Learning, experimentation and preference discovery

The mainstream approach to decision theory adopted by economists assumes what Herbert Simon calls the substantive rationality of individuals, that is, it assumes rational behavior as the optimal outcome for achieving goals in given circumstances. This approach distinguishes the economic view of decision theory from the psychological view, which assumes a weaker notion of procedural rationality, which sees behavior as rational as long as it is the result of appropriate mental deliberation. Although the classical approach of economists to decision theory, assuming the substantive rationality of the consumer, has considerable descriptive and predictive power, it comes at the cost of many behavioral paradoxes. Some paradoxes of choice, such as the phenomenon of preference reversal, have led to the complete rejection of the existence of preferences by many psychological theories. Many theories follow this rejection of maximizing behavior, including preference construction theories. The theory of preference discovery, developed by Charles Plott, takes a middle ground position between economics and psychology on this issue. He states that the agent's rationality evolves with his experience, from the first stage where, in the absence of experience in a given choice task, the observed choices are chaotic, because the agent, even if rational, experiments in order to learn his own preferences. In the second stage, individual preferences are discovered, but the perception of other actors' behavior as rational is still missing, leading to the persistence of paradoxes when interactions between actors occur. Finally, in the third stage, the perception of others as rational also begins, and the individual behaves as substantively rational. The theory of preference discovery claims that there are well-defined and stable preferences, but nevertheless they can lead to unstable observed choices if the individual is unaware of them. The main observation on which this theory is built is that many behavioral paradoxes, including preference reversal and the possession effect, disappear in repeated experiments in the presence of stimuli. The main purpose of this study is to identify the preference discovery process. We achieve this goal by constructing a theoretical model. We distinguish three specific research objectives, which together make up the main objective of the study. 1) identification of the process and determinants of preferences discovered by experimental consumption, 2) identification of the processes, methods and properties of learning behaviors of the consumer discovering his own preferences, and 3) identification of the testable limitations of the model, from the perspective of the revealed preferences approach. The study focuses on the construction of a theoretical model of preference discovery. We consider a consumer with unknown preferences that are discovered through the consumption experience. At the same time, we indicate the source of uncertainty, so we assume an objective, not a subjective, state space. The subjective state space is often considered advantageous because states of introspective uncertainty are usually unobservable. However, in relation to the discovery of preferences and learning through consumer experience, we consider the subjectivity of the state space to be an unnecessary requirement, because without loss we can treat states as corresponding to all possible real relations of consumer preferences, with relations introspectively treated by the consumer as impossible represented as empty states. With the exogenous state space defined in this way, we can study preference discovery experiments as unrelated to flexibility, which is the dominant approach in the literature. As in most modern research, we believe that the consumer is formulating subjective probabilities over a space of potential preference relationships. However, by using the exogenous state space, we are able to determine the consumer's subjective probability in a more direct way. In addition, we introduce the concept of similarity of alternatives as a primitive of the model. We believe that this facilitates a more detailed study of indirect learning than is possible in contemporary work. In our setting, consumption can therefore additionally inform the consumer about the relative ranking of alternatives based on their similarity to those already experienced, allowing for a more accurate identification of the theoretical properties of intermediate learning. Dominant approaches also fail to account for systemic errors in updating subjective probability. In the study, we consider a more general situation where noisy information as well as under- or over-reaction to it is acceptable, in order to further investigate consumer learning behavior. Together with the proposed more direct approach, this allows for a better identification of concepts such as the value of information and its shape, e.g. concavity to experience. Also from the perspective of the revealed preferences approach, we propose to study the testable model constraints and invoking parameters with an approach similar to that adopted in research from the trend of dynamic stochastic utility.