Data-Driven Crypto Risk Factors

Abstract: Usually, risk factor literature assumes that these factors driving returns are observable (at least theoretically) and are usually obtained by a regression model. An alternative approach is to assume that factors are in fact unobservable and generate these out of pure statistical methodology. One such technique is principal component analysis (PCA), which can be used to generate orthogonal (or independent) factors that drive variability of returns. In this paper, we apply PCA to extract factors out of a top 1000 (by market cap) crypto asset universe, after data cleaning and processing, leaving a net 103 assets studied. We find that the first 10 factors explain about 60% of variability in returns, which is much lower than the usual explainability with mature markets like equities, possibly suggesting additional factors in the crypto universe or hinting at a yet amateur market. We also see the effect of survival bias at play by observing a notable increase in total variability when we reduce the data coverage constraints. Lastly, we construct eigen portfolios and compare performance of these with an equally weighted benchmark.

Download the paper to dive into this analysis of risk factors in the crypto universe using principal component analysis.

‍*Note: Cloudwall and the technology behind its Serenity System were acquired by Talos in April 2024. Learn more.