Last Updated on
The Bitcoin Machine Learning Anomaly Detection Model (aka BTC Beta) has successfully exited beta. As a result, it is now deployed as the production model for Bitcoin trade signals. This has some important implications for those using Bitcoin Auto Trade.
This post will cover the reason behind the change, what to expect, and considerations for existing customers.
The Bitcoin 4.x machine learning model
In October of 2019, we released a new machine learning model for Bitcoin trading signals. This model was based on an entirely new approach, designed to identify anomalies in the market that tend to occur before abnormal swings in price.
To learn more about how the model works:
- Overview: Bitcoin Anomaly Detector
- Detailed view: Bitcoin Trading with Machine Learning Anomaly Detection
Its key advantage is that it is very robust and works in a broad set of market conditions without needing an adjustment period.
Our original models (up through 3.x) learn and adjust to changing market conditions, but this can mean there will be situations in which those models take time to adjust.
For example, at the end of 2019, Bitcoin Cash dropped around 63% before making a fast turn around. This sustained 63% drop taught our 3.x model to be very conservative and protective. As a result, it did not jump in on the recent upswing. The new 4.x model is able to better react regardless of market conditions and changes.
The BTC Beta period
Most of our customers know the 4.x model as “BTC Beta.” We initially released the 4.x models as beta, allowing customers to see its performance and also use it with Auto Trade.
We have tracked its performance publicly on our Trade History page.
The primary finding during the beta period is that the model was too aggressive. This aggressiveness might work out in the long run, but it can create unacceptable losses in the short run.
Taking this feedback, we were able to modify some boundary parameters and let our tools work on new solutions. It was able to tweak slightly, maintaining similarly great performance while also cutting losses sooner. This became the 4.1 model. There is a great discussion on this topic in our Community Forums here: BTC Beta Stop Loss Has Tightened.
Successfully exiting beta
After seeing the performance in the real world and making this tweak, it was agreed the 4.x Model passed beta and was ready for production.
Additionally, most of our Auto Trade customers have already chosen to follow the BTC Beta model.
The final nudge was seeing the possibility of “learning lag” that can occur in the 3.x models (as shown by BCH and LTC over the last couple of weeks). This shows how the robustness of the 4.x models is critical to have in these ever-evolving markets.
Changes to the Member Dashboard
As you will note, there are a couple of changes on the Member Dashboard:
- All references to “BTC Beta” are removed
- The BTC signal information now matches the 4.x data
- BTC is removed from the Trade Meter
- BTC is graphed with the Anomaly Detector Visualization
Changes to the Auto Trade Dashboard
- “BTC Beta” is no longer a trading pair selection
- If you were on “BTC Beta,” you are now on “BTC”
- If you were already Auto Trading against “BTC” you are still on “BTC” but now following the 4.x model
What you need to do
Please access the Auto Trade Dashboard to confirm your Auto Trade is still active and shows BTC.
This picture is of an account that was trading BTC-Beta-USD. After the upgrade, it properly reflects BTC-USD as the trading pair.
Please check your Auto Trade dashboard.
If you were on BTC Beta for Auto Trade, your Dashboard should now look like the one above. You should no longer see the word “Beta.”
If you face any issues with this Auto Trade transition, please Contact Support.
What’s next for Crypto-ML’s machine learning
While we have a large list of research and development areas, we will be focused on delivery 4.x model upgrades to each of our other support cryptocurrencies:
- Bitcoin Cash
- Bitcoin Short