Bakker (2017) found numerous significantly profitable trading strategies. However, profitability is highly unstable and declines over time.

Cutting away all the scientific and mathematical jargon it appears that none of the trading rules is significantly profitable after 2014. To understand what this means you must understand the scientific notion of significance. Simply put, it means the result was not due to chance behavior. Bakker (2017) essentially destroys current popular views on TA.

The strategies under scrutiny were:

• Double moving average
• Support & resistance
• Channel breakout
• Relative strength indicator
• On-balance volume average
• Bollinger bands

Each of these trading strategies were compared to a buy-and-hold benchmark strategy to compare the effectiveness of each single trading rule. The average number of trades done over all significant trade strategies equal 34.336 — roughly one trade per 70 minutes. What was interesting is the facts that the length of the holding period had less influence on the significance of each trading strategy — despite the lower number of transaction costs.

Bollinger Bands proved to be the most effective in the study of Bakker (2017) and provided the most profit on a 5-minute interval. The trading rule manages to both effectively exploit both upward and downward price deviations. The trading rule is primarily profitable in periods of high volatility, especially when volatility peaks.

Bollinger Band strategy was only effective when exploiting many small price fluctuations.

HOWEVER, results indicate that excess returns are highly unstable over time. The ratio of good vs bad trades based on a BB strategy is especially remarkable. The model had over 56k of profitable trades compared to around 17k losing trades. This might seem to paint the BB-strategy as highly effective. But, the average loss of a losing position is considerably higher than the average return of a profitable trade. The average duration of a losing position is also twice as large than the average duration of a profitable trade. This makes the BB-strategy only effective when exploiting many small price fluctuations.

Another problem occurs however when you trade based on a high-frequency BB-rule, transaction costs. The impact of these costs completely erased the profitability of the best performing trading rule.

Note that the research is based on averages from 2013–2017 on the Bitcoin transaction costs from the BitStamp exchange. With the implementation of Segwit — and thus lower transaction costs these results may now be different.

Note however that, due to the presence of transaction costs all long positions were losing compared to the buy-and-hold benchmark strategy. The mean return of a profitable trade using BB was substantially higher than the mean loss of a losing trade however. This makes using Bollinger Bands an effective strategy, though not more effective than the buy-and-hold strategy.

While Bakker (2017) did identify profitable individual trading rules, none of these used in isolation will be able to predict all price movements. As picking only one leads to loss of information from other trading rules. To account for this Bakker developed a complicated neural network incorporating many of these rules into a complex trading strategy. Bakker’s trading rule outperformed every single trading rule and is impossible to trade on manually.

In conclusion, profitable trading based on a technical analysis is most likely left for machine-learning algorithms and not for the ‘popular chartists’ on Twitter. Despite some predictive power in TA it seems ‘hodling’ might be more effective after all.