As you may have learned from our previous blog posts (read 1, 2) or Youtube videos (watch 1, 2), there are many differences between the forex and crypto markets. They define the modeling methodology, or how we trade, and what strategies we use with an emphasis on three parameters: the size of moves, liquidity, and price patterns.
In the crypto market, the main approach we take is the trend following signals long-term, and we optimize it on a daily basis to make sure that we enter and exit at the right times.
In the forex market, we do the opposite – short-term swing.
And the reason for it lies in the price patterns difference.
Price patterns difference
There is a standard 80/20 rule saying that markets trend about 20% of the time and spend the other 80% grinding through the trading ranges, pullbacks, and other counter-trend actions.
What is the main idea of swing trading that we apply at EndoTech?
In crypto, there are 30% to 40% of the trend moves. While in forex, there are only 5% to 10% of trends.
Finally, while crypto trend moves are clean and almost noiseless, forex trends are comparable in size to the countertrend actions. Trends are only two or three times larger than the noise. It is easier to take advantage of the noise than of the trends.
At EndoTech, we developed several proprietary indicators that identify the temporary overbought and oversold situations. We use them to enter the trades. For the exits, we combine these indicators with time-based stop loss and profit target exit rules.
Goal function
The main aspect of the modeling is to define a goal function. It means to define the success or failure of the model. Since we only see the past, our goal is not about profitable modeling of the past. We search for the best identification and utilization of the patterns.
Similar to all other professions and their respective fields we aim to be accurate and create success for our clients. For example, physicians’ final goal is to keep us healthy. They are as successful as many issues they identify and treat. Exactly the same way, our goal is to reach an accuracy of at least 66% of predictions. We also have to lose on average no more than 1.5% of an average winning trade. This is enough to make a winning strategy.
Profit/Loss is equal to the number of winning trades multiplied by the average profit of the winning trades minus the number of losing trades multiplied by the average loss of the losing trades. If a 66% accuracy is reached, and the average win/loss is sufficiently small – this strategy will be profitable and quite stable.
Pros and cons of the swing trading
Swing trading is much more professional and demands attention to detail. Swing trades should have perfect timing for entry and exit. While in our crypto following systems we can make errors equal to 20%-30% of the exit. In swing trading every percent counts, trades are shorter and there are more trades overall.
And here comes the fun part of what artificial intelligence does for us. The devil is in the details and the artificial intelligence helps us understand them. Because many professional traders and bots are involved, and they use advanced information about the market. So should we.
There are predefined economic events and the market knows how to react. A week with no economic events or one event at its end, will dramatically differ from a week full of events.
Global trends
The second aspect concerns the global markets trends that are dictated by the global trends in the economy.
And this is where the D.AI.SY scientific project of fundamental analysis comes into picture.
Our team of AI financial analysts and quants designed their model for the EUR/USD pair. It predicts the next week’s price movement and volatility. They use the feature selection to decide what market economic indicators in the U.S. affect the EUR/USD price. The models are built with KNN, RandomForest, TableNet, and XGboost .
The acquired result of the best model surprisingly was at an accuracy level of 86%. Then we combined swing trading economic calendar and the fundamental analysis model. And we reached the goal of a 66% winning rate with less than 1.5% in average win/loss rate.
