This paper focuses on learning efficient equilibria in repeated games. The author proposes a stochastic learning rule that selects a subgame-perfect equilibrium where players receive efficient payoffs. The paper details how this rule works, involving bounded-memory models, approximate best responses, and periodic testing of these models against observed actions. A key element is the probability of experimenting with new models being inversely related to recent payoffs. The article demonstrates that this rule selects efficient equilibria and discusses how different specifications of experimentation probabilities can lead to the Kalai–Smorodinsky and maxmin bargaining solutions.

This academic article from Econometrica, published by The Econometric Society and available on JSTOR, explores the concept of subjective equilibrium in the context of repeated games. It contrasts this idea with the more established notion of Nash equilibrium, which assumes players have accurate knowledge of their opponents' strategies. The authors, Ehud Kalai and Ehud Lehrer, propose that subjective equilibrium reflects a scenario where players optimize their strategy based on their personal beliefs about others, which may not be entirely correct. The paper focuses on infinitely repeated perfect-monitoring games and investigates the conditions under which subjective equilibrium behavior aligns with or approximates Nash equilibrium behavior.

This academic paper, published in the journal Games and Economic Behavior, discusses the possibility of Bayesian learning in infinitely repeated games. The author examines prior research suggesting limitations on rational learning in such games, particularly the difficulty of achieving both belief consistency and strategy learnability. The paper proposes a modified concept called "optimizing learnability," which focuses on the ability of players to learn optimal strategies rather than any possible strategy. The author argues that this redefined notion of learnability, when combined with ε-consistency, is sufficient for convergence to approximate Nash equilibrium, even in a broader class of games than previously thought possible. The text includes formal definitions, lemmas, and a theorem to support its claims, along with examples illustrating the application of the theory in coordination games.

This paper published in the September 1993 issue of Econometrica: Journal of the Econometric Society and available through JSTOR, presents a theoretical study on how players in infinitely repeated games with rational learning will eventually converge to playing a Nash equilibrium. The authors, Ehud Kalai and Ehud Lehrer, explore how players with subjective beliefs about their opponents' strategies, under certain conditions including absolute continuity of true strategies with respect to beliefs and perfect monitoring, will learn to predict the game's future play accurately. This Bayesian updating process, driven by each player's goal to maximize their expected utility, leads to a state where the players' actions approximate an epsilon-Nash equilibrium of the repeated game. The paper contrasts this approach with models assuming bounded rationality or myopic play, highlighting the importance of players knowing their own payoff matrices and having beliefs compatible with the truth.

keepSave to notecopy_alldocsAdd noteaudio_magic_eraserAudio OverviewflowchartMind Map

These sources discuss QALIGN, a new technique for aligning Large Language Model (LLM) outputs with desired preferences at the time of generating the response, rather than requiring expensive prior training or fine-tuning of the model weights. The core idea is to use Markov chain Monte Carlo (MCMC) sampling, guided by an external reward model (RM), to iteratively refine outputs and converge towards a preferred distribution. This approach offers significant resource efficiency compared to traditional methods like Reinforcement Learning from Human Feedback (RLHF) or Direct Preference Optimization (DPO), potentially making advanced alignment research more accessible, particularly in environments like business schools. The text highlights QALIGN's ability to leverage human-in-the-loop or AI feedback via RMs and shifts the research focus from model architecture to the crucial areas of reward model design and evaluation. Experiments demonstrate that QALIGN can outperform existing test-time methods like best-of-n (BoN) and majority voting (MV), showing consistent performance improvements as computational resources increase, unlike methods that can degrade from over-optimizing imperfect RMs.

This paper discusses a high-dimensional choice model for online retailers to address the challenge of understanding substitution patterns among a large number of products, which is difficult with traditional models. By leveraging consumer clickstream data and combining econometric and machine learning methods, specifically the graphical lasso technique, the authors aim to learn flexible substitution patterns based on product utility shocks. They demonstrate through synthetic data and a real-world empirical setting that their method provides more accurate demand forecasts and significantly improves the precision of estimating price elasticities, crucial inputs for operational decisions like inventory management, assortment planning, and pricing strategies. The research also highlights the value of integrating machine learning with econometrics and utilizing clickstream data in retail operations.

This paper describes the development and evaluation of a novel method for estimating parameters in structural econometric models using pretrained neural networks. The core idea is to pre-train an estimator using a large number of datasets simulated from a specific structural model, enabling rapid and accurate estimation on new datasets with minimal computational cost and researcher effort. The authors demonstrate this approach with a sequential search model, showcasing that the pretrained estimator achieves comparable or superior accuracy to traditional methods like Simulated Maximum Likelihood Estimation (SMLE) while being significantly faster. The research highlights the potential for off-the-shelf estimators to increase the accessibility of structural models and facilitate their integration into real-time algorithms.

This paper investigates how political ideology influences a buyer's likelihood of negotiating prices, particularly in informal market settings like purchasing houses or used cars. Through both archival data analysis of real estate transactions and Google search trends, and controlled laboratory studies, the research demonstrates that conservative buyers exhibit a greater propensity to negotiate compared to their liberal counterparts. This difference is attributed to conservatives' stronger endorsement of free-market ideology, which is shown to mediate the relationship between political leaning and negotiation behavior. Furthermore, the studies suggest that conservatives tend to moralize the act of price negotiation and are more likely to negotiate regardless of the seller's identity or the potential outcome, supporting the notion that these behaviors are driven by ideological beliefs beyond simple economic gain. The findings highlight that marketplace actions are shaped by both economic considerations and deeply held ideological convictions.

We explore several alternative perspectives on why individuals in a household might seek their partner's input when deciding to dispose of items, suggesting the reasons go beyond simple ownership rights and valuation. It proposes that factors such as perceived knowledge gaps about a partner's use or attachment to an item, the desire to manage interpersonal risk and avoid blame for removing something important to the partner, and the need to overcome the inertia associated with disrupting the status quo of the shared living space all contribute to the need for joint decision-making. Finally, the text considers how the act of disposal may more strongly activate communal norms related to the shared household environment than the act of acquiring something new.

This paper explores the financial implications of new product preannouncements for firms, particularly within the software and hardware industries. It leverages agency and signaling theories, along with rational learning theory, to examine how these announcements influence shareholder value. The research finds that while short-term abnormal returns are only positive when the preannouncement provides specific information about the product, long-term abnormal returns are significantly positive, especially when firms provide ongoing updates and possess high reliability in delivering on past preannouncements. The study employs the calendar-time portfolio methodology to analyze these long-term effects, offering practical guidelines for managers on managing communications to maximize financial value from preannouncing products.