Research

Work in progress

CautionThe Timing of Academic Specialization: Impacts on Student Outcomes

Joint work with Alexei Karas

In many educational systems, including most of Europe, students apply to college by choosing a specific degree program, and have little control over the composition of their curriculum beyond this choice. In most US colleges — especially in Liberal Arts and Sciences (LAS) programs that focus on interdisciplinary education — students do not have to commit to a discipline at enrollment. They can explore their interests by taking courses in diverse fields, and can change the direction of their studies multiple times without having to transfer to a different program and risk losing study credits. While this exploration stage may lead to better match quality between students and their chosen fields of study, a less coherent and focused curriculum may reduce the alignment of academic records and future educational or occupational aspirations. Understanding this trade-off in how the timing of academic specialization affects academic and labor market outcomes can be valuable for optimizing undergraduate program structures.

ImportantSchool Choice, School Switching, and Optimal Assignment

Joint work with Bas van der Klaauw and Hessel Oosterbeek

Close to 20% of secondary school students in Amsterdam – and elsewhere – transfer to another secondary school at some point, even when initially placed in their top-ranked school. Such switches are costly for the students involved and disrupts the learning environment of their former and new classmates. This study investigates whether these transfers are the result of initial school choices that are “hard-to-rationalize”. Using data from the Amsterdam secondary school match linked to administrative registers, we estimate models of school demand and switching behavior. Our results show that switching is indeed predicted by “hard-to-rationalize” initial school choices, which we quantify by a low predicted probability of choosing the preferred school. Over 60% of switchers can be correctly identified at the admission stage. Simulations suggest that informing predicted switchers and encouraging them to adjust their initial preferences can reduce the average expected switching probability by up to 15%.

TipBank Failure Prediction: Supplementing Standard Approaches with the Thematic Analysis of Reasons of Bank Closure

Joint work with Alexei Karas

We combine textual and quantitative evidence to study bank closures in Russia. From Central Bank of Russia (CBR) press releases, we identify four failure themes: self-termination (S), transparency violations (T), mismanagement or fraud (M), and balance-sheet distress (CAEL). We validate this typology by showing that thematic spikes track major macroeconomic shocks, regulatory reforms, and changes in CBR leadership. Incorporating the validated typology into logistic regression and random forests failure prediction models demonstrates that closure type is central to predictability. At a one-month horizon, random forests accuracy increases from 0.78 to 0.96 when S-, T-, and M-type failures are excluded, and logistic regression accuracy rises from 0.73 to 0.86. Gains persist, though attenuate, at longer horizons, and S/T/M subsets remain the least predictable.

Thesis work

Rozsos, T. (2024). Switch happens: were school switchers initially assigned to the wrong schools?. MPhil thesis supervised by Prof. Dr. Bas van der Klaauw (VU), Prof. Dr. Hessel Oosterbeek (UvA).

Rozsos, T. (2022). Developing bank failure prediction models: Exploring the value of failure type heterogeneity. Bachelor’s thesis supervised by Dr. Alexei Karas (UCR).