Final Report Summary - DYSMOIA (Dynamic Structural Economic Models: Identification and Estimation) This research project was first composed of subprojects regarding economic questions that involve inter-temporal tradeoffs for economic agents: education or learning while working, as investments for the future, consumption and savings through financial or non financial assets, asset pricing, stocking behavior of consumers etc. Investigating each of these various cases using empirical economic data involves careful modeling of preferences and expectations of economic agents in dynamic settings and carefully designed empirical strategies. In the field of earning dynamics studies, earning profiles along the life cycle can be described by three individual specific parameters which summarize individual returns and costs of investment, initial values of human capital stocks and the shadow price of human capital, later in the working life. Estimates, using a French cohort of males working in the private sector and observed from the start of their career until 30 years later (either a single cohort from 1977 to 2007 or multiple ones from 1987 onwards), help predicting well the dynamics of earnings. This allows the inequality of earnings among private sector workers at each point of their life-cycle to be described in clearer way. Starting from a U-shaped profile of inequality at the beginning of the life-cycle, it reaches a plateau after about 15 years. These results thus bring new elements to the debate since earnings inequality has been stable in France in contrast to the UK or the US over this period. The structural model can be also extended in order to understand certain forms of attrition (exits and reentries) that plague studies of earnings dynamicsIn a second successful strand of the project, we study allocation mechanisms of students to schools or colleges. Our data are diverse since they are collected in Beijing, high school in Paris, a Federal university in Northeastern Brazil or master students in Toulouse and use diverse mechanisms such as immediate acceptance or deferred acceptance with truncation. We investigate how these mechanisms affect the selection of students and their welfare. We use a theoretical setting known as college choice in the economic literature and which has lead recently to various proposals of attractive allocation mechanisms. Our aim is to prolong this literature in mechanism design in an empirical setting. Our statistical framework generally consists in a standard discrete choice models for student or college preferences and our estimation procedures rely on a careful modeling of expectations in a dynamic game. Using these estimates, we can then compute counterfactuals that compare the effect that truncation or other application costs have on congestion costs and the quality of the matching of students to colleges. This leads to policy recommendations for policy makers regarding the second-best optimality criteria helping to choose between allocation mechanisms.New data has also be generated by the DYSMOIA project to better understand these economic questions. The PI and his team collected data among high school students and their choice of majors at University in the new allocation system in Brazil that was put in place in recent years. We also collected and analyzed experimental data at the University of Toulouse among graduate students, regarding their choice of a second year master program, by using various popular mechanisms. Finally, we gathered data from undergraduate students in the same University in Brazil as well as in Toulouse to estimate learning models which allow the rules used by the University for course failure or grading to be evaluated. This is part of a recent push forward in empirical economics to marry the experimental approach, that allows experimental variation to be obtained and structural analyzes that complements it. By using a small number of experimental points, structural analysis is founded on extra- and interpolation and this allows economic analyzes to be sturdier.Last but not least, econometric estimation methods can be designed by adapting the standard set of statistical tools used in panel data or cross-section set-ups. It is also of interest to design new methods that fit better to the type of data we collect or that are more flexible in terms of the models we estimate. This is why part of this research investigates issues in partial identification arising because of using data censored by intervals or issues in non-parametric estimation that leads to imposing minimal requirements on functional forms.
