City Research Online

Abstract

How would a portfolio perform under alternative macroeconomic conditions? Traditional scenario analysis in finance relies heavily on historical data, thus limiting risk assessment under rare or unobserved macroeconomic environments. We introduce MacroVAE, a variational autoencoder that generates realistic return sequences based on macroeconomic indicators to enable counterfactual scenario analysis. Trained on historical futures returns and macroeconomic data from global economies, MacroVAE generates return sequences conditioned on specified macroeconomic scenarios. The model uses convolutional ResNet blocks with Feature-wise Linear Modulation for stable macroeconomics-driven generation. In rolling out-of-sample evaluation, MacroVAE outperforms state-of-the-art generative baselines in replicating empirical distributions and financial stylized facts. We demonstrate two applications: counterfactual scenario analysis under alternative macroeconomic conditions, and forward-looking stress testing across diverse inflation-growth combinations, including ones rarely or not observed historically. MacroVAE enables systematic exploration of macroeconomic conditions, expanding the toolkit for portfolio risk management.

Publication Type:	Conference or Workshop Item (Paper)
Additional Information:	This work is licensed under a Creative Commons Attribution 4.0 International License.
Publisher Keywords:	Generative machine learning models, synthetic data, time-series, asset allocation, financial risk management
Subjects:	H Social Sciences > HB Economic Theory H Social Sciences > HG Finance
Departments:	School of Science & Technology School of Science & Technology > Department of Computer Science
SWORD Depositor:	Symplectic Administrator

Export

Downloads