Poziv na predavanje "Latent Factor Estimation in High-Dimensional Financial Time Series"

Sažetak predavanja

A fundamental question in finance is "What explains price movements of financial securities?" - with direct implications to risk management and portfolio optimization. A key component in explaining price movements is understanding the mutual movement and underlying correlation structures of large numbers of securities. In this talk, unsupervised learning methods for estimating latent factors in high-dimensional financial time series will be considered. A special focus will be placed on latent factor models and clustering methods which use the underlying correlation structures to obtain lower-dimensional representations. The talk will also cover improved covariance estimation methods and portfolio optimization approaches based on the estimated latent factors, with results using historical market data.

Životopis predavača

Stjepan Begušić received his PhD degree in 2020 from the University of Zagreb, with the doctoral thesis "Estimation of latent factors from high-dimensional financial time series using unsupervised learning". He has worked as a teaching and research assistant at UNIZG-FER since 2016, where he worked on projects funded by the Croatian Science Foundation, EU funding agencies, and partners from the industry. He is currently a postdoctoral researcher at UNIZG-FER, working on the project "Deep Reinforcement Learning Algorithms for Risk Management" (DREAM) funded by the Croatian Science Foundation. He is the co-author of multiple journal and conference papers on the topics of statistical and machine learning methods for finance, with a focus on high-dimensional problems in financial risk modeling. He is a member of the Laboratory for Financial and Risk Analytics, and is a member of IEEE and the IEEE Signal Processing Society.