Applying Machine Learning Portfolio Modeling to Bitcoin

Investors considering whether to add bitcoin to their multi-asset class portfolio face the challenge of how to think about both the impact of a bitcoin allocation and how to set and manage their exposure. A simple way to start can be to examine a portfolio by comparing what its historical return would be with and without bitcoin, which is what we have done in Getting off Zero, our previous report. However, the major shortcoming of this approach is that bitcoin’s price history is incredibly short, and its historical returns have been abnormally high as the asset class matures (bitcoin’s market capitalization went from a value of less than $10 million to more than $1 trillion in approximately 10 years).

In this paper, we explore a different way to understand the dynamic impact that bitcoin may have on a portfolio, using new, proprietary portfolio modeling technology from Fidelity. This technology can be used to support investment decisions and help explore scenarios but is not meant to give a specific forecast or portfolio recommendations.

• Our new model uses machine learning and high-performance cloud computing to overcome many of the limitations inherent in short time series data, as well as some shortcomings of traditional statistical investment models.
• To enable personalized decision support, the model allows for flexible returns assumptions. Here, we look at arguably more realistic return assumptions that bitcoin may achieve going forward, compared with its outsized performance of the past. The characteristics of bitcoin’s return outcomes considering such expected return assumptions are explored in a portfolio context.
• We also extend our previous analysis from a simple 60% equity and 40% bond portfolio to one that is more representative of today’s institutional and multi-asset class portfolio that could include real estate, commodities, precious metals, and cash.
• Finally, we introduce the mathematical concept of evaluating a portfolio’s risk-return “efficient” frontier, defining investment risk not only on a basis of volatility but also extreme loss potential. We advance this “return loss efficient frontier” as more informative for an emerging asset class like bitcoin, in which investors want to take advantage of the potential upside gain but still control overall portfolio loss.

Watch The Value Exchange: Applying Machine Learning Portfolio Modeling to Bitcoin for more insights.

Written by:

Chris Kuiper, CFA, Director of Research, Fidelity Digital Assets

Aaron Gao, Head of Investment Application Development,

Fidelity Advanced Technologies for Investment Management (ATIM)

Yonatan Tekleab, PhD, Director of Data Analytics, Fidelity ATIM