Do simple factor models help explain future return variations across different cryptocurrencies, as they do for stocks? In their April 2019 paper entitled “Common Risk Factors in Cryptocurrency”, Yukun Liu, Aleh Tsyvinski and Xi Wu examine performances of cryptocurrency (coin) counterparts for 25 price-related and market-related stock market factors, broadly categorized as size, momentum, volume and volatility factors. They first construct a coin market index based on capitalization-weighted returns of all coins in their sample. They then each week sort coins into fifths based on each factor and calculate average excess return for a portfolio that is long (short) coins in the highest (lowest) quintile. Finally, they investigate whether any small group of factors accounts for returns of all significant factors. Using daily prices in U.S. dollars and non-return variables (excluding top and bottom 1% values as potential errors/outliers) for all coins with market capitalizations over $1 million dollars from Coinmarketcap.com during January 2014 through December 2018 (a total of 1,707 coins, growing from 109 in 2014 to 1,583 in 2018), they find that:
- Mean (median) market capitalization of all coins in the sample is $357 million ($8.2 million). Mean (median) daily dollar volume is $18.3 million ($0.1 million).
- Average coin market index weekly return is 1.3%, with standard deviation 11.7% and fat tails.
- Strategies that are long (short) the quintile of coins with the highest (lowest) expected returns have statistically significant average next-week excess gross returns for 9 of 25 factors:
- Market capitalization (3.4% per week).
- Price at end of portfolio formation week (3.9% per week).
- Maximum price during portfolio formation week (4.1% per week).
- 1-week, 2-week, 3-week and 4-week momentum (2.7%, 3.3%, 4.1% and 2.5% per week, respectively).
- Dollar volume traded (3.2% per week).
- Standard deviation of dollar volume traded (3.0% per week).
- For all significant factors, variations of average next-week gross returns across quintiles are nearly systematic.
- A 1-factor (market) model of coin returns performs poorly explaining differences in coin next-week gross returns.
- A 3-factor (market, size, 3-week momentum) model captures most of differences in coin next-week gross returns. In other words, after controlling for these three factors, none of the nine successful factors have statistically significant gross alphas.
- Shorting Bitcoin rather than shorting the factor quintile of coins with the lowest expected returns produces similar results.
- Momentum factors are much stronger among coins with large market capitalizations than those with small market capitalizations.
In summary, evidence suggests that cryptocurrency traders should prefer small-capitalization coins or large-capitalization coins with strong returns over the past few weeks.
Cautions regarding findings include:
- The cryptocurrency market is immature (wild), so its early behaviors may not be representative of current behaviors.
- All returns are gross, not net. Trading frictions for weekly factor portfolio reformation involving illiquid coins may be very high and may vary by market capitalization, such that net findings may be very different from gross findings. Moreover, as noted in the paper, shorting is likely infeasible for many coins and may be costly where feasible (but shorting Bitcoin is an alternative).
- Testing many factors on the same sample introduces data snooping bias, such that results for the best-performing factors overstate expectations. Also, choosing the optimal 3-week momentum for the 3-factor model imports bias into the 3-factor model.
See also:
