Bayesian vs. No-Regret Learners in Asset Markets
A study on arXiv (2502.08597) compares Bayesian and no-regret learning agents in asset markets with stochastic payoffs. It bridges survival and market dominance concepts from economics with regret minimization theory. Key finding: regret is crucial in market selection, but low regret alone does not ensure survival; an agent with logarithmic regret can be driven out by a Bayesian learner with a finite prior that includes the correct model. Bayesian learning is fragile, while no-regret learning is more robust and requires less environmental knowledge.
Key facts
- Study compares Bayesian and no-regret learners in asset markets.
- Regret plays a key role in market selection.
- Low regret does not guarantee survival.
- An agent with logarithmic regret can be eliminated by a Bayesian learner with a correct model prior.
- Bayesian learning is highly fragile.
- No-regret learning is more robust and requires less knowledge.
- The study bridges survival and regret minimization theories.
- Published on arXiv with ID 2502.08597.
Entities
Institutions
- arXiv