Essays on Tail Risks in Macroeconomics

Abstract

[eng] This thesis contributes to two problems identified in the literature: i) How do US financial conditions impact funding markets (credit and stocks) in a large set of countries around the world under different scenarios of macro-financial distress?; and ii) What role can be played by high-frequency data, real variables, and machine learning techniques in improving the forecasting performance of macroeconomic tail risk measures? In Chapter 2, I prove answers to the former question, while in Chapters 3, 4, and 5 I deal with the latter question. From a methodological perspective, I use time series econometrics, quantile regressions, mixed data sampling methods, machine learning models, and forecasts evaluation tests to address the various research questions. Furthermore, this dissertation has implications for risk management, monetary policy, financial stability, and forecasting. In Chapter 2, I systematically document vulnerable funding episodes in the world economy. That is, financial conditions in the United States have significant predictive power in the lowest quantiles of credit growth and stock market prices around the global economy. I also show that vulnerable funding can be explained, mainly contemporaneously, by the relative market size in the case of credit markets and by the financial links with the US (measured by the total direct investment of the US as a percentage of the country’s GDP) in the case of the stock market. The policy implication of this work is clear. I show that international funding markets are a source of persistence and amplification of financial conditions shocks across the global economy. This means that a deterioration of US financial conditions calls for policy actions in other economies around the world. In the second part of my dissertation, I tackle the problem of producing accurate, out-of-sample tail forecasts for output growth, unemployment and inflation. In Chapter 3, I show that both real and financial variables reported with a daily frequency provide valuable information for monitoring periods of economic vulnerability. I further show that is possible to provide an early warning of a downturn in GDP in pseudo real-time and that this framework works well during episodes of distress. The flexible approach reported allows me to emphasize the importance of both economic theory and economic intuition when interpreting the results of forecast combinations and for improving the point forecast itself. All in all, I contribute to a better understanding of the economic signals that can be extracted from this daily information when seeking to anticipate downturns in the economy. In Chapter 4, I construct daily unemployment at risk around consensus forecasts conditional on the Aruoba-Diebold-Scotti business conditions index, using a quantile mixed sampling model. My results suggest that this indicator has better nowcasting properties than those provided by other daily financial conditioning variables, and provides early signal of unemployment rate increases, especially during episodes of distress. The results are relevant for risk monitoring and nowcasting purposes of central banks and other institutions. In Chapter 5, I investigate potential future inflation risks in a large group of countries, using inflation density forecasts based on a set of global factors as predictors. I provides evidence that, in general, global inflation factors improve the accuracy of density forecasts. Also, I show that state-of-the-art machine learning techniques provide superior predictive performance. I document heterogeneous patterns of inflation risk measure across world regions. The results of this chapter are relevant from the perspective of a central bank or an international organization, as they often want to assess risk across different regions. In this regard, I find that global factors are generally robust predictors of density forecasts across countries. This also calls attention to a synchronized reaction of the largest central banks around the world, which is likely to contribute to sustain global price stability.