No full-text available
To read the full-text of this research,
you can request a copy directly from the author.
Learning with Regret
Abstract
Choices in economic games are predicted better by models that look back at what might have been, instead of looking forward to maximum gain.
... Research in computational models incorporating the regret signal is becoming important (Cohen, 2008;Marchiori and Warglien, 2008). Most of the current models use predictionerror signal for learning the internal parameters of the system. ...
NeuroEconomics is a rapidly developing multidisciplinary research area that employs neuroscience techniques to explain economic theories of human behaviour and makes efforts to bring researchers from the disciplines of psychology, neuroscience, and economics to a common platform. The foundations of this science lie in the pioneering experiments of Kahneman and Tversky (1979). The past decade has seen a remarkable effort by neuroscientists and economists in applying neurophysiological (animal experiments recoding single cell behaviour) and neuroimaging (such as event related potentials, functional magnetic resonance imaging) techniques to understand brain activity while experimental tasks that involve economic decisions are being performed. In this review we would like to present studies from single cell recording to behavioural level while presenting various components related to the process of individual decisions. Along with the future scope of computational modelling, the existing models related to individual decision making will be presented. Taking further, the upcoming research area of collective decision making will be reviewed.
Game theory is an interdisciplinary approach to the study of human behavior. Games describe a widely accepted framework for
representing interactive decision-making. Artificial Neural Networks (ANNs) are universal approximators and have the ability
of learning. Combining ANNs with game representation, we introduced a new architecture by which the learning abilities of
ANNs are utilized to predict game behavior. Based on previous work, we investigated further the potential value of neural
networks for modeling and predicting human interactive learning in repeated games. We conducted simulation studies based on
the new model using experiments data which are provided by authors other than this paper. Through computer simulations and
comparing with other models, we demonstrated that our model is superior in many respects to other models on ten experiments.
The sealed-bid first-price auction of a single object in the case of independent privately-known values is the simplest auction setting and understanding it is important for understanding more complex mechanisms. But bidders bid above the risk-neutral Nash equilibrium theory prediction. The reasons for this "over bidding" remain an unsolved puzzle. Several explanations have been offered, including risk aversion, social comparisons, and learning. We present a new explanation based on regret and a model that explains not only the observed over bidding in sealed-bid first-price auctions, but also behavior in several other settings that is inconsistent with risk aversion.
There is a good deal of miscommunication among experimenters and theorists about how to evaluate a theory that can be rejected
by sufficient data, but may nevertheless be a useful approximation. A standard experimental design reports whether a general
theory can be rejected on an informative test case. This paper, in contrast, reports an experiment designed to meaningfully
pose the question: “how good an approximation does a theory provide on average.” It focuses on a class of randomly selected
games, and estimates how many pairs of experimental subjects would have to be observed playing a previously unexamined game
before the mean of the experimental observations would provide a better prediction than the theory about the behavior of a
new pair of subjects playing this game. We call this quantity the model’s equivalent number of observations, and explore its
properties.
Much of human learning in a social context has an interactive nature: What an individual learns is affected by what other individuals are learning at the same time. Games represent a widely accepted paradigm for representing interactive decision-making. We explored the potential value of neural networks for modeling and predicting human interactive learning in repeated games. We found that even very simple learning networks, driven by regret-based feedback, accurately predict observed human behavior in different experiments on 21 games with unique equilibria in mixed strategies. Introducing regret in the feedback dramatically improved the performance of the neural network. We show that regret-based models provide better predictions of learning than established economic models.
Human decisions cannot be explained solely by rational imperatives but are strongly influenced by emotion. Theoretical and behavioral studies provide a sound empirical basis to the impact of the emotion of regret in guiding choice behavior. Recent neuropsychological and neuroimaging data have stressed the fundamental role of the orbitofrontal cortex in mediating the experience of regret. Functional magnetic resonance imaging data indicate that reactivation of activity within the orbitofrontal cortex and amygdala occurring during the phase of choice, when the brain is anticipating possible future consequences of decisions, characterizes the anticipation of regret. In turn, these patterns reflect learning based on cumulative emotional experience. Moreover, affective consequences can induce specific mechanisms of cognitive control of the choice processes, involving reinforcement or avoidance of the experienced behavior.