By Alex Botte, CFA, CAIA, Head of Client and Portfolio Solutions at Runa Digital Assets, LLC, an investment firm specializing in digital asset portfolios.
Executive Summary
- This post is the first in a series in which we construct a digital asset-specific risk model. We begin by establishing the “layer 1,” foundational factor in our risk model, which is crypto beta.
- There is compelling evidence for a shared risk factor in digital assets. In fact, the degree of co-movement of the top digital assets is stronger than the co-movement of the top stocks.
- We believe that this crypto beta should not only form the foundation of a digital asset-specific risk model, but also form the core of a digital asset portfolio allocation, similar to equity beta’s role in equity risk models and portfolio allocations. Crypto beta can also be helpful in investment decision making by using it to evaluate active manager performance.
Introduction
According to a Fidelity institutional investor survey, one of the biggest downsides institutional investors see to investing in digital assets is price volatility.[i] This is understandable. Just compare the volatilities of digital assets to traditional assets like stocks and bonds like we do in Exhibit 1. Bitcoin and Ethereum, the two largest digital assets by market capitalization,[ii] have experienced 80% and 110% annualized volatilities, respectively, while stock and bond indices were much lower at 19% and 5%, respectively. Digital assets have also carried much higher volatilities compared to indices of alternative assets like hedge funds and commodities.
So far, these stock, bond, hedge fund, and commodity comparisons use indices, which are diversified and exhibit lower correlations than the individual assets within them. But even an individual stock, like Apple, that is riskier than a diversified stock index has meaningfully lower volatility than Bitcoin and Ethereum.
Exhibit 1: Average Annual Returns and Annualized Volatilities (Sorted By Volatility)
January 3, 2017 - February 28, 2022 using daily data. Bitcoin and Ethereum price data is sourced from Messari. US Bonds, US Stocks, and Commodities are represented by the iShares Core U.S. Aggregate Bond ETF, the SPDR S&P 500 ETF Trust, and the iShares S&P GSCI Commodity-Indexed Trust respectively, and are sourced from Yahoo Finance. The price data for Apple is also sourced from Yahoo Finance. Hedge Funds are represented by the Barclay Hedge Fund Index, and the return data is sourced from BarclayHedge. Hedge Funds use monthly data, and January and February 2022 returns are estimates. Returns are calculated arithmetically.
The good news is that crypto’s extra risk has historically been compensated, as Bitcoin and Ethereum have enjoyed the highest average annual returns (in the triple digits!) relative to stocks, bonds, hedge funds, and commodities.[iii] In fact, over this period, Bitcoin and Ethereum were the only assets, except for Apple, to experience average annual returns that exceeded their annualized volatilities.
The higher volatilities of digital assets necessitate a crypto-specific risk model for investors to understand and manage these risks. In this piece (the first of several posts on building a digital asset-specific risk model), we seek to quantify and define the foundational factor in our risk model: broadly shared digital asset risk, or “crypto beta.”
Applying the CAPM to Crypto
Before we jump into the analysis, we first provide a brief background on risk models. One of the original risk models for stocks was the Capital Asset Pricing Model, or the CAPM, which was introduced in the 1960s. The idea behind the model was that a stock’s return can be explained by three components: (1) the rate that a theoretical risk-free asset returns, (2) the return that can be explained by the stock’s relationship to the overall equity market, and (3) the idiosyncratic, or unique, return of that stock.
Exhibit 2: Capital Asset Pricing Model Formula
Sources: Treynor (1961, 1962), Sharpe (1964), Lintner (1965), and Mossin (1966).
The second component is important. It is the portion of risk and returns that can be attributed to the stock’s relationship to an undiversifiable market risk (commonly called “equity beta”), which is shared across stocks broadly. Stocks generally move together as we will show later on in this post. Of course they don’t move together in perfect lockstep, but stocks tend to have relatively high and persistent correlations with one another. Over time, investors have been compensated in the form of positive returns for accepting equity beta. This equity risk forms the foundation of most equity-specific and even broader, multi-asset risk models.
We believe that a similar concept to “equity beta” exists in crypto and that we can quantify and define it. Thus, the rest of this post focuses on using the idea of the CAPM to present a case for building a similar model that is tailored to digital assets.
Quantifying Crypto Beta
We first need to determine the extent to which there is a shared risk in digital asset markets, similar to equity beta in the original CAPM. We start our analysis by looking at the historical correlations of the top digital assets (see the Appendix for more information on these assets).[iv]
Exhibit 3: Correlations of the Top Digital Assets
December 31, 2020 - February 28, 2022 using daily data. Price data is sourced from Messari.
The average correlation over this period was 46%, and 99% of the correlation pairs were positive. The two least correlated assets (those with the green columns/rows) are Unus Sed Leo and Dogecoin, which are two very unique coins. The former is an exchange token used in the iFinex ecosystem and the other is a “meme” coin. If we exclude these two assets from the analysis, the average correlation rises to 50%, and every single correlation pair was positive.
An easier way to visualize Exhibit 3 is to bucket the correlations and determine the percentage of pairs in each bucket.
Exhibit 4: Histogram of Correlations of the Top Digital Assets
December 31, 2020 - February 28, 2022 using daily data. Price data is sourced from Messari.
Exhibit 4 shows that the vast majority of pairwise correlations are between 25% and 75%, which seems to indicate that there is a decent amount of co-movement across this set of digital assets. Let’s compare these results to equity markets, where we know a common risk factor exists. We can use the 30 stocks in the Dow Jones Industrial Average.[v]
Exhibit 5: Correlations of the Top Stocks
December 31, 2020 - February 28, 2022 using daily data. Price data is sourced from Yahoo Finance.
Similar to what we observed for digital assets, the average correlation among these top stocks was positive at 26% and the vast majority (99%) of the correlation pairs were positive. However, compared to the digital asset correlation matrix, we see a lot more green and yellow (lower correlations) versus red and orange (higher correlations).
We add the distribution of stock correlations to the histogram that we presented in Exhibit 4.
Exhibit 6: Histogram of Correlations of the Top Digital Assets and Top Stocks
December 31, 2020 - February 28, 2022 using daily data. Digital asset price data is sourced from Messari. Stock price data is sourced from Yahoo Finance.
Digital asset correlations were higher than stock correlations, although there is reason to believe correlations between digital assets will decline over time toward the level observed for stocks as the industry matures and investment is less driven by narratives and more by fundamentals.
Nevertheless, we know that it is accepted that stocks have a common risk factor (equity beta), and the results here clearly show that the level of co-movement of the top digital assets is stronger than the co-movement of the top stocks. The higher the co-movement, the higher the shared risk, so this is certainly compelling evidence for a crypto beta.
We can further test this idea by employing a Principal Component Analysis (PCA), which is a mathematical tool that can extract key information from large data sets, like the one we have here for digital assets. The result of PCA is different variables (principal components) that can explain variation in the digital asset data. There will be one variable for each asset in our sample.
Exhibit 7: Variance Explained by the PCA Variables (or Principal Components)
December 31, 2020 - February 28, 2022 using daily data. Digital asset price data is sourced from Messari.
Each bar in Exhibit 7 represents the amount of risk across our data set that can be explained by that variable, and the sum of the bars equals 100%. This technique was able to find a single variable that alone can explain over 41% of the variance in this digital asset data set. PCA is an unsupervised technique that is completely data-driven, so this first variable doesn’t have a qualitative label. However, the variable will have a “weight” to each asset in our sample, which will allow us to better understand what this variable represents.[vi]
Exhibit 8: Weights of Variable (Principal Component) #1
December 31, 2020 - February 28, 2022 using daily data. Digital asset price data is sourced from Messari.
The key takeaway from Exhibit 8 is that this first risk factor is pointing in the same direction for every asset in our sample. The largest risk driver is simply having directional exposure to crypto. This once again confirms that there is a common risk driver in this asset class, giving us confidence to create the foundational factor in our risk model, crypto beta, similar to how equity beta is the foundation of the CAPM.
Defining Crypto Beta
Practitioners proxy equity market risk and returns using broad-based equity indices like the MSCI World Index or the S&P 500 Index. We can do something similar for digital assets to proxy crypto beta. There are already digital asset indices like the Bloomberg Galaxy Crypto Index that exist. These, similar to their equity counterparts, are typically constructed using the largest assets by market capitalization (e.g., the Bloomberg Galaxy Crypto Index has historically had 5-6 constituents).[vii]
We’ll take a similar approach and create a simple, market-cap weighted index of the top five digital assets. It is rebalanced quarterly, and it excludes stablecoins.[viii] As expected, bitcoin makes up a predominant portion of this index construction, but its share has been declining over time. This is consistent with bitcoin’s decline in market dominance over this period.
Exhibit 9: Composition of a Simple Crypto Beta Index
December 31, 2020 - February 28, 2022 using daily data. Digital asset price data is sourced from Messari. Market capitalizations for weightings are sourced from Messari; the top five assets at each rebalancing date are sourced from CoinMarketCap.
We’ll analyze the historical performance of this simple index and compare it to Bitcoin and Ethereum as well as a 50/50 combination of the two assets. Below we see that the average annual return of this simple index since January 2021 was 84%, just outpacing its annualized volatility of 83%. Bitcoin returns and risk were a bit lower, while Ethereum clearly outperformed over this period. Similar to the equity market risk premium, historically, the market has (handsomely) compensated investors for accepting crypto beta, no matter which of these four measures you use.[x]
Exhibit 10: Summary Statistics of Different Measures of Crypto Beta
December 31, 2020 - February 28, 2022 using daily data. Digital asset price data is sourced from Messari. Market capitalizations for weightings are sourced from Messari; the top five assets at each rebalancing date are sourced from CoinMarketCap. Returns are calculated arithmetically.
Conclusion
In summary, the analysis strongly supports the existence of a shared risk factor in digital assets. We believe that this crypto beta can be helpful to digital asset investors in several ways:
- It can form the foundation of a digital asset-specific risk model, similar to equity beta’s role in the CAPM. In future posts, we plan to demonstrate how digital asset risk models can be expanded to include other risks, similar to how the CAPM was expanded to include other factors.
- It can form the core of a digital asset portfolio allocation, similar to equity beta’s role in equity portfolio allocations.
- It can be used to evaluate active managers’ performance to determine if they are adding value above and beyond passive crypto beta.
Appendix: More Information on the Top Digital Assets Used in the Analysis
Inception Date is the date of first available prices on Messari. Sectors are determined by Runa Digital Assets. Descriptions are sourced from CoinMarketCap. Market capitalizations are from CoinMarketCap’s historical snapshot on February 27, 2022.
Footnotes:
[1] 54% of surveyed investors believed that price volatility is one of the greatest barriers to investment. Source: https://www.fidelitydigitalassets.com/bin-public/060_www_fidelity_com/documents/FDAS/2021-digital-asset-study.pdf
[2] As of February 25, 2022 according to CoinMarketCap.
[3] It may make sense to use log returns for highly volatile assets. Average annual log returns over this period for Bitcoin and Ethereum were 72.7% and 111.4%, respectively, which are still higher than the holding period returns of any of the other assets in Exhibit 1.
[4] We used the top 50 assets by market capitalization according to CoinMarketCap’s historical snapshot on February 27, 2022. We excluded stablecoins, Wrapped bitcoin, Bitcoin BEP2, assets that were incepted after December 31, 2020, and assets with missing data, leaving us with a total of 38 assets for this analysis.
[5] Source for the Dow Jones Industrial Average (DJIA) constituents: https://www.slickcharts.com/dowjones We selected the DJIA because it had a similar number of assets to the number in our digital asset universe. The DJIA is also focused on large caps, similar to our focus on the top digital assets by market capitalization.
[6] The weights displayed in Exhibit 8 are the eigenvectors.
[7] Source: https://data.bloomberglp.com/professional/sites/10/BGCI-Factsheet-February-2020.pdf
[8] We relied on the historical market capitalizations from CoinMarketCap here. We begin on December 31, 2020, using the market capitalizations from CoinMarketCap’s December 27, 2020 historical snapshot. The specific rebalance dates are as follows: March 28, 2021, June 27, 2021, September 26, 2021, and December 26, 2021.
[9] Admittedly this is a short period, however, Exhibit 1 also shows the positive compensation for accepting crypto beta (measured by Bitcoin and Ethereum) over a longer period that includes the 2018 crypto bear market.
About the Author:
Alex Botte, CFA, CAIA is the Head of Client and Portfolio Solutions at Runa Digital Assets, LLC, an investment firm specializing in digital asset portfolios. Alex was previously a Vice President on the Client Portfolio Management team at Two Sigma Advisers LLC where she provided transparency to Two Sigma’s fund investors in addition to producing investment management-related content and thought leadership.
Prior to Two Sigma, she was at AQR Capital Management, where she most recently served as a Product Specialist for the firm’s Global Asset Allocation strategies. Prior to AQR, she worked in Prime Services at Barclays. Alex holds a Bachelor’s of Science in Applied Economics and Management from Cornell University.
A special thank you to Robert Gingrich, PhD, Bryan Johnson, CFA, CAIA, Jennifer Murphy, CFA, and Max Williams for their helpful contributions and edits to this article.
This article (i) is only for informational and educational purposes, (ii) is not intended to provide, and should not be relied upon, for investment, accounting, legal or tax advice, and (iii) is not a recommendation as to any portfolio, allocation, strategy or investment. This article is not an offer to sell or the solicitation of an offer to buy any securities or other instruments.