Banca de QUALIFICAÇÃO: Matheus José Silva de Souza

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : Matheus José Silva de Souza
DATE: 09/03/2023
TIME: 20:00
LOCAL: Vídeo Conferência
TITLE:

Recognizing Economic Behavior

KEY WORDS:

Economics, Machine Learning, Model Evaluation, Artificial Neural Networks, Aspirations, Preferences over Menus, Random Choice

PAGES: 99
BIG AREA: Ciências Sociais Aplicadas
AREA: Economia
SUMMARY:

This work comprises three independent chapters that aims to contribute to the understanding of human behavior in terms of revealed preference. The first chapter covers a literature review on economics theoretical modeling progress as models are tested on real world environments, focusing on the relationships that arise between theoretical models and experimental tools, namely econometrics and machine learning (ML), which ultimately should be seen as tools for economics and not the actual target, as stated by Goodhart’s law. This chapter, then, serves as a work to guide one on the discovering of the roots of economic modeling and gives an up-to-date on the most recent tools applied to data-driven studies. The second chapter propose a meaningful way to use ML algorithms to evaluate economic models on data, using restrictiveness and completeness measures, related to the ability of a model to explain data due to its potential on identifying structure or due to its looseness to be compatible with many data sets. We find out that artificial neural networks (ANN), in particular multi-layer perceptron (MLP), even trained with a small balanced data set, seems to be a promising way to point out behavior structure of data under the light of a selected range of economic models. Furthermore, we find out that imposing reflexiveness axiom to data plays an important role when one is willing to identify the its underlying structure and completeness measure can be used to bridge deterministic and stochastic models, enabling to evaluate the joint potential of two of these models to understand underlying preferences and uncertainty of data together. The last chapter models how agents update their prior beliefs, represented by a Random Choice Rule (RCR) with a Finite Random Expected Utility (FREU) representation, as new information becomes known. It also shows that Random Consistency is a necessary and sufficient condition for a RCR to be an update of another after the Decision Maker learns new information and it may contract or expand her subjective state spaces. We also address the matter of unforeseen contingencies representation, presenting an extension to previous works by characterizing its opposite direction, when the subjective states of the FREU representation of a Random Choice Rule is contained in the subjective state space of the representation of a Preference Over Menus. Finally, we also present a discussion on the conditions under which a collection of Random Choice Rules represent a partition of a broader Random Choice Rule of a Preference Over Menus.

BANKING MEMBERS:
Presidente - 1642911 - DANIEL OLIVEIRA CAJUEIRO
Externo à Instituição - GIL RIELLA - FGV
Externo à Instituição - MATHEUS SCHMELING COSTA - FGV