UC Santa Barbara

Complex policy problems like climate change and illicit finance require a diverse methodological repertoire and an agnostic approach to selecting the appropriate analytical tool to accomplish discrete inferential tasks. Drawing from the disciplines of political science, economics, and statistical data science, this dissertation tackles three distinct problems on causal evaluation, measurement, and missing data. The first paper evaluates the causal effect of a climate mitigation policy on the carbon emissions of the UK. Using a synthetic control estimator, this chapter finds that post-treatment emissions in the UK were 10% lower than what they would have been without the climate policy. The results imply that voluntary climate reforms that make concessions to domestic producers are still able to meaningfully reduce emissions, even in the absence of a legally binding global climate treaty. The second paper presents a novel methodology to measure illicit trade flows and originates the "atlas of misinvoicing", the first database to provide comprehensive bilateral estimates of the dollar amount of misinvoiced trade disaggregated by commodity sector for 167 countries during 2000-2018. Results show that African countries lost on average $86 billion a year in gross illicit outflows during that period, and that the biggest source of illicit trade on the continent was the natural resources sector. The findings suggest that combating illicit financial flows will be crucial to providing finance for sustainable development and to promoting domestic resource mobilization in poor countries. The third paper proposes a machine learning approach to ameliorate the problem of missing data from developing countries, where administrative systems for data collection tend to be weaker. Some African countries do not provide customs declarations, which the "atlas" method requires as input data. This chapter predicts illicit trade using machine learning models that are trained on readily available data without relying on official trade statistics. Findings show that the models are able to recover 70% of the variation in illicit trade outcomes. This demonstrates the promise of predictive approaches to augment existing measures of illicit finance in data-constrained settings. Broadly, the chapters in this dissertation can be understood as operating in the different scientific frameworks of causal, descriptive, and predictive inference, respectively. Tackling difficult environmental and developmental problems will require a willingness to traverse methodological siloes in order to identify the best tool for the job – this dissertation contributes to pushing the search for solutions forward.