The popularity of cryptocurrencies grows, and so do the startups with brilliant ideas on exploiting this lucrative market. Most of them fail within two years, but others achieve high financial success.
When we talk about trading bots, it’s no different — most old-school algorithm bots failed miserably, while the fresh ones powered by AI are under everyone’s radar. Today we’ll give you a crash course on machine learning and how to take that knowledge and pick a winner out of the sea of new crypto bots.
First, we’ll destroy one common misconception about AI. Trading bots until now, no matter how complicated their structure is, don’t really use artificial intelligence. They rely on mathematical formulas and indicators such as moving averages to make decisions. One Button Trader is the first publically available automation tool for trading that is 100% powered by AI and machine learning.
Like every other technology, AI has its foundation. In this case, it is neural networks. They are a complex mix of non-linear functions that would make even the best mathematician struggle to solve them. Luckily, machines are very efficient and can handle such tasks with minimal expert knowledge.
There are different strategies to train AI bots. We prefer to use an evolutionary one (think of Darwinian Theory) as it has shown the most promise in both literature and practice.
A simple example of an evolutionary strategy:
· Creation of Master model (Let’s call it Adam)
· We create N number of mutated models from Adam by applying random noise (we make ourselves some imperfect clones)
· Next, we evaluate those clones in the environment for testing.
· We adjust the Master model by the weighted sum of the reward multiplied by their respective random noise (Adam feeds from the knowledge and mistakes of his doubles)
· We loop back to step 2 until satisfied (Adam achieves godlike status)
Now comes the question: If we can train a single AI to become so good at something, why do we need that many variations of bots? Well, the answer is simple — while every trading bot uses an evolutionary strategy to learn, not all of them use the same pathway, structure & learning curve.
Some of them are very efficient in computing tasks, while other show promises with good compute-to-memory usage. In every case, one thing is for, and it’s that not a single bot created on the OB Trading platform is exactly identical to another one.
Trading AIs from one family adapt to the environment by the same methodology but may trade differently, based on when they first entered the market and began their learning.
Keep reading to find out more differentiations and how to make the right pick for you!
So the time for you to make a pick has come, but how can you decide? All the bots at OB Trader seem adorable, and the only difference for an untrained eye would be their names. Well, that couldn’t be any further from the truth because there are plenty of technical aspects that differentiate the bots.
Let’s look at several of them:
· AI Model
· Trading Strategy
· Market Scan Frequency
· Available Trading Pairs
· Use Of Limit Order Trading
Currently, we rely on 3 different types of AI mechanisms: Filter, GRU & Transformer, all of which you can read in-depth in our Whitepaper.
Initially, we experimented with a basic ML model named Filter for our early bots, such as Astral. By using a simple 2 parallel linear layers system, accompanied by 3 FilterBlocks for great compute-to-memory usage we saw decent results in our initial tests.
Check out Astral v2 Case Study! For simplicity and a proven track record, you can go with this group of bots.
Later, we switched to a more complex multi-layer GRU model that allowed the following (Horizon & Ascendant) bots the ability to reuse some computation from previous steps. This allowed for more efficient computation as well as removing bias from historical data.
The results — Horizon Bot — Trained to Buy Low and Sell High. If you want balance and great risk management, then this is the right choice!
Finally, we took the latest and most powerful AI model Transformers and used the FAVOR+ element to approximate attention in our Performer bot. This is our most promising bot up until now. If you want to go with the latest technology, then we strongly recommend bots from this series.
Future: We’re constantly testing new AI, optimizing previous ones & developing easier guidance on how to pick the best bot for your goals.
If you’ve read our article on Why Choose an AI Trading Bot, you’re familiar with our 3 categories of trading strategies — safe players, bull machines & pirates. Here’s a quick refresher:
· Safer Bots With Great Risk Management
Here we got Astral, who generalizes well on uprising markets and uses an early stop mechanism for increased safety. The other two great choices are Ascendant & Solar, both of which take only moderate positions, minimizing the risk and showing stable results in different market conditions!
· Bull Machines For Max Gains During Uptrends
Clipper was specifically designed to “clip” all the extra profits during extreme bull runs. Thanks to his AI, he can identify trends in their start and grab the tops of the price hills.
Performer, on the other hand, uses up to 4 times more data inputs than other bots. He rarely misses in making extra profits. What’s more fascinating is that he demonstrated solid control during bear markets and saved several investors’ portfolios.
· Finally, we look at Horizon, who crusades sideways as a pirate.
He adapts quickly, trades between candlesticks, and is the only one in the family to use Limit Orders. He doesn’t fear sudden price changes and has shown great success in sideways markets.
There are endless possibilities for AI trading bots. While most traditional bots failed, the majority of the new high-tech ones show promise right from the beginning. You can choose from different Wall-Es based on the ML tech, historical success in a particular pair, or current trade of the market.
As of Oct 2021, our bots scan the market either every 1H or 4H, but we are working on more variants. We are also mastering the stop-loss mechanism of our bots and are secretly developing the first-ever AI Bot That Shorts. Stay on track with everything on our roadmap.
We regularly prepare insightful reports and case studies about crypto trading and the blockchain industry.
Congrats! You successfully joined One Click Crypto waiting list. Keep an eye on your inbox, you will get updates soon.
In the meantime, join our communities to be extra cool