WTruthful]. p. 1

Where the paper says it

The one new force in the combinatorial setting is that a bid on a bundle can pull marginal traders across submarkets and so move prices indirectly; we show that as long as this feedback is not too strong it does not overturn the single-good logic, and for the two-good case we prove this under explicit price-influence and stability conditions. Complementarities help: stronger synergies lead highvalue traders to specialise in supplying full bundles, which thickens the bundle market and lowers both sources of welfare loss—the structural cost of No Assembly and the strategic cost of shading.

Simulations show both losses are small in moderately thick markets. p. 0
Where the paper says it

The key incentive step—that bidding on a bundle creates no first-order strategic distortion beyond the single-good logic—is proved for two goods; for larger item sets it remains a maintained condition. Multi-agent reinforcementlearning simulations decompose the welfare loss and indicate that No Assembly, not strategic shading, is the binding finite-market friction, with both losses small in moderately thick markets and declining with complementarity amongst goods.

The theory concludes that a decentralized bundle auction can achieve efficient allocation and informative prices once the market is moderately thick and goods are genuine complements, requiring no central authority for assembly. p. 2

Where the paper says it

Taken together, the theory and the simulations deliver a single practical message. A decentralised bundle auction can serve both purposes of a market—efficient allocation and informative prices—once it is moderately thick and the goods are genuine complements, and it needs no central authority to assemble bundles on traders’ behalf.

The binding friction at realistic sizes is the inability to assemble, not strategic misreporting; complementarity is shown to thicken bundle markets and restore efficiency by inducing endogenous self-selection of natural assemblers into full-bundle roles. p. 2
Where the paper says it

A decentralised bundle auction can serve both purposes of a market—efficient allocation and informative prices—once it is moderately thick and the goods are genuine complements, and it needs no central authority to assemble bundles on traders’ behalf. The friction that matters at realistic market sizes is the inability to assemble, not strategic misreporting; and complementarities, far from making matters worse, are exactly the force that thickens bundle markets and restores efficiency.

Key results include the conditional large-market rate result for the two-item combinatorial BBDA: lambda eta buyer(S) = O(1/eta) and lambda eta seller(S) = O(1/eta) as eta to infinity. p. 2

Where the paper says it

Our central device is a submarket reduction. Conditional on the market sorting traders into thick, regular bundle submarkets, each submarket behaves in the large-market limit like a single-good buyer’s bid double auction run on the traders who selected into it, and the established single-good results then apply submarket by submarket.

The structural loss bound is conjectured as E ELstructural at most C eta, which numerical verification supports. p. 1
Where the paper says it

The one new force in the combinatorial setting is that a bid on a bundle can pull marginal traders across submarkets and so move prices indirectly; we show that as long as this feedback is not too strong it does not overturn the single-good logic, and for the two-good case we prove this under explicit price-influence and stability conditions. Complementarities help: stronger synergies lead highvalue traders to specialise in supplying full bundles, which thickens the bundle market and lowers both sources of welfare loss—the structural cost of No Assembly and the strategic cost of shading.

The conjectured No-Assembly price gap satisfies E[delta p, eta] at most C 1e-c eta + C 2/eta. p. 1
Where the paper says it

As a result, it is not known whether the large-market virtues of single-good double auctions, namely disappearing strategic behaviour, competitive pricing, and price discovery, survive once goods are traded in bundles and assembly is not allowed. That is the gap we address.

The empirical results confirm that the structural loss from No Assembly exceeds the allocational loss in every (eta, kappa bar) cell, with complementarity helping to reduce both components as synergy (kappa bar) increases. p. 2

Where the paper says it

Across market sizes and synergy levels the structural cost is the larger of the two, a few percent of available surplus at moderate market size. Both costs shrink as the market grows and as complementarities strengthen, so that at roughly eight traders on each side of the market the combined loss is small.

The binding friction is finite-market dependent, shrinking at a rate close to 1/eta as market size grows. p. 0
Where the paper says it

The key incentive step—that bidding on a bundle creates no first-order strategic distortion beyond the single-good logic—is proved for two goods; for larger item sets it remains a maintained condition. Multi-agent reinforcementlearning simulations decompose the welfare loss and indicate that No Assembly, not strategic shading, is the binding finite-market friction, with both losses small in moderately thick markets and declining with complementarity amongst goods.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Bid Shading Vanishing: The system can be designed such that bid shading vanishes as market thickness increases, meaning buyers stop reporting below their true values because the gain from unilateral deviation shrinks.

  2. Competitive Price Discovery: The system will ensure that clearing prices concentrate on competitive levels and track the common value, allowing outsiders to use prices to guide decisions effectively.

  3. Structural Friction Mitigation: The platform design must address the No Assembly friction by using a mechanism where a buyer’s bundle S must be supplied by a single seller offering exactly S, ensuring trades occur only for pre-assembled bundles.

  4. Welfare Benchmarking: The system can utilize the defined welfare metrics to assess performance, specifically measuring the gap between the surplus a central planner could achieve if assembly were allowed and the surplus the auction achieves under truthful bidding.

Abstract

Licences can forbid assembling bundles from several sellers or splitting them among buyers, a restriction we call No Recombination. We study its welfare cost in a double auction as the number of traders on each side of the market grows. We ask whether that cost vanishes as a share of welfare, as the gain from misreporting does with a single good. With two goods and Gaussian shocks, consider the prices that would prevail under free recombination. As the market grows, the loss as a share of optimal welfare vanishes if and only if expected demand for the two-good bundle equals expected supply of whole bundles at those prices. Otherwise the loss persists. In our market, sellers can never gain by misreporting. With two goods, buyers who commit to bidding their values plus fixed margins misreport in symmetric equilibria by amounts that vanish as the market grows. Under these conditions, in large markets No Recombination outweighs strategic bidding as a source of loss.

Sources

Related papers