UCLA

This dissertation contains three essays in applied microeconomics. The first chapter evaluates

the welfare costs induced by limiting the number of choices in deferred acceptance

mechanisms. I show that when the number of choices is capped, some students have to be

strategic and that increasing the size of the submittable list can result in better matches,

and therefore lead to welfare improvement. I use Iranian college entrance dataset to estimate

a novel discrete choice model for centralized university systems, in which I relax the

independence of unobserved preference shocks assumption. I validate the model with out of sample data from a quasi-experimental policy change, in which the list cap was increased

by 50 percent. In my counterfactual analysis, I calculate that a list cap of 10 choices instead

of 100 would incur a 14.2 percent welfare loss. This is equivalent to a 453 km increase in

the home-university distance, which is 2.6 times the average distance traveled by Iranian

students. I also show that a more restrictive list cap does not affect students at the top and

bottom of the ranking, but hurts students with average scores and benefits students in the

lower quartile. In the second chapter, I use the aforementioned dataset to find determinants

of major choice. I estimate a rank ordered logit model of major choice and show that labor

market variables, specifically earnings and unemployment play a significant role in choice

of majors by students. The model shows that students prefer majors with higher expected

income and expected employment rate. This study also suggests that many students care more about the school they are applying to, rather than the major. Several explanations is

possible, for example prestige of some schools might be one reason. Credit constraints that

families face or the cultural barriers might also play a role for those students who prefer to

stay in their hometown even at the price of studying a major they are not very interested

in. Finally, in the third chapter I use neural networks to predict the number of quarters

that it takes a student with certain characteristics to graduate from UCLA. I also define a

survival model, in such those who did graduate before sixth year were survivors and those

who couldn't were the failures.