Overview

Abstract (English)

This thesis presents three chapters that use and develop microeconometric methods for microdata analysis in economics. The first chapter studies how social interactions influence entrepreneurs’ risk-taking decisions. We conduct two risk-taking experiments with young Ugandan entrepreneurs. Between the two experiments, the entrepreneurs participate in a networking activity where they build relationships and discuss with each other. We collect data on peer network formation and on participants’ choices before and after the networking activity. We find that participants tend to make more (less) risky choices in the second experiment if the peers they discuss with make on average more (less) risky choices in the first experiment. This suggests that even short term social interactions may affect risk-taking decisions. We also find that participants who make (in)consistent choices in the experiments tend to develop relationships with individuals who also make (in)consistent choices, even when controlling for observable variables such as education and gender, suggesting that peer networks are formed according to unobservable characteristics linked to cognitive ability. The second chapter studies whether tax-preferred saving accounts policies in Canada are suited to all individuals given they different income path and given differences in tax codes across provinces. The two main forms of tax-preferred saving accounts – TEE and EET – tax savings at the contribution and withdrawal years respectively. Thus the relative returns of the two saving vehicles depend on the effective marginal tax rates in these two years, which in turn depend on earning dynamics. This chapter estimates a model of earning dynamics on a Canadian longitudinal administrative database containing millions of individuals, allowing for substantial heterogeneity in the evolution of income across income groups. The model is then used, together with a tax and credit calculator, to predict how the returns of EET and TEE vary across these groups. The results suggest that TEE accounts yield in general higher returns, especially for low-income groups. Comparing optimal saving choices predicted by the model with observed saving choices in the data suggests that EET are over-chosen, especially in the province of Quebec. These results have important implications for “nudging” policies that are currently being implemented in Quebec, forcing employers to automatically enrol their employees in savings accounts similar to EET. These could yield very low returns for low-income individuals, which are known to be the most sensitive to nudging. Finally, the third chapter is concerned with methodological problems often arising in regression discontinuity designs (RDD). It considers the problem of rounding errors in the running variable of RDD, which often make the treatment variable unobservable for some observations around the threshold. While researchers usually discard these observations, I show that they contain valuable information because the outcome’s distribution splits in two as a function of the treatment effect. Integrating this information in standard data driven criteria helps in choosing the best model specification and avoid specification biases. This method is promising, especially for improving estimates of causal effects in very large database (where the number of observations discarded can be very large), such as the LAD used in Chapter 2.

Abstract (French)

Please note that abstracts only appear in the language of the publication and might not have a translation.