Abstract

This dissertation consists of four chapters on applying the Synthetic Control Method to rare events with a significant impact. Initially pioneered by Abadie et Al., the Synthetic Control Method is a policy analysis tool developed to tackle the weaknesses of traditional policy analysis models such as the Difference-in-Difference. In the first chapter of this dissertation, this model is used to reconcile a recurring issue in the disaster literature: why some countries recover better than others from disasters and the role that political institutions play in this recovery. The results show that regulatory power is the most significant institutional quality variable that determines post-disaster recovery. Ranking in the top 30\% of countries regarding regulatory power is linked to GDP recovery rates from disasters that are higher than predicted GDP. The variable with the most negligible impact was corruption, as proxied by the corruption perception index. A 1-point increase in this index was linked to a 0.05\% increase in the recovery rates compared to predicted GDP. On the other hand, the degree of democratization or level of democracy is insignificant in determining the size or level of recovery. Finally, over five years after the occurrence of a disaster, countries that experienced negative recovery rates of GDP per capita had this value shrink by about 13\%. In contrast, countries with positive recovery rates of GDP per capita ended up with a GDP per capita ahead of its predicted value by 8\% over the same five-year period.

One of the most devastating disasters of the last 50 years is the COVID-19 pandemic that put the entire world at a standstill. The second chapter of this dissertation summarizes the literature surrounding anti-contagion policies and highlights a gap in the literature in untangling the impact of individual anti-contagion policies. This gap is tackled in the third chapter, which investigates the relative importance and impact of individual anti-contagion policies in reducing death rates in the United States. Restrictions on gatherings proved to be the most significant policy in reducing death rates, lowering them on average by four out 100,000 COVID deaths per day 60 days after the implementation of such a policy. School closings and public transportation closings were the least effective policies reducing death rates by 0.2 and 0.5 per 100,000 over the same period.


In the fourth chapter, the traditional Synthetic Control Method is modified to account for cumulative and interrupted events through the Multi Synthetic Control Method. This method is tested on previous examples used in the literature and is shown to be robust to uninterrupted events. When applied to anti-contagion policies in the United States, the Multi Synthetic Control Method finds that the standard Synthetic Control Method can underestimate the true impact of a policy by up to 150\%. The values obtained from the Multi Synthetic Control Method for the same event as compared to the base Synthetic Control Method were significantly different, ranging between 20\% to 150\% different in absolute value.


Significant improvements have been made to the original Synthetic Control Method since its inception. In this thesis, additional improvements are proposed to improve this method's accuracy. In the first chapter, a new method of selecting the vector of relative importance (known as the $V$ vector) is discussed. This method improves the accuracy of obtaining this vector for regressions where the variance of the treated variable and the number of co-factor variables are high.

The results of this dissertation show the ability of the Synthetic Control Method to tackle all kinds of policies. Policy-makers aiming to take on upcoming waves or different mutations of the COVID-19 virus should consider the effectiveness of different policies and the implication of their stringency in affecting death rates and economic variables, and the trade-off between them.