UC Riverside

This dissertation discusses the application of machine learning techniques on the economic causal inference and forecasting.

In Chapter 2, we develop a new method, the $L_1$-regularized soft decision tree, to identify the relevant features for the heterogeneous treatment effect and confounding effect under a very flexible nonlinear potential outcomes framework for causal inference. Compared to other methods, we show that our approach can identify relevant factors without the widely used assumption of additive nonlinearity. By embedding the debiased soft decision tree into the $L_1$-based variable selection framework, we show that the $L_1$-regularized soft decision tree outperforms other variable selection methods such as lasso in nonlinear settings.

Chapter 3 focuses on macroeconomic forecasting literature. This chapter introduces unFEAR, an unsupervised feature extraction clustering method aimed at facilitating crisis prediction tasks. We use unsupervised representation learning and a novel autoencoder method to extract from economic data information relevant to identify time-invariant non-overlapping clusters comprising observed crisis and non-crisis episodes. Each cluster corresponds to a different economic regime characterized by an idiosyncratic crisis generating mechanism.

Chapters 4 and 5 summarize the literature of economic forecasting using two attractive machine learning techniques. Chapter 4 focuses on the Bagging and Random Forests methods. We explore Bagging, Random Forest, and their variants in various aspects of theory and practice. We also discuss applications based on these methods in economic forecasting and inference. Chapter 5 discusses Boosting. Boosting can estimate the variables of interest consistently under fairly general conditions given a large set of explanatory variables. Boosting is fast and easy to implement, which makes it one of the most popular machine learning algorithms in academia and industry.