arXiv subjects

Aroon Narayanan

Publications and source records attributed to Aroon Narayanan.

5 recordsLinked to original sources

arXiv2023

Risk Preferences of Learning Algorithms

Agents' learning from feedback shapes economic outcomes, and many economic decision-makers today employ learning algorithms to make consequential choices. This note shows that a widely used learning algorithm, $\varepsilon$-Greedy, exhibits emergent risk aversion: it prefers actions with lower variance. When presented with actions of the same expectation, under a wide range of conditions, $\varepsilon$-Greedy chooses the lower-variance action with probability approaching one. This emergent preference can have wide-ranging consequences, ranging from concerns about fairness to homogenization, and holds transiently even when the riskier action has a strictly higher expected payoff. We discuss two methods to correct this bias. The first method requires the algorithm to reweight data as a function of how likely the actions were to be chosen. The second requires the algorithm to have optimistic estimates of actions for which it has not collected much data. We show that risk-neutrality is restored with these corrections.

cs.LG↗

arXiv2022

Social learning via actions in bandit environments

I study a game of strategic exploration with private payoffs and public actions in a Bayesian bandit setting. In particular, I look at cascade equilibria, in which agents switch over time from the risky action to the riskless action only when they become sufficiently pessimistic. I show that these equilibria exist under some conditions and establish their salient properties. Individual exploration in these equilibria can be more or less than the single-agent level depending on whether the agents start out with a common prior or not, but the most optimistic agent always underexplores. I also show that allowing the agents to write enforceable ex-ante contracts will lead to the most ex-ante optimistic agent to buy all payoff streams, providing an explanation to the buying out of smaller start-ups by more established firms.

econ.TH↗

arXiv2021

Single-peaked domains with designer uncertainty

This paper studies single-peaked domains where the designer is uncertain about the underlying alignment according to which the domain is single-peaked. The underlying alignment is common knowledge amongst agents, but preferences are private knowledge. Thus, the state of the world has both a public and private element, with the designer uninformed of both. I first posit a relevant solution concept called implementation in mixed information equilibria, which requires Nash implementation in the public information and dominant strategy implementation in the private information given the public information. I then identify necessary and sufficient conditions for social rules to be implementable. The characterization is used to identify unanimous and anonymous implementable social rules for various belief structures of the designer, which basically boils down to picking the right rules from the large class of median rules identified by Moulin (1980), and hence this result can be seen as identifying which median rules are robust to designer uncertainty.

econ.TH↗

arXiv2021

Ex-post implementation with interdependent values

We characterize ex-post implementable allocation rules for single object auctions under quasi-linear preferences with convex interdependent value functions. We show that requiring ex-post implementability is equivalent to requiring that the allocation rule must satisfy a condition that we call eventual monotonicity (EM), which is a weakening of monotonicity, a familiar condition used to characterize dominant strategy implementation.

econ.TH↗

arXiv2017

Coverage Analysis in Millimeter Wave Cellular Networks with Reflections

The coverage probability of a user in a mmwave system depends on the availability of line-of-sight paths or reflected paths from any base station. Many prior works modelled blockages using random shape theory and analyzed the SIR distribution with and without interference. While, it is intuitive that the reflected paths do not significantly contribute to the coverage (because of longer path lengths), there are no works which provide a model and study the coverage with reflections. In this paper, we model and analyze the impact of reflectors using stochastic geometry. We observe that the reflectors have very little impact on the coverage probability.

cs.IT↗