The basics of the trend-adaptive algorithm are the same as the Bollinger Bands algorithm.
The medium-term trend is determined by the 120-minute weighted moving average and the 480-minute weighted moving average.
The long-term trend is determined by the 702-minute weighted moving average and the 1440-minute weighted moving average.
The timing of purchase is determined based on the mid-term trend and the lower Bollinger Band, and the maximum number of purchases, profit-cut ratio, and loss-cut ratio are determined based on the long-term trend.
As a result of backtesting with 400,000 data, we were able to confirm that it is superior to the Bollinger Band algorithm in performance.
*2021.09.29: Modified robobits_adv2.py file
I couldn't create a 1440 minute moving average with 3 minute data, so I modified it to 5 minute data.
base_candle_url5 = "https://crix-api-cdn.upbit.com/v1/crix/candles/minutes/5?code=CRIX.UPBIT.{}&count=400" .format(coin_name)
df5 = util.get_web_1m_data(base_candle_url5)
df[ 'wma120' ] = calc.get_wma(df[ 'c' ], 120 )
df[ 'wma480' ] = calc.get_wma(df5[ 'c' ], 96 ) #96*5=480
df[ 'wma720' ] = calc.get_wma(df5[ 'c' ], 144 ) #144*5=720
df[ 'wma1440' ] = calc.get_wma(df5[ 'c' ], 288 ) #288*5=1440