CryptoHawk AI Long-Short Strategy — AI & Trading Strategy in Action

CryptoHawk AI

CryptoHawk AI Long-Short Predictions

Machine learning and data science depend heavily on significant time-series data to identify and leverage the patterns to achieve meaningful predictions. The CryptoHawk AI Long-Short prediction model is based on Bitcoin and Ethereum because they have the most history compared to other coins. The resulting model, which includes many market indicators and co-integration, is based on what has been a relatively bullish history. This is why the model performs well in detecting buy signals, whereas the sell/short signals have been less effective and favor investors with shorting options.

After more than a year and a half of predictions and disruptive crypto market volatility, our data science team has experimented with several different models and strategies to leverage our Long-Short signals.

This document will help you combine the strongest and most consistent predictions with a buy and stop-loss strategy.

CryptoHawk AI Long-Short Buy & Stop-Loss Threshold Strategy

As its name indicates, the CryptoHawk AI Long-Short model was modeled to consider short selling. The Buy & Stop-Loss Threshold Strategy acknowledges that most casual traders do not engage in short selling. Our data science team analyzed the performance of the predictions and applied two thresholds, one for one upper gain limit and one for a lower sell/stop-loss limit have been established.

In this exercise, we display several upper and lower limit thresholds. This strategy uses the first buy signal from the machine learning (ML) strategy and holds it until it reaches the upper or lower limits. A sell signal is then generated when gain/loss reaches the established thresholds.

When the gain reaches the lower threshold limit, we check what the suggestion from the ML strategy at that point is. If it also calls a sell signal, then we show High confidence; otherwise, we are moderate about our call. As we move forward, we will see the comparison of different strategies during different time intervals of the test dataset. The Moving Average (MA) and ML are used in our research to establish a baseline and analyze our predictions. The threshold strategies are the ones we are demonstrating to illustrate the various performances. We also apply a set fee of 0.1% for each transaction.

2020–12–01–2021–01–31 Analysis for BTC-USD

By applying the different threshold distributions to the predictions to mark a sell and stop-sell position, we discovered that the strategy offered significant gains compared to a market hold strategy. The most impressive gains were measured on the 12% and -3% settings, with an initial balance of $100 returning an end trading balance of $407.46. The following detail the different distributions applied to the strategy.

Threshold at 3% & -1%

profit 3% loss 1%
Metrics
Formula
Value
init_balance
Initial balance
$100
end_balance
After trading balance
$219.20
win_rate
winning trades / total Trades
0.3142
profit_factor
sum_win_returns/sum_lossing_returns
1.0446
win_loss_ratio
avg_win/avg_loss
2.4148
total_trades
Total Trades
938
winning_trades
Winning Trades
295
losing_trades
Losing Trades
643
mean_of_returns
np.mean(close.pct_change().dropna())
0.0001
std_of_returns
np.std(close.pct_change().dropna())
0.0088
AI_sharpe_ratio
mean_of_returns/std_of_returns
0.0143
VBT Sharpe Ratio
sqrt(periods)*mean(returns)/std(returns)
1.1687
CPC Index
profit_factor*win_rate*win_loss_ratio
0.7926
moderate
exit false by model but SL hit
519
excellent
exit true by model and SL hit
124

Threshold at 6% & -2%

profit 6% loss 2%
Metrics
Formula
Value
init_balance
Initial balance
$100
end_balance
After trading balance
$204.41
win_rate
winning trades / total Trades
0.2987
profit_factor
sum_win_returns/sum_lossing_returns
1.0663
win_loss_ratio
avg_win/avg_loss
2.7086
total_trades
Total Trades
374
winning_trades
Winning Trades
112
losing_trades
Losing Trades
262
mean_of_returns
np.mean(close.pct_change().dropna())
0.0001
std_of_returns
np.std(close.pct_change().dropna())
0.0088
AI_sharpe_ratio
mean_of_returns/std_of_returns
0.0143
VBT Sharpe Ratio
sqrt(periods)*mean(returns)/std(returns)
1.0673
CPC Index
profit_factor*win_rate*win_loss_ratio
0.8649
moderate
exit false by model but SL hit
205
excellent
exit true by model and SL hit
57

Threshold at 9% & -3%

Threshold Profit 9% loss 3
Metrics
Formula
Value
init_balance
Initial balance
$100
end_balance
After trading balance
$266.76
win_rate
winning trades / total Trades
0.3081
profit_factor
sum_win_returns/sum_lossing_returns
1.0993
win_loss_ratio
avg_win/avg_loss
2.8985
total_trades
Total Trades
197
winning_trades
Winning Trades
60
losing_trades
Losing Trades
137
mean_of_returns
np.mean(close.pct_change().dropna())
0.0001
std_of_returns
np.std(close.pct_change().dropna())
0.0088
AI_sharpe_ratio
mean_of_returns/std_of_returns
0.0143
VBT Sharpe Ratio
sqrt(periods)*mean(returns)/std(returns)
1.3085
CPC Index
profit_factor*win_rate*win_loss_ratio
0.8649
moderate
exit false by model but SL hit
108
excellent
exit true by model and SL hit
29

Threshold at 12% & -4% (Highest Performance)

Threshold profit 12% loss 4
Metrics
Formula
Value
init_balance
Initial balance
$100
end_balance
After trading balance
$407.46
win_rate
winning trades / total Trades
0.3182
profit_factor
sum_win_returns/sum_lossing_returns
1.1731
win_loss_ratio
avg_win/avg_loss
3.1955
total_trades
Total Trades
131
winning_trades
Winning Trades
41
losing_trades
Losing Trades
90
mean_of_returns
np.mean(close.pct_change().dropna())
0.0001
std_of_returns
np.std(close.pct_change().dropna())
0.0088
AI_sharpe_ratio
mean_of_returns/std_of_returns
0.0143
VBT Sharpe Ratio
sqrt(periods)*mean(returns)/std(returns)
1.6952
CPC Index
profit_factor*win_rate*win_loss_ratio
1.1732
moderate
exit false by model but SL hit
73
excellent
exit true by model and SL hit
17

Threshold at 15% & -5% 

Threshold Profit 15% loss 5
Metrics
Formula
Value
init_balance
Initial balance
$100
end_balance
After trading balance
$260.80
win_rate
winning trades / total Trades
0.3133
profit_factor
sum_win_returns/sum_lossing_returns
1.1548
win_loss_ratio
avg_win/avg_loss
3.2652
total_trades
Total Trades
82
winning_trades
Winning Trades
25
losing_trades
Losing Trades
57
mean_of_returns
np.mean(close.pct_change().dropna())
0.0001
std_of_returns
np.std(close.pct_change().dropna())
0.0088
AI_sharpe_ratio
mean_of_returns/std_of_returns
0.0143
VBT Sharpe Ratio
sqrt(periods)*mean(returns)/std(returns)
1.2709
CPC Index
profit_factor*win_rate*win_loss_ratio
1.1496
moderate
exit false by model but SL hit
46
excellent
exit true by model and SL hit
11

Summary

Our Machine Learning model is best combined with strategies that are supported by backtesting and being mindful of the volatility of crypto. Financial and technical indicators, social media, regulations, seasonality, absence of traditional supply chains, and business reports contribute to positioning crypto predictions in a class of their own. During such a paradigm shift, unexpected events have disrupted how people perceive and engage with cryptocurrencies. Complementary predictions with strategies are key to a successful outcome despite the market’s high uncertainties and volatility.

Our machine learning and data science team are not only strengthening our AI models but also exploring and sharing strategies that make CryptoHawk AI a valuable predictive tool for crypto traders.

With this information and the upcoming API for our Pro CryptoHawk AI subscribers, the use of our predictions and published backtested strategies become a powerful tool for trades building their trading bots.

Contact Information

    1. Thierry Hubert, CTO
    1. ‭+1 (833) 344-4629‬
    1. thierry@digimax-global.com

Data Science and AI Contributing Team

    1. Dr. Shahzad Cheema, Chief Data Scientist
    1. Dr. Muhammad Kamran Malik, Lead Data Scientist‬
    1. Hassan Masood, Data Scientist

For more information about CryptoHawk AI, visit www.CryptoHawk.ai or download our mobile app at Apple App Store or Google Play

More Posts

linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram