Are simple moving averages (SMA) useful for timing difficult-to-value Bitcoin? In their January 2018 paper entitled “Bitcoin: Predictability and Profitability Via Technical Analysis”, Andrew Detzel, Hong Liu, Jack Strauss, Guofu Zhou and Yingzi Zhu investigate the use of 5-day, 10-day, 20-day, 50-day or 100-day SMAs to predict Bitcoin returns. Specifically, they test a trading strategy that holds Bitcoins (cash) when current Bitcoin price is above (below) a selected SMA. They assume cash earns the U.S. Treasury bill (T-bill) yield. Using daily Bitcoin prices and T-bill yield, along with data for other variables/assets for comparison, during July 18, 2010 through December 12, 2017, they find that:
- Buying and holding Bitcoin earns annualized daily excess (relative to T-bill yield) return 224%, with annualized volatility 111%, Sharpe ratio 2.0 and maximum drawdown -89.5%.
- Bitcoin returns are predictable using any of the price SMAs tested, such that the simple trading strategy described above:
- Has gross annualized returns in the range 203%-216%, with annualized volatilities in the range 83.8%-97.0%.
- Has gross annualized Sharpe ratios that decrease systematically from 2.6 for a 5-day SMA to 2.2 for a 100-day SMA .
- Suppresses gross maximum drawdown to the range -70.3% for 50-day and 100-day SMAs to -66.4% for a 5-day SMA.
- Has breakeven Bitcoin-cash switching frictions increasing systematically from 1.1% for a 5-day SMA to 4.5% for a 100-day SMA.
- Trading strategy performance is robust for the first and second halves of the sample period.
- The same trading strategy also beats buy-and-hold on a gross basis for the NASDAQ Composite Index during 1996 through 2005 when dot-com stocks are hard to value, increasing gross annualized Sharpe ratio from 0.25 for the index to the range 0.42-0.71 and suppressing maximum drawdown -77.9% for the index to as low as -25.7%.
- It is arguable that investors trying to interpret prices cause short-term trending for hard-to-value assets.
In summary, evidence indicates that SMAs are effective timing tools, on a gross basis, for hard-to-value assets such as Bitcoin.
Cautions regarding findings include:
- Reported returns, Sharpe ratios and maximum drawdowns are gross, not net. Bitcoin to date has exhibited long execution times and large trading frictions, potentially exceeding the breakeven frictions noted above.
- Bitcoin market depth may not have supported large dollar trades across much of the sample period.
- Testing multiple strategy variations on the same sample introduces data snooping bias, such that the best-performing variation overstates expectations.
