Essays in Macro-Finance and Statistics

Abstract

This dissertation consists of two chapters on my work in macro-finance, and one chapter on the statistical properties of k-clustering.The first chapter proposes an objective disaster risk measure based on a real economic outcome that captures the tail risks harbored by credit expansions and market booms across three disaster dimensions: crashes in real GDP growth, financial crisis, and equity market crashes. Our findings on the correlation between disaster risk and asset prices substantiate a disconnect between realized tail risks and investors’ perceived tail risks, and point to the fragility of rare disaster models proposed in literature. This chapter is based on the working paper “Time Varying Disaster Risk and Asset Prices” with Matt Baron and Wei Xiong. The second chapter presents a study on the role of sectoral credit in aggregate credit booms, financial crises, and crashes in GDP growth. We find that the aggregate credit cycle masks co-existing sectoral credit cycles that individually exhibit different characteristics, and that the use of sectoral credit data substan- tially improves out-of-sample forecasts of macroeconomic risks. Our findings suggest that using measures of “the” credit cycle for regulatory oversight can be highly misleading. This chapter is based on the working paper “Varieties of Credit Booms” with Karsten Müller and Emil Verner. The third chapter presents a k-clustering procedure based on Lp distance, gener- alizing the popular k-means and k-medians estimator. We establish novel results on the strong consistency, asymptotic normality, and bootstrap validity of the estimator, and derive concentration bounds and asymptotics of the associated risk criterion. A byproduct of this work is a framework for proving asymptotic normality and bootstrap validity of regular M-estimators that unifies developments in empirical processes. We will derive sufficient conditions based on the combinatorial properties of the risk function class that only require smoothness of expectations of the risk function, thereby enabling one to sidestep the analytical challenges of using the classical stochastic differentiability approach when the risk function is non-smooth and non-convex.