LQG 2024/10/08 – 18:30 – Beyond Time Series Information: Beta Estimation via Machine Learning – Tizian Otto – at Bloomberg – IP *OLO* Hybrid

by

A methodology to approximate a stock’s true unobservable market beta while mitigating systematic biases using machine learning.

LQG 2024/10/08 – 18:30 – Beyond Time Series Information: Beta Estimation via Machine Learning

Seminar by Tizian Otto

2024/10/08 – In-Person and On-Line

The LQG is very grateful to Bloomberg – for hosting this LQG seminar.

Obtaining good estimates of stock-level market betas is undoubtedly important for many market participants. Use cases range from determining a firm’s cost of capital to constructing portfolios that aim to exploit alpha signals while neutralising market risk. Unfortunately, a stock’s true sensitivity toward the market portfolio is unobservable and generating meaningful forecasts is particularly difficult. The common approach is to run time series regressions at a stock-level to obtain historical betas and take those estimates as predictions for a stock’s future beta.
In line with the literature, our research suggests that historical betas possess systematic biases. Utilising information beyond returns can help improve beta estimation. We show that our machine learning (ML)-based beta estimates outperform leading benchmarks statistically and economically with lower stock-level forecast errors and better hedging capabilities at the portfolio level.
We propose a methodology to approximate a stock’s true unobservable market beta while mitigating systematic biases. Aiming to produce the best estimates of “true” betas, we apply ML techniques to extract valuable information from stock-level characteristics as well as macroeconomic indicators. Reconsidering the well-known trade-off between bias and noise, we highlight the relevance of the portfolio size on the optimality of beta estimates.

 

Tizian Otto

Researcher in Portfolio & Risk Analytics

Tizian Otto, Ph.D., is a Researcher in the Portfolio & Risk Analytics Team at Bloomberg in London where he is leading the acceleration of machine learning adoption.  Prior to that, he was a Research Fellow at Yale University as well as a Visiting Research Scholar at Stanford University and Harvard University.
During his doctoral and postdoctoral studies, Tizian worked on the development of machine learning-based models to forecasting return, risk, and liquidity in different financial markets.
Holding a Ph.D. in Finance from the University of Hamburg, Germany, he reviewed and published several articles on this topic in leading academic and practitioner journals.