Cryptoasset markets have experienced a paradigm shift over the past decade: originally regarded as a purely speculative investment, institutional investors are starting to appreciate cryptoassets’ unique return drivers. As of 2020, 45% of institutional investors in Europe and 27% in the US have exposure to cryptoassets, either directly or via futures contracts. Despite considerable interest in this evolving asset class, the question of what factors drive expected returns in cryptoassets remains largely unexplored.
On the one hand, Cheah and Fry (2015) argue that Bitcoin has a fundamental value of zero and contains a significant speculative component. On the other hand, Cong, Li, and Wang (2020) develop a theoretical model and emphasize that the fundamental value of cryptoassets positively depend on their network size. Since empirical asset pricing studies on cryptoassets are limited and often omit cryptoassets’ fundamentals in their analysis, the relation between cryptoasset prices and blockchain fundamentals remains an open question.
My recent working paper two goals. First, I examine the existence of a value premium in cryptoasset returns. That is, I assess whether cryptoassets with high ratios of active addresses to the network value (value cryptoassets) yield higher average returns than cryptoassets with low ratios of active addresses to the network value (growth cryptoassets). Active addresses refer to the number of unique wallet addresses that conduct an on-chain transaction, whereas the network value of a cryptoasset corresponds to its market capitalization. Second, I examine whether a value factor has explanatory power for average cryptoasset returns in the cross-section.
My first result is easy to summarize. Value cryptoassets yield significantly higher returns than growth cryptoassets. Importantly, the value premium is large in economic magnitude with an average return of 2.1% per week and a Sharpe ratio of 0.33. Second, the value factor I identified explains the common variation in cryptoasset returns. A four-factor model directed at capturing the value pattern in average returns performs better than a three-factor model that only includes the market, size, and momentum factor. These results have significant implications for the understanding of cryptoassets. Most importantly, the results suggest that cryptoasset prices are related to their fundamentals and are not purely speculative investments.
My empirical study
I compile a unique dataset that combines blockchain fundamentals with financial data for 652 cryptoassets from July 4th, 2017 to October 6th, 2020. Blockchain fundamentals include information on the number of active addresses, dollar transaction volume, and the number of transactions. The main empirical analysis is conducted weekly and centers around a value anomaly measure that is defined as the ratio between the average number of active addresses over the past 30 days to network value (aanv30). Intuitively, the aanv30 ratio is a simple proxy of a cryptoassets’ network size relative to its market capitalization. Analogous to the book-to-market ratio for stocks, value cryptoassets are characterized by a high aanv30 ratio while growth cryptoassets are characterized by a low ratio.
A. Cross-section of Expected Returns
My first result is depicted in Figure 1, which shows average weekly portfolio returns from unconditional bivariate sorts. At the end of each Tuesday, cryptoassets are allocated independently into three aanv30 groups and small, neutral, and large market capitalizations. For each of the nine resultant portfolios, I calculate value-weighted returns from week to week.
I find a pattern in average cryptoasset returns that is related to the aanv30 ratio. Value cryptoassets with high aanv30 ratios yield higher average returns than growth cryptoassets with low aanv30 ratios. Importantly, the increasing pattern in average returns from growth towards value portfolios is persistent across all size terciles. This pattern in average returns represents the value premium in the cryptoasset market. In short, value cryptoassets yield higher returns than growth cryptoassets.
The value premium for cryptoassets is most profound within portfolios that are small and have low market beta low momentum, and high idiosyncratic volatility. Further, cross-sectional regressions confirm that the aanv30 ratio carries a positive return premium even after controlling for market beta, size, momentum, and idiosyncratic volatility. The value premium is also robust to different weighting schemes and holding periods. In a next step, I test if a value factor has explanatory power for average cryptoasset returns in the cross-section.
Figure 1. 3×3 independent size-aanv30 sort
Averages of weekly percent excess returns for value-weighted portfolios formed on size and average active addresses over the past 30 days to network value (aanv30); 04-July-2017 to 06-October-2020, 170 weeks. At the end of each Tuesday (=t), cryptoassets are allocated independently into three size and three aanv30 groups. The intersection of the two sorts produces nine portfolios. Then, value-weighted returns are calculated from Tuesday (t) to Tuesday in the consecutive week (t+1).
Averages of weekly percent excess returns for value-weighted portfolios formed on size and average active addresses over the past 30 days to network value (aanv30); 04-July-2017 to 06-October-2020, 170 weeks. At the end of each Tuesday (=t), cryptoassets are allocated independently into three size and three aanv30 groups. The intersection of the two sorts produces nine portfolios. Then, value-weighted returns are calculated from Tuesday (t) to Tuesday in the consecutive week (t+1).
B. A Four–Factor Model
To develop this model, I make use of four cryptoasset–specific common risk factors: excess market return (MKT), size (SMB), momentum (WML), and value (HML). The first three factors have previously been introduced by Liu, Liang, and Cui (2020). I propose the value factor as an additional common risk factor in the returns on cryptoassets and test the four-factor factor model’s performance to explain the common variation in cryptoasset returns. The value factor is constructed from 3×3 unconditional sorts on size and the aanv30 ratio. The four-factor asset pricing model is defined as follows:
Ri,t − RF,t = αi + βiMKTt + siSMBt + wiWMLt + viHMLt + si,t (1)
When an asset–pricing model perfectly captures expected returns, the intercepts are indistinguishable from zero when regressing cryptoasset’s excess returns on the factors. For all factor models investigated, I reject the null hypothesis that the intercept is zero at the 1% level. This result indicates that all models are incomplete descriptions of expected cryptoasset returns. However, of particular interest is whether the addition of the value factor improves the models’ ability to describe average cryptoasset returns. Thus, I compare the regression intercepts between factor models that exclude the value factor to models that include the value factor. The results suggest that adding the value factor to existing factor models significantly improves the model’s performance.
A two-factor model that contains the market and value factor provides the best description of expected returns. Potential redundancies of factors are evaluated by spanning regressions, which regress each of the four factors on the remaining three. The results suggest that in the four-factor model, the size factor is redundant for describing average returns. The second result can be summarized as follows: a four-factor model directed at capturing the value pattern in average returns performs better than a three-factor model.
Implications
The findings presented in my working paper have significant implications for the understanding of cryptoassets. First, the documented value premium shows that cryptoasset return patterns go beyond established anomalies in stock markets (e.g. size and momentum). Second, the results suggest that cryptoasset prices are related to their blockchain fundamentals, which challenges the view that cryptoassets are purely speculative investments. Third, as the four-factor model still provides an incomplete description of expected cryptoasset returns, more research is required in this field. Importantly, my findings offer new research opportunities that go beyond the interpretation of cryptoassets as a speculative bubble. In the future, many new patterns in cryptoasset returns may be discovered.
Luca Liebi is a Ph.D. candidate and research assistant at the School of Finance of the University of St. Gallen, Switzerland. His research interests are empirical asset pricing and risk management with a special focus on cryptoassets. His recent paper is available on SSRN.
