How do the major crypto-assets (Bitcoin, Ripple, and Ethereum) stack up against conventional asset classes? In their August 2018 paper entitled “Risks and Returns of Cryptocurrency”, Yukun Liu and Aleh Tsyvinski apply standard tools of asset pricing to measure crypto-asset exposures to:
- 160 equity factors.
- Macroeconomic factors (non-durable consumption growth, durable consumption growth, industrial production growth, and personal income growth).
- Major non-U.S. currencies (Australian Dollar, Canadian Dollar, Euro, Singapore Dollar and UK Pound).
- Precious metals (gold, platinum and silver).
They also investigate potential predictors for cryptocurrency returns analogous to those of traditional asset classes (momentum, investor attention, price-to-“dividend” ratio, realized volatility and supply). Finally, they measure exposures of various industries to crypto-asset returns. Using daily crypto-asset prices for Bitcoin since January 2011 and for Ripple and Ethereum since early August 2013, all through May 2018, along with contemporaneous data for other variables as outlined above, they find that:
- Crypto-assets have:
- Average returns and standard deviations of returns an order of magnitude higher than those for traditional asset classes. Returns have positive skewness.
- Sharpe ratios for daily and weekly (monthly) frequencies that are greater than (about the same as) that of the U.S. stock market.
- Substantial probabilities of extreme returns. For example, Bitcoin drops (jumps) by at least 20% on 0.5% (1%) of days.
- Crypto-asset returns exhibit:
- Low and statistically insignificant exposures to market, size, value, profitability and investment factors. Nor do they exhibit meaningful loadings on 155 other equity factors. Compared to Bitcoin, Ripple and Ethereum have higher 1-factor alphas, smaller betas and stronger exposures to the value factor.
- No significant exposures to any precious metals, except Ethereum to gold.
- Low and statistically insignificant exposures to macroeconomic factors, except Ethereum to durable consumption growth.
- Regarding crypto-asset return predictors:
- There is significant intrinsic (absolute or time series) momentum at daily and weekly frequencies for all three cryptocurrencies. For example, average next-week Bitcoin return for the top (bottom) fifth of last-week returns is 11.2% (2.6%) with Sharpe ratio 0.45 (0.19). However, the effect for Ethereum is less significant than those for Bitcoin and Ripple.
- Investor attention predicts future returns over horizons of one to several weeks. For example:
- Average next-week Bitcoin return for the top (bottom) fifth of last-week Google search on “Bitcoin” is 11.2% (1.1%) with Sharpe ratio 0.48 (0.08).
- A one standard deviation increase in the ratio of Google searches on “Bitcoin hack” to “Bitcoin” predicts a 2.8% decrease in Bitcoin return next week.
- There is little evidence that a proxy for price-to-“dividend” ratio (based on Bitcoin wallet count), realized volatility or cost of mining predict crypto-asset returns.
- Consumer Goods and Healthcare industry returns have significantly positive exposures to Bitcoin returns, while Fabricated Products and Metal Mining industry returns have significantly negative exposures. Finance, Retail and Wholesale industry returns have no exposure.
In summary, evidence indicates that crypto-assets have little or no relationships to traditional asset classes, exhibit some predictability based on short-term momentum and investor attention and are more important (positively or negatively) for some industries than others.
Cautions regarding findings include:
- Crypto-asset markets are immature, such that past behaviors/returns may not be representative of the future.
- Above analyses of crypto-asset return predictors are in-sample. An investor operating in real time may have drawn different conclusions at different times.
- Relating the same crypto-asset data to many different variables introduces snooping bias, such that the strongest findings overstate expectations.
For more perspectives, see “Crypto-asset Research Survey”.
