Release 7 of Crypto-ML’s Machine Learning Trading System
Last Updated on February 17, 2021 by admin
Crypto-ML is pleased to announce Release 7 of its machine learning crypto trading platform. This upgrade is different than past releases. Not only does it bring a new machine learning component to our trading system, but it also launches an intelligent switching system that can select between various strategies for opening and closing trades.
This upgrade contains one of our most exciting innovations yet. As far as we know, this is a completely novel approach in the field and not used in any other financial ML platform.
New Machine Learning Approach: Manipulation Detection
The most immediately obvious change with Release 7 is a new trading system that uses our existing neural networks but interprets the output in a different manner.
Specifically, this system generates trades when it detects the markets are behaving abnormally.
Internally, we dubbed this “manipulation detection” as it was identifying when the markets were being pushed in an unusual manner, not following traditional trading patterns.
Whereas our existing system (Release 6 and earlier) is particularly effective at understanding typical market moves, this new system is adept at catching those unusual times that result in large breakouts and crashes.
This new method accepts two key realities:
- Markets generally move in a reasonably predictable manner, but key moments express chaotic, unpredictable behavior.
- Machine learning and statistical models are good at understanding core behavior and patterns but are less adept at understanding edge cases.
The new approach attempts to identify times when we are moving into chaotic edge cases and then predict where price will move.
This concept can be broken down as follows: let’s assume machine learning can predict 90% of behavior. But perhaps it’s in that other 10% that the markets really move strongly. If we can identify when we’re in that 10% and act accordingly, we’ll be able to take advantage of the big market movements.
Rather than trying to control chaotic areas of statistics and market movements, this new approach embraces those realities.
These systems have actually been running live on our backend. The results are here:
Please note, this is not back testing. These are real actual results generated against production data. We’re providing here to give you a sense of what to expect.
Shortly after going live, this system captured a rally resulting in its first two end-to-end trades:
- BTC +12.75% in about 12 hours
- ETH +9.81% in about 12 hours
As always, this new approach builds upon our previous systems. In essence, it is taking advantage of existing capabilities in a new way.
Intelligent Strategy Switching
With this release, we are not replacing our existing mechanism with the new “manipulation detector” mechanism. Rather, they are both running in parallel.
An additional new component now monitors these mechanisms and can issue trades from either, depending on the market conditions.
Further, as new approaches and strategies are added into the mix, this intelligent switching will cycle between them, with the goal of delivering the highest-probability trades for any given condition.
Differences You Can Expect
In the initial phases of Release 7, you may notice some differences in the trading style.
Trades issued by the “manipulation detector” will exhibit these traits:
1. May result in larger gains and losses
This system lets rallies go, whereas our traditional system sets profit targets and captured small wins along the way.
Likewise, the stop system is very different and has the potential to generate a larger loss, though this is closely monitored.
2. Better ability to capture rallies and avoid crashes
By its very nature, the manipulation detector is designed to predict and act on those exceptional moments in the market. As a result, it should better capture exceptional rallies and avoid the market during exceptional drops.
3. Will not publish stops or profit targets
All controls are internal to the system and are dynamic. More details may be shared on the Member Dashboard in the future, but currently, when the “manipulation detector” issues a trade, you will notice less information is available.
4. Similar trade frequency
Regardless of the bot issuing trades, we should see a similar frequency of trades as we have over that latter half of 2020. There may be periods of rapid trading and there may be periods of quiet.
Because of this, we recommend tightly managing your trading fees. Read out post on how to reduce crypto trading fees for more.
5. Crash Protection may not be used
We recently released a “Crash Protection” feature that is designed to work as an emergency feature during market crashes. Please watch for additional updates in this area.
Recommendation: You may always place your own stop losses on the exchange. You can also exit trades at any time. Doing this will not interfere with our Auto Trade system.
6. API Changes
If you are a Developer or Professional customer leveraging our WebSocket API, you will notice differences in output.
Namely, you will see a new “bot” entry identifying which strategy issued the trade. If “Manipulation” generated the trade, you will not see target and stop entries.
Please check the API Documentation for changes.
Crypto-ML is constantly pushing the envelope and developing novel ways to trade cryptocurrency. These changes will provide an immediate and noticeable impact but will also set our platform up for unparalleled flexibility going forward.
Crypto-ML provides machine learning for crypto traders and investors. Gain crystal-clear signals and deep market insights. Add predictive capabilities to your toolbox. Learn more and join for free.