Experimental explorations of bounded rationality

Abstract

This thesis explores bounded rationality -the recognition that individuals, when making economic decisions, face cognitive limitations that prevent them from processing and analyzing all available information comprehensively- in three domains: probabilistic thinking, strategic choice and networked markets. In the domain of probabilistic thinking, we propose a quasi-Bayesian theory of inference that explains the overprecision bias, an excessive certainty regarding the accuracy of one's beliefs, which manifest by subjective confidence intervals that are often too narrow when compared with Bayesian ones. We run laboratory experiments to explore overprecision when people learn about the empirical frequency f of some random event. Motivated by the literature on limited attention, we hypothesize that, when there is a large number of potential values of f, individuals mentally operate with simplified representations of the objective state space. Their mental models are however sophisticated in that they co-move with the signals observed, focusing on the values of f most consistent with the evidence available. As a result, they elaborate accurate unbiased point estimates of f , but also overly narrow confidence intervals since they become too confident about them, as they hardly reflect on those values of f that would call more into doubt their conclusions. In the domain of strategic choice, we run laboratory experiments to explore off-equilibrium behavior in anchored guessing games- a class of games introduced in Ballester, Vorsatz, and Ponti (2023) that generalizes guessing games such as the p-beauty contest game-, build a level-k model that explains the empirical data and determine whether the performance of subjects in the Cognitive Reflection Test or CRT (Frederick, 2005) is a valid predictor for their behavior in this novel type of games. Several studies, (e.g., Brañas-Garza et al., 2012), have found for a variety of games that subjects with a higher CRT socre make choices that are closer to Nash equilibrium. We show theoretically, in an extended level-k framework with free subjective beliefs, that there are two factors that could contribute to subjects with a higher CRT score displaying such observable effect in anchored guessing games: (a) their iterative reasoning process could employ a higher reasoning level or (b) the starting point (or seed) for such iterative process could be more advantageously positioned (or both). We find strong evidence that subjects with a higher CRT socre make choices that are closer to Nash equilibrium than subjects with a lower CRT score, also in anchored guessing games with an interior equilibrium. We then find empirical evidence that the main factor that contributes to this fact is that they iterate more often in their reasoning process than subjects with a lower CRT score. We also find that, regardless of CRT score, subjects move to using seeds closer to their optimum calues as the experiment progresses. However, we find no clear evidence of subjects with a higher CRT score starting their iterative reasoning proccesses in seeds closer to their optimum values than subjects with a lower CRT score. Finally, through a dynamic analysis of the estimated mean level of reasoning and seed distance over the experiment, we also find evidence of a learning or adaptation process, which can be characterized by a warm-up phase (in which iubjects reduce their seed distance), followed by a learning phase (in which they increase their resoning level, at a much faster rate in subjects with a higher CRT score) and then a saturation phase in which no further improvementes are made. In the domain of networked markets, we present a research project, currently "work in progress", that explores bounded retionality in agents that engage in bargaining in networked markets and use simplified models of the ner¡twork in their decision-making process. The specific networked bargaining setting that we analyze is similar to the one described in Talamàs (2017), i.e., non-cooperative bilateral bargaining with network stationary and strategic choice of partner, but modified to have bounded rational agents that employ local models of the network in their decisions, instead of fully Bayesian agents. Our objective is to analyze how the interplay among the architecture of the trading network, the buyers' valutions, the sellers' costs, and the depth of their limited models of the network, impact price formation, price dispersion, allocations and effiency in the networked market. Since we are still in the process of obtaining jey results, we present the motivation for the project and a brief literature review on the subject of networked bargaining, we outline some theoretical and methodological considerations for conducting our research and introduce our theoretical framework, which is based on the novel game-theory solution concept of Local Bayesian Equilibrium (LBE) defined in Sadler (2015). We are currently implementing and validating this model through computational experiments that run in a custom-made networked bargaining simulator.