Christian Menden Livres

Focusing on the intersection of data augmentation (DA) and machine learning (ML), this dissertation addresses challenges posed by poor data quality in ML applications. It explores various DA methods, aiming to clarify their benefits and obstacles in practical use. Through a multiple-case study, the research demonstrates how DA can enhance the performance and applicability of ML techniques, specifically within supply chain management, thereby contributing valuable insights to both fields.

The analysis of the financial cycle and its interaction with the macroeconomy has become a central issue for the design of macroprudential policy since the 2007-08 financial crisis. This paper proposes the construction of financial cycle measures for the US based on a large data set of macroeconomic and financial variables. More specifically, we estimate three synthetic financial cycle components that account for the majority of the variation in the data set using a dynamic factor model. We investigate whether these financial cycle components have significant predictive power for economic activity, inflation and short-term interest rates by means of Granger causality tests in a factor-augmented VAR set-up. Further, we analyze if the synthetic financial cycle components have significant forecasting power for the prediction of economic recessions using dynamic probit models. Our main findings indicate that all financial cycle measures improve the quality of recession forecasts significantly. In particular, the factor related to financial market participants’ uncertainty and risk aversion – related to Rey’s (2013) global financial cycle – seems to serve as an appropriate early warning indicator for policymakers.