Paper: arXiv 2411.05032
Authors: Aleksei Pastushkov
Abstract
We examine the dynamics of informational efficiency in a market with asymmetrically informed, boundedly rational traders who adaptively learn optimal strategies using simple multiarmed bandit (MAB) algorithms. The strategies available to the traders have two dimensions: on the one hand, the traders must endogenously choose whether to acquire a costly information signal, on the other, they must determine how aggressively they trade by choosing the share of their wealth to be invested in the risky asset. Our study contributes to two strands of literature: the literature comparing the effects of competitive and strategic behavior on asset price efficiency under costly information as well as the actively growing literature on algorithmic tacit collusion and pseudo-collusion in financial markets. We find that for certain market environments (with low information costs) our model reproduces the results of Kyle [“1989”] in that the ability of traders to trade strategically leads to worse price efficiency compared to the purely competitive case. For other environments (with high information costs), on the other hand, our results show that a market with strategically acting traders can be more efficient than a purely competitive one. Furthermore, we obtain novel results on the ability of independently learning traders to coordinate on a pseudo-collusive behavior, leading to non-competitive pricing. Contrary to some recent contributions (see e.g. [“Cartea et al. 2022”]), we find that the pseudo-collusive behavior in our model is robust to a large number of agents, demonstrating that even in the setting of financial markets with a large number of independently learning traders non-competitive pricing and pseudo-collusive behavior can frequently arise.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 3.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper employs advanced mathematical modeling based on multi-armed bandit algorithms and strategic game theory to analyze market dynamics, but it relies entirely on theoretical modeling and agent-based simulations without any backtesting on real market data or production-level code implementation.
Research Flowchart
flowchart TD Start["Research Goal"] --> Methodology["Methodology: Adaptive MAB Agents"] Methodology --> Inputs["Data/Inputs: Costly Info & Wealth Allocation"] Inputs --> Computation["Computational Process: Multi-Agent Simulation"] Computation --> Comparison["Comparison: Competitive vs. Strategic"] Comparison --> Outcomes["Key Findings"] Outcomes --> F1["Efficiency depends on Info Cost"] Outcomes --> F2["Pseudo-collusion & Non-competitive Pricing"] Outcomes --> F3["Robustness to Large N Agents"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes an agent‑based model in which asymmetrically informed, boundedly rational traders adaptively learn optimal strategies using multi‑armed bandit (UCB) algorithms, choosing both whether to acquire a costly information signal and how aggressively to invest in the risky asset.
Source sentence
We examine the dynamics of informational efficiency in a market with asymmetrically informed, boundedly rational traders who adaptively learn optimal strategies using simple multiarmed bandit (MAB) algorithms. The strategies available to the traders have two dimensions: on the one hand, the traders must endogenously choose whether to acquire a costly information signal, on the other, they must determine how
Assumes 6
- All traders are risk‑neutral. model
Source sentence
As by assumption all traders are risk-neutral in our model, a trader is willing to invest in the risky 5 Which can be thought of as either a stock or a derivative contract.
- Acquiring the information signal incurs a fixed cost C and perfectly reveals the next period payoff. model
Source sentence
The first dimension is whether to become informed by acquiring a costly information signal at cost C. If the trader i becomes informed, their expectation of the next period’s payoff of the risky asset is equal to its actual value, i.e. the information signal perfectly reveals the value of the next period’s payoff: 𝐸𝑖𝑡(𝐹𝑡+1) = 𝐹𝑡+1 (2) If the trader chooses to become uninformed,
- Uninformed traders base expectations on the costlessly observable historical average return. model
Source sentence
The first dimension is whether to become informed by acquiring a costly information signal at cost C. If the trader i becomes informed, their expectation of the next period’s payoff of the risky asset is equal to its actual value, i.e. the information signal perfectly reveals the value of the next period’s payoff: 𝐸𝑖𝑡(𝐹𝑡+1) = 𝐹𝑡+1 (2) If the trader chooses to become uninformed, their expectation of 𝐹𝑡+1 is based on
- The supply of the risky asset is fixed at N shares each period and is provided by a liquidity‑seeking representative trader. market
Source sentence
The individual schedules are aggregated by the market venue to an aggregate demand schedule. The supply of N shares can be 8 thought of as provided by a liquidity-seeking trader who is insensitive to the “fundamental” payoff value 𝐹𝑡.
Show 2 more
- Traders use a stateless Upper Confidence Bound (UCB) multi‑armed bandit learning algorithm with only one tunable parameter. model
Source sentence
In contrast, the MAB learning algorithm we employ, and particularly its Upper Confidence Bound (UCB) implementation, has only 1 parameter.
- The simulation includes a moderately large number of agents (M = 100). model
Source sentence
An additional innovation pursued in this study is that our model includes a moderately large number of agents (100), whereas prior literature on algorithmic collusion in financial markets typically models interactions of less than 10, and in many cases only 2-4 agents.
Data
Synthetic data only.
Costs
Not applicable
Finds 5
- For low information costs the model reproduces Kyle (1989) results: strategic trading leads to worse price efficiency than the purely competitive case. theoretical
Source sentence
We find that for certain market environments (with low information costs) our model reproduces the results of Kyle [1989] in that the ability of traders to trade strategically leads to worse price efficiency compared to the purely competitive case.
- For high information costs the strategically acting market can be more efficient than a purely competitive one. theoretical
Source sentence
For other environments (with high information costs), on the other hand, our results show that a market with strategically acting traders can be more efficient than a purely competitive one.
- Independently learning traders can coordinate on pseudo‑collusive behavior, resulting in non‑competitive pricing. theoretical
Source sentence
Furthermore, we obtain novel results on the ability of independently learning traders to coordinate on a pseudo-collusive behavior, leading to non-competitive pricing. Contrary to some recent contributions (see e.g. [Cartea et al. 2022]), we find that the pseudo-collusive behavior in our model is robust to a large number of agents, demonstrating that even in the setting of financial markets with a large number of
- Pricing is more competitive in the high informational‑cost regime, with the price line closer to the efficiency upper bound. performance
Source sentence
Notably, the pricing is more competitive in the high informational cost regime (Figures 12-14), as the blue line on average is closer to the upper bound.
- Non‑competitive pricing persists even with a relatively large number of UCB‑driven agents (M = 100), contrary to prior findings for smaller agent sets. negative
Source sentence
In our model, non-competitive pricing persists despite there being a relatively large number of competing, UCB-driven agents 18 (M=100).
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.