Measuring and pricing macroeconomic uncertainty: a machine learning econometric approach

Abstract

This thesis measures and prices macroeconomic (or aggregate) uncertainty with non-parametric (AI/ML) methods, benchmarking against the current parametric standard in the literature. Long-short term memory deep neural networks (LSTMs) are the current method of preference to measure time varying phenomena such as macroeconomic uncertainty in chapter 2. Before examining whether a non-parametric measure of macroeconomic uncertainty is priced in the cross section of US stock returns in chapter 4, chapter 3 inquires into the common empirical finding of a negative uncertainty premium. To do so, chapter 3 exploits monthly data for the US AMEX, Nasdaq and NYSE stocks between 1993 and 2022, to build a dynamic hedging strategy across calm and turbulent sub-periods, examining the corresponding uncertainty premia. All along, parametric and non-parametric macroeconomic uncertainty measures are compared between (e.g. deploying VARs) and in terms of their pricing effects (e.g. with suitable tests for nested and non-nested specifications). The results indicate that the non-parametric measure of uncertainty has superior explanatory and predictive power for stock returns compared to the traditional parametric measures.

More information

Identifiers

ORCID for Hector Calvo-Pardo: orcid.org/0000-0001-6645-4273

ORCID for Jose Olmo: orcid.org/0000-0002-0437-7812

Catalogue record

Export record