CEU Electronic Theses and Dissertations, 2025
| Author | Menglibayev, Nurdaulet |
|---|---|
| Title | Synthetic Control Methods: Balance Between Overfitting and Underfitting with Limited Data |
| Summary | This thesis investigates the role of model complexity in the Synthetic Control Method (SCM) by systematically evaluating its impact through replications of seminal studies by Abadie et al. (2010, 2015). Using three variants of SCM with increased levels of complexity, I assess the tradeoff between underfitting and overfitting in prediction. The results demonstrate that, in these cases, more complex models can outperform the original specification, indicating the presence of underfitting in classical applications of SCM. Moreover, I find that introducing time-varying weights—an extension not justified by the classical data-generating process assumed in standard SCM—can significantly improve predictive performance. However, the most complex specification tested, which incorporated most layers of flexibility, produced consistently worse predictions than the original model, clearly illustrating overfitting. These findings underscore the importance of balancing model complexity in SCM applications and highlight the value of empirical tuning beyond theoretical assumptions. |
| Supervisor | Lieli, Robert Pal |
| Department | Economics MA |
| Full text | https://www.etd.ceu.edu/2025/menglibayevnurdaulet.pdf |
Visit the CEU Library.
© 2007-2025, Central European University