Identification and Estimation of a Multi-Armed Bandit Model of Subjective Expectation Formation
(Identificação e Estimação de um Modelo Multi-Armed Bandit de Formação de Expectativas Subjetivas)
Bandit Models & Subjective Expectations
CNPq, Brazil
OBJECTIVE
The main objective of my research proposal is to develop a structure of econometric modeling (both identication and estimation) that can be applied to the issue of student's choice of university major in Brazilian universities. It is intended to adapt models like Delavande and Zafar (2014), Arcidiacono, Hotz, Maurel, and Romano (2014), and Estrada and Gignoux (2014) for the case where high school students are about to choose their university major. Actually, differently, from countries like the United States, Brazilian students choose ex-ante, with a very low probability of posterior change, their university major).
By "adaptation" of existing models we mean that we will contribute to the stock of existing analytical tools into two directions, say, by elicitation and modelization of probabilities of counterfactual choices as well as by including a formal process of belief revision of newly (exogenously and endogenously) learned information where agents choose optimally what type of information to acquire as well as how much of it. Indeed, as far as we know there is no past attempt to build an estimable model with such characteristics.
As a secondary objective we will apply our model to a newly collected microdata data set (PAEST - Pesquisa de Aspiraçõoes e Expectativas dos Estudantes Concludentes do Ensino Médio - Survey on Aspirations and Expectations of High School Students) idealized and collected by myself, Prof. Thierry Magnac (University of Toulouse) and other members of our research group with a budget from the European Community.
This data set has been instrumental to understand important issues about high school student choice of undergraduate major as well as the role of the SISU (Unied Selection System), especially the socioeconomic circumstances surrounding those about to make undergraduate major choices during their high school last year. Importantly, this CNPq's project proposal intends to continue a joint research agenda (European Research Commission Advanced Grant 2011 - Project: College or school choice: matching mechanisms) initiated in 2012 in partnership with Prof. Thierry Magnac, which aims to build up a dynamic model of choice of university major and of grading as well as effort exerted to be successful.
Indeed, as exams have both a dimension of selection of the most talented (although the selection is imperfect) but also of those who exert most effort, we face a tricky analytical issue of identication and estimation of micro determinants of academic access and performance. At first, the agenda focused on the impacts of selection mechanisms (vestibular) the performance of the university. It was developed and estimated a model of selection of entries in Brazilian universities. Now, with better data from PAEST we can further our analysis to understand schooling choices. The proposed CNPq's research project proposal is an offshoot of that initial agenda and they stand to deepen knowledge about the functioning of the "black box" of the academic production function.