Date of Degree


Document Type


Degree Name





Lilia Maliar

Committee Members

Wim Vijverberg

Sebastiano Manzan

Subject Categories



Machine Learning, Synthetic Control, Econometrics


This dissertation consists of three chapters on machine learning modeling in economics. Chapter 1 - Robust PCA Synthetic Control: In this chapter, I propose an algorithm for comparative studies called robust PCA synthetic control. My algorithm builds on the synthetic control model of Abadie et al., 2015 and the robust synthetic control model of Amjad et al., 2018. I apply all three methods (robust PCA synthetic control, synthetic control, and robust synthetic control) to answer the hypothetical question, what would have been the per capita GDP of West Germany if it had not reunified with East Germany in 1990? I then implement two placebo studies. Finally, I test the outcome of each method for robustness. Additionally, I implement robust PCA synthetic control on the case of the Facebook privacy scandal in 2018 to investigate its impact on Facebook stock price. This paper demonstrates that robust PCA synthetic control can outperform the robust synthetic control model of Amjad et al., 2018 in placebo studies and is less sensitive to the weights of synthetic members than the synthetic control model of Abadie et al., 2015. Chapter 2 - Improving Time Series Extrinsic Regression: In this chapter, I propose an innovative deep learning model (ROCKET-XGBoost) for time series analysis. I first review the current deep learning models for time series analysis and explain the concept of time series extrinsic regression. Building on the ROCKET model initially proposed by Dempster et al., 2020, I suggest applying elements of XGBoost in order to improve accuracy. Using the data sets gathered by Tan et al., 2020, I show that ROCKET-XGBoost has greater accuracy compared to the other deep learning and machine learning methods for time series extrinsic regressions. Not only does ROCKET-XGBoost maintain the efficiency of ROCKET, but it can also improve its estimation precision. Chapter 3 - The Contribution of the Minimum Wage to US Wage Inequality, A Penalized Spline Approach: In this chapter, I reassess the effect of minimum wage on U.S. earnings inequality using Current Population Survey data from 1979 to 2012 and the findings in Autor et al., 2016. I apply a penalized spline technique that addresses potential biases in parametric estimation in earlier works. Using this method, I find that, in contrast with the conclusion of Lee, 1999, the spillover effect of minimum wage on the upper tail and lower tail of wage distribution, where the minimum is nominally nonbinding, is small and not significant.

Included in

Econometrics Commons