Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space

arXiv:2609.37820 · q-fin.CP, econ.GN, q-fin.EC · Submitted 2026-09-29 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space".

Warren: Here's the gist of the paper. Firms follow changing economic opportunities, but rivalry changes their response <ref:2609.37820#pg2> We develop ESCAPE, a rational-share game of distribution-valued positioning with heterogeneous capability costs, establish a unique equilibrium, and derive an exact reallocation restriction separating opportunity and crowding contributions <ref:2609.37820#pg3> The central theoretical result concerns how the new position’s revenue advantage changes between environments <ref:2609.37820#pg4> Rational firms can enter a region that is becoming more crowded, because the example’s full gain is positive while its crowding contribution is negative <ref:2609.37820#pg5> The relative-response component accounts for 90.5% of the weighted squared composition-response contrast between strategic and independent firms in the benchmark economy <ref:2609.37820#pg6> Corporate-news evidence documents persistent but changing peer structure, showing that optimized peer reconstructions retain 86.5% of their aggregate validation advantage one year later even as absolute distance to both frozen reconstructions increases <ref:2609.37820#pg7> Economic Model Setup The firm’s repositioning problem involves choosing where to operate as the economic value of its opportunities changes <ref:2609.37820#pg8> The firm's payoff is defined by the expression πi(qi, q−i; rt) which incorporates the value of serving current opportunities, the cost of positioning there, and the effect of rivals on opportunity shares <ref:2609.37820#pg9> The model uses a proportional-share contest technology where revenue is allocated according to appeal shares <ref:2609.37820#pg10> Firms differ in their capability centers, which make nearby activities less costly at the margin, and this heterogeneity generates strategic differentiation Decomposition of Gains The full reallocation gain is decomposed into opportunity and crowding contributions based on the change in environment The full gain obeys J full i ≥ 0, (1) and can be separated into opportunity and crowding contributions The exact two-firm example demonstrates that rational firms can enter a crowded region because the full gain is positive while its crowding contribution is negative The opportunity component measures how the revaluation of demand regions changes that relative revenue holding opportunities fixed, and the crowding component measures how rival movement changes the new position’s relative revenue holding opportunities fixed Information Persistence The corpus evidence documents that peer structure persists even while firm positions move materially, with a fitted peer combination retaining most of its advantage even as both optimized and equal-weight reconstructions become more distant This persistence suggests that similarity and peer relevance are state-dependent, allowing an analyst to distinguish erosion in absolute fit from a loss of relative usefulness Measurement Margins The observation experiment separates economic scale and observed composition as distinct measurement margins, showing that the positive composition separation establishes that rivalry changes the allocation across activities as well as total scale The ratio of aggregate destination to validation advantages is 86.

Title and authors: Ruth: So we're talking about this paper called "Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space," and the authors are Marcus Gawronsky and Chun-Sung Huang. What's the big picture here for someone listening right now?

Warren: It sounds like it’s diving into how firms adjust their positions when opportunities shift, but rivalry changes how they react, which is a key tension in markets.

Ruth: Exactly. The paper focuses on this idea that firms aren't just reacting to where the economic value of opportunities moves; they are also adjusting based on what their rivals are doing.

Iris: I wonder if we're talking about how firms move around in a way that seems like they’re chasing opportunity, but that movement is actually influenced by strategic interactions.

Ruth: That’s right. They develop this model called ESCAPE to figure out a rational equilibrium for distribution-valued positioning when there are varying capability costs for each firm.

Warren: So what does this ESCAPE model actually do? Does it give us a concrete picture of how firms make their moves?

Ruth: It establishes a unique equilibrium, and crucially, it derives an exact reallocation restriction that separates the gain from just the opportunity shift versus the contribution from crowding.

Iris: Separating those two contributions sounds like a big step for identification. So what’s the main theoretical result they get out of this model?

Ruth: The central result is about how much revenue advantage a new position has when it moves between different economic environments, showing that it depends on the environment.

Warren: That makes sense because if the advantage changes depending on where you are, then what we think is optimal today might not be optimal tomorrow.

Iris: And they show that rational firms can even enter a region that’s getting more crowded, because the full gain from moving is positive while the crowding contribution is negative.

Ruth: That's a really interesting point, because it means you don't always have to stay in less crowded spots to make good moves.

Title and authors: Warren: But what about how much rivalry actually matters in practice? The paper mentions something about the relative response component taking up ninety point five percent of the weighted squared composition-response contrast between strategic and independent firms.

Iris: That number is pretty striking, especially since it says that this contrast is numerically zero if the capabilities are identical. So what’s that saying about how we measure competition?

Ruth: It's suggesting that you can still distinguish this interaction when you look at corpus-scale measurement, even if the underlying capabilities of the firms are the same.

Warren: So it seems like we need a certain level of data volume to see these strategic effects in action. How does this relate to what we saw in corporate news evidence about peer structure?

Iris: The paper’s observation experiment documents that peer structure persists even when firm positions move significantly, showing that optimized peer reconstructions retain eighty-six point five percent of their aggregate validation advantage one year later, even as the distance to both frozen reconstructions gets larger.

Ruth: That persistence is telling. It suggests that what we call similarity and peer relevance isn't just about how close you are physically or in terms of absolute position.

Warren: It implies that the usefulness of a peer might be state-dependent, meaning it changes depending on the current economic situation, rather than being a fixed property.

Iris: And this distinction is what separates shared movement from changes in relative economic structure for financial analysis, which is something we need to keep in mind.

Ruth: So we’re looking at how the paper suggests that when we look at these language model representations of firm positions, there are specific ways to measure this separation between absolute and relative movement.

Warren: Before we get into those technical details, let's talk about what the paper actually suggests for improving on this framework. What are their main takeaways for future research?

Iris: They suggest a few things, like using a squared composition contrast to distinguish between absolute movement of firm positions and changes in the relative economic structure.

Title and authors: Ruth: That’s a concrete way to test that separation, which accounts for ninety point five percent of the total contrast in heterogeneous capabilities. It helps analysts separate shared movement from changes in relative economic structure.

Warren: And another thing they highlight is identifying when a common opportunity improvement compensates for unfavorable crowding, using that full reallocation gain being positive even when its crowding component is negative, as shown in the exact two-firm example.

Iris: That example shows that the full gain can be positive while the crowding contribution is negative, which means you can enter crowded regions if your opportunity gain is strong enough.

Ruth: So they’re pointing toward a way to identify when a firm should move into a more crowded space based on this calculation.

Warren: And finally, they point to the importance of measuring the difference between optimized and equal-weight reconstructions, which helps track that persistence in peer structure over time.

Iris: It seems like the core idea is moving beyond just looking at where firms are now and instead measuring how their relative positions are evolving against a backdrop of changing opportunities.

Ruth: The paper "Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space" gives us tools to separate the effects of opportunity shifts from the effects of strategic interaction in firm repositioning.

Warren: It suggests that understanding how firms respond differently to the same shift is a more important part of this than just tracking where they end up.

Iris: So for those listening, think about how you measure similarity—is it about absolute location or is it about relative relevance?

Ruth: Right, and that brings us to the conclusion of this discussion on the paper "Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space."

Warren: This work shows that even when firms have different capabilities, rivalry can actually lead to firms entering crowded areas if their opportunity gains are strong enough.

Iris: And it gives us a way to measure that interaction by separating the opportunity component from the crowding component.

Ruth: It’s a framework for understanding how strategic interaction changes the allocation of positions when economic opportunities are changing.

The paper's summary: Ruth: So we've seen how firms adjust their positions in this new language model space when opportunities change, and now we're going to look at what the authors actually conclude about that whole picture.

Warren: They boil it down to this idea that it’s not just about where the opportunity is moving, but how much the other guys are doing influences your move.

Ruth: Exactly. It’s all about this concept of decomposition—separating the total gain from what’s just a shift in demand versus what's due to crowding from rivals.

Warren: And they show that even if you're trying to chase a good opportunity, you can still end up in a crowded area because your opportunity gain is strong enough to overcome the negative crowding contribution.

Ruth: That’s the core result there. It means rational firms don’t have to play it perfectly safe and avoid crowds just for the sake of avoiding rivalry; sometimes entering a crowded space can be good if you're gaining value from that specific location.

Warren: But what about how much this actually matters in the real world? They find that this interaction—the relative response component—is responsible for ninety point five percent of all the contrast between strategic and independent firms when capabilities are different.

Ruth: That’s a big number. It suggests that even if we try to ignore direct competition, the way firms react strategically accounts for most of what separates them.

Warren: And they back that up with this experiment looking at corporate news evidence. They found that optimized peer reconstructions keep eighty-six point five percent of their advantage one year later, even when the actual distance to those peers is growing.

Ruth: So it shows that similarity and relevance aren't just about how close you are in absolute terms; they’re state-dependent, meaning what matters changes with the situation.

Warren: That leads into a few practical ways we can use this framework. They suggest using a squared composition contrast to separate movement based on whether it’s absolute change or relative structure change.

Ruth: And that contrast accounts for ninety point five percent of the total contrast in capabilities, which is how you try to isolate that strategic effect.

Warren: Another practical thing they point out is looking at that exact two-firm example where the full gain is positive and the crowding contribution is negative, showing when an improvement on demand actually compensates for being crowded.

Ruth: So it’s not just about maximizing the immediate opportunity; it’s about checking if that move actually makes sense when you factor in how others will react.

Warren: This paper sets up a way to measure that interaction by separating the opportunity component from the crowding component, which is super useful for understanding market dynamics.

Ruth: It really gives us a lens to look at firm repositioning where we can see if they’re just blindly following opportunity shifts or if they're making smarter strategic moves against their rivals.

Warren: But it does have limits. The authors acknowledge that this model relies on a proportional-share contest technology, which is an idealization of how revenue actually gets distributed in the real language model space.

Ruth: Right, and that’s a fair caveat—the math is clean, but the underlying economic reality might be messier than their setup allows.

Warren: So next time you look at firm positioning in these models, think about separating the gain from the shift versus the gain from crowding; that's where all this paper points us.

The paper's improvements: Iris: So we've wrapped up on what the paper found, and now we're going to look at how they suggest making this model better for real analysis.

Ruth: They point out a couple of ways to improve the framework, specifically focusing on how we measure similarity and structure over time.

Warren: They suggest using a squared composition contrast to tell the difference between absolute movement of positions and changes in the relative economic structure.

Iris: That’s useful because it lets us isolate that strategic effect, and they say this contrast accounts for ninety point five percent of the total contrast in capabilities.

Ruth: It’s about getting a cleaner signal, so we can separate shared movement from changes in how things are actually related between firms.

Warren: Then there’s the idea of tracking the persistence of peer information more carefully. They suggest measuring the difference between optimized and equal-weight reconstructions over time.

Iris: So it’s not just about looking at where a firm is now, but seeing how that peer structure holds up even when you move further away from those frozen positions.

Ruth: That persistence suggests that similarity isn't fixed; it depends on the current economic state, which is a big shift for how we think about competitive relevance.

Warren: And they also stress the importance of identifying when a common opportunity improvement actually compensates for unfavorable crowding in that exact two-firm example we talked about earlier.

Iris: That helps us figure out exactly when a firm should move into a crowded region, based on whether the gain from that specific location is worth it.

Ruth: It gives us a tool to check if the opportunity shift is strong enough to overcome the crowding penalty in real-time analysis.

Warren: So these improvements are basically about giving analysts better tools to disentangle the effects of opportunity shifts from pure strategic interaction over time.

Iris: It’s about moving beyond just observation and building a way to measure that evolution more precisely.

Ruth: If we can get this decomposition right, it gives us a much clearer picture of what drives firm behavior in these complex spaces.

Warren: Before we look at those technical details, keep in mind that the model is still based on this proportional-share contest technology, which is an idealization of how revenue actually gets distributed.

Iris: Right, and that’s a limitation—the math works great for this structure, but we have to remember it doesn't perfectly map onto every possible market mechanism.

Conclusion: Tom: So we’ve gone through the math and the results of "Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space," and now we need to wrap up what this means for us.

Iris: Basically, they showed how you can separate the gain from just shifting demand versus what's due to crowding from rivals, which is a pretty clean way to see the interaction.

Ruth: It really puts a framework on how firms make decisions when opportunities are changing but rivalry keeps them in check.

Warren: The implication for trading is that you can start thinking about whether your move is just chasing a better opportunity or if it’s actually a strategic response to where the competition is heading.

Ruth: They prove that even when you’re trying to optimize your position, crowding doesn't always mean you have to stay put; sometimes the full gain from moving can be positive even if the crowding component is negative.

Warren: That’s a practical thing for risk management because it changes how we calculate potential drawdowns when we look at strategic moves.

Iris: I just want to check on the identification again—the core variation they are using is that decomposition of total gain into opportunity and crowding contributions, which seems solid for testing this claim.

Ruth: It is solid, but remember that the model itself assumes a proportional-share contest technology, so we have to keep that in mind when applying it to real markets.

Warren: Yeah, the scaling matters; we’re looking at how these effects behave as you look at larger and larger measurement records.

Iris: That’s true—the authors themselves flagged that the model doesn't capture every nuance of how actual revenue is distributed in practice, just what happens under their simplified setting.

Ruth: So to sum up, "Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space" gives us a way to measure the difference between absolute movement and changes in relative structure using that squared composition contrast.

Warren: It’s a tool for understanding why firms move the way they do, separating the noise from the signal of strategic interaction.

Iris: I think it provides a necessary check on whether we are just looking at correlation or actual causal effects in these competitive environments.

Ruth: That’s right. It’s a solid piece of work for understanding how competition shapes positioning in complex systems like language models.

Warren: Moving on, next week we're looking at something totally different—how geometric observables help us detect financial regimes, which is a completely different way to look at market structure.

University of Cape Town

q-fin.CP, econ.GN, q-fin.EC

Submitted: 2026-09-29

Updated: 2026-11-05

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

Importance score: 74/100

The gist: Firms follow changing economic opportunities, but rivalry changes their response <ref:2609.37820#pg2> We develop ESCAPE…

Key concepts

ESCAPE
A game theory model designed to analyze how firms choose where to operate based on changing economic opportunities and how rivals react. It incorporates heterogeneous capability costs, allowing the model to predict strategic positioning in dynamic environments.
Opportunity vs. Crowding Contribution
The total gain from repositioning is split into two parts: one measuring how demand regions change (opportunity) and another measuring how rivals' movements affect the new position's relative revenue holding opportunities (crowding). This separation helps determine if a firm can still benefit from moving into a crowded market.
Information Persistence
This concept describes how the structure of competitors remains stable over time, even as firms make large positional shifts. Evidence suggests that optimized peer reconstructions maintain most of their advantage long after absolute distance increases, indicating that relative relevance is state-dependent.

Terminology

Summary

Firms follow changing economic opportunities, but rivalry changes their response <ref:2609. p. 0

Where the paper says it

Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space Marcus Gawronsky, Chun-Sung Huang Department of Finance and Tax, University of Cape Town September 2026 Abstract Firms follow changing economic opportunities, but rivalry changes their response. We develop ESCAPE, a rational-share game of distribution-valued positioning with heterogeneous capability costs, establish a unique equilibrium, and derive an exact reallocation restriction separating opportunity and crowding contributions.

37820#pg2> We develop ESCAPE, a rational-share game of distribution-valued positioning with heterogeneous capability costs, establish a unique equilibrium, and derive an exact reallocation restriction separating opportunity and crowding contributions <ref:2609. p. 0
Where the paper says it

Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space Marcus Gawronsky, Chun-Sung Huang Department of Finance and Tax, University of Cape Town September 2026 Abstract Firms follow changing economic opportunities, but rivalry changes their response. We develop ESCAPE, a rational-share game of distribution-valued positioning with heterogeneous capability costs, establish a unique equilibrium, and derive an exact reallocation restriction separating opportunity and crowding contributions.

37820#pg3> The central theoretical result concerns how the new position’s revenue advantage changes between environments <ref:2609. p. 1
Where the paper says it

The difference between their responses measures the incremental effect of rivalry within the model. The central theoretical result concerns how the new position’s revenue advantage changes between environments.

37820#pg4> Rational firms can enter a region that is becoming more crowded, because the example’s full gain is positive while its crowding contribution is negative <ref:2609. p. 1
Where the paper says it

The resulting full reallocation gain obeys J full i ≥ 0, (1) and can be decomposed into opportunity and crowding contributions. An exact two-firm example establishes the economically central possibility: rational firms can enter a region that is becoming more crowded, because the example’s full gain is positive while its crowding contribution is negative.

37820#pg5> The relative-response component accounts for 90. p. 0
Where the paper says it

Rivalry instead changes how differently firms respond to the same opportunity shift. In the benchmark many-firm economy, the relative-response component accounts for 90.

5% of the weighted squared composition-response contrast between strategic and independent firms in the benchmark economy <ref:2609. p. 0
Where the paper says it

In the benchmark many-firm economy, the relative-response component accounts for 90. 5% of the weighted squared composition-response contrast between strategic and independent firms; it is numerically zero under identical capabilities.

37820#pg6> Corporate-news evidence documents persistent but changing peer structure, showing that optimized peer reconstructions retain 86. p. 0
Where the paper says it

The contrast remains distinguishable under corpus-scale measurement. Corporate-news distributions document persistent but changing peer structure: optimized peer reconstructions retain 86.

5% of their aggregate validation advantage one year later even as absolute distance to both frozen reconstructions increases <ref:2609. p. 0
Where the paper says it

Corporate-news distributions document persistent but changing peer structure: optimized peer reconstructions retain 86. 5% of their aggregate validation advantage one year later even as absolute distance to both frozen reconstructions increases.

37820#pg7>

Economic Model Setup

The firm’s repositioning problem involves choosing where to operate as the economic value of its opportunities changes <ref:2609.37820#pg8> The firm's payoff is defined by the expression πi(qi, q−i; rt) which incorporates the value of serving current opportunities, the cost of positioning there, and the effect of rivals on opportunity shares <ref:2609.37820#pg9> The model uses a proportional-share contest technology where revenue is allocated according to appeal shares <ref:2609. p. 5

Where the paper says it

(6) This is the unit-exponent proportional-share contest technology with positive outside appeal (Xu et al. , 2022); the ratio allocates operating revenue according to proportional appeal shares.

37820#pg10> Firms differ in their capability centers, which make nearby activities less costly at the margin, and this heterogeneity generates strategic differentiation

Decomposition of Gains

The full reallocation gain is decomposed into opportunity and crowding contributions based on the change in environment The full gain obeys J full i ≥ 0, (1) and can be separated into opportunity and crowding contributions The exact two-firm example demonstrates that rational firms can enter a crowded region because the full gain is positive while its crowding contribution is negative The opportunity component measures how the revaluation of demand regions changes that relative revenue holding opportunities fixed, and the crowding component measures how rival movement changes the new position’s relative revenue holding opportunities fixed

Information Persistence

The corpus evidence documents that peer structure persists even while firm positions move materially, with a fitted peer combination retaining most of its advantage even as both optimized and equal-weight reconstructions become more distant This persistence suggests that similarity and peer relevance are state-dependent, allowing an analyst to distinguish erosion in absolute fit from a loss of relative usefulness

Measurement Margins

The observation experiment separates economic scale and observed composition as distinct measurement margins, showing that the positive composition separation establishes that rivalry changes the allocation across activities as well as total scale The ratio of aggregate destination to validation advantages is 86.

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

The system can distinguish between absolute movement of firm positions and changes in relative economic structure by calculating a squared composition contrast, which accounts for 90.5% of the total contrast in heterogeneous capabilities. This allows financial analysts to separate shared movement from changes in relative economic structure.

2

The AI system can identify when a common opportunity improvement compensates for unfavorable crowding, as the full reallocation gain can be positive even when its crowding component is negative, which is demonstrated by the exact two-firm example.

3

The system can track the persistence of peer information by measuring the difference between optimized and equal-weight reconstructions, showing that optimized peer reconstructions retain 86.5% of their aggregate validation advantage one year later even as absolute distance to both frozen reconstructions increases.

Abstract

When firms respond to new market opportunities, their capacity choices alter the future discounted free cash flows available to both competitors and investors. In competitive markets, these investment decisions publicized in news articles and analyst reports, measuring the changes in capital structure, product mix, firm leadership, risk exposure and operating procedure made by the firm. Modern search engines rely on Large Language Models to measure the relevancy of source documents to given search query. These measurements let researchers represent these news articles as clouds of dense vectors whose information geometry reflects the dissimilarity of firms over time. Building on Xu et al. (2022), we develop a model of costly competitive adjustment in which firms allocate capacity in which firms make firm capacity adjustment towards free cash flow shocks. Using a novel sandwich estimator, we analyse the sign and magnitude of this predicted adjustment using multi-agent simulation, and validate our findings on a sample of news and analyst reports from Nasdaq-listed companies between 2018-2022.

Sources

Related papers