Crypto AI trading bots: A beginner's guide

• Algorithms dominate many large traditional markets, generating billions in profits for major firms. 

• Artificial intelligence (AI) can enhance these algorithms by: a) Identifying novel opportunities, b) Responding to large amounts of data, and c) Learning adaptively based on performance.

• Crypto traders looking to deploy their own algorithm have two options: Use a subscription-based service or bot marketplace, or create their own bot from scratch by coding their own indicator or execution software.

Crypto AI trading bots leverage sophisticated mathematical models, machine learning algorithms and automation to execute trading strategies on behalf of traders. 

These advanced systems continuously analyze vast amounts of market data, identify patterns, and adapt their strategies in real-time to optimize performance.

In contrast, traditional pre-programmed algorithms operate based on fixed parameters and follow a backtested script. While these conventional algorithms can be effective, they lack the ability to independently evolve and adapt to changing market conditions. This rigidity can limit their effectiveness in dynamic and unpredictable cryptocurrency markets.

AI-driven bots represent a significant upgrade for automated trading systems. Their ability to learn from new data, adjust strategies on the fly, and predict market movements with greater accuracy expands the functionality and productivity that automated trading systems can offer.

1. Several academics expect AI to impact many aspects of our lives, including financial markets. For crypto traders, AI offers the ability to autonomously execute data-driven trades, which is particularly beneficial for those unable to monitor markets closely.

2. Reports indicate that around 60-75% of trading volume in major US, European, and Asian traditional financial markets is generated by algorithmic trading. It is possible that a significant percentage of crypto trading volume is also driven by bots. Therefore, it is advantageous for traders to understand how these systems work and how to integrate them into their investment strategies.

3. Through machine learning, AI has the potential to enhance the performance of trading algorithms. These advances may expedite the process of identifying, testing and deploying strategies, while also being able to adapt to an evolving market.

AI-driven algorithms are a form of systematic trading where traders follow a rules-based approach that may involve using indicators and automation to execute trades. 

This approach contrasts with discretionary trading, where traders make subjective decisions based on various sources of information about the future direction of a market.

While both systematic and discretionary trading have their merits, AI-driven trading bots introduce several unique benefits:

• Algorithms remove many of the human flaws present in discretionary trading. They are consistent, data-driven, do not have emotions, do not sleep, and can monitor multiple markets simultaneously without missing key moments. 

• AI means that algorithms may be able to do the job of a discretionary trader and more, by examining historical data to generate and backtest strategies, while also being able to learn adaptively by reviewing performance. AI is able to do this because it excels at crunching the numbers on large data sets and identifying outliers. 

• Regular algorithms can execute trades much faster than their human counterparts. AI-driven bots are therefore not only faster, but they may be able to use multiple streams of information to respond intelligently to major market events.

AI influences how algorithms (and traders) behave in a number of ways:

• AI is used to combine, monitor and assess a vast number of hugely divergent streams of data. This enables AI-driven algorithms to respond rapidly to important shifts in market sentiment, price, economic forecasts and major geo-political events. 

• AI can identify common technical patterns in price charts, which in turn can be used to generate setups.

• Therefore, AI might be used to quickly close or enter a trade, automatically adjust the distribution of an investment portfolio, or provide valuable reporting to key decision makers.

• Via machine learning - a subfield of AI - AI can combine large tranches of structured and unstructured data to make predictions

• Through adaptive learning, AI can be employed to improve the performance of algorithms over time. This is similar in principle to a discretionary trader reviewing their trading journal to identify what they are doing well, and what needs to be improved. 

• Identify patterns that can be combined with traders’ intuition to make better decisions

If you are interested in learning more about real-world, AI-driven products, SNP’s Kensho and BlackRock’s Aladdin are two projects that employ AI for business insights and portfolio management, respectively, as well as various other functions.

The following list summarizes some of the published research relating to the use of AI in crypto markets:

• One study demonstrated that researchers were able to predict Bitcoin (BTC) movements using machine learning with 66% accuracy. 

• A similar study found that machine learning could predict the daily market movements of 100 leading cryptocurrencies with 52.9% to 54.1% accuracy.

• The use of neural networks in experimental conditions also showed promise for predicting the price of Bitcoin.

In a recent article, Forbes examined the role of AI in crypto trading, referencing the following projects:

• SingularityNET: is a marketplace for AI products, including bots that can be used for market analysis. 

• GNY.io: is looking to build ‘...predictive machine learning tools for the crypto trading community.

Another widely referenced AI trading bot platform is Cryptohopper, which claims to have inbuilt artificial intelligence to help decide which strategies to deploy.

All trading systems, even AI-driven crypto trading bots, are prone to various risks. 

Given that algorithms are directed by computer code, the degree to which they are secure depends on who wrote the software, whether it has been independently reviewed and how often it is revised.

Experienced traders/developers who create their own trading bots may have greater confidence in their reliability and security. Because self-coded bots are completely transparent, they allow a trader to understand and control every aspect of the bot’s operation, making the whole system potentially more secure. 

In contrast, third-party, subscription-based trading bots pose several inherent risks:

Hacking

Third-party software is vulnerable to hacking. If you connect any software to your trading platform via an API, and someone manages to take control of that software, that person effectively controls your account. This may allow them to use your capital however they like. 

In March 2018, after taking control of numerous accounts via Binance’s API, hackers used the funds in those accounts to pump the price of Viacoin. More recently in 2023, the trading bot platform 3Commas was exploited, resulting in $22 million being siphoned from users’ accounts.

Failure

Algorithms, like humans, are susceptible to errors that can stem from various factors. Flawed coding or disconnection from a trading platform’s API can result in an algorithm failing to act as it is programmed to. For example, if an algorithm enters a long position and the stop-loss is managed by the software, a failure in the software could prevent the position from closing during a sharp price drop. This risk is especially pronounced when using leverage, potentially leading to catastrophic losses. There are numerous documented instances of algorithm failures in both traditional finance and crypto markets, resulting in substantial losses.

Lack of edge

An algorithm may appear to be profitable based on backtesting, but when the strategy is deployed live into the market, it does not replicate its historical performance. As the saying goes, “past performance is not indicative of future results,” and this holds true for algorithms. 

Backtesting itself is subject to many pitfalls such as over-fitting, where you ascribe more credibility to a strategy than actually exists. If you repeatedly backtest and revise a strategy using the same small tranche of historical data, you are likely to be misled into thinking you have an edge. In fact, all you have found is a strategy that worked only during that discrete phase of price action. Therefore, when that same strategy is deployed into a forward test, it performs poorly, because it is not informed by the full scope and breadth of market behavior. Rather, it operates using only a small snapshot of market behavior, which is merely noise in the grand scheme of things. 

Scams

Scammers have used the guise of trading bots that promise significant returns to extract value from unsuspecting retail traders. The CFTC cautions against buying into “AI-created algorithms,” as scammers have exploited public interest in this field, resulting in billion-dollar ponzi schemes.

For more information, check out our Kraken Learn Center guide, How to keep your crypto safe.

This topic could merit several articles on its own, but here is a simple guide that describes the basic framework:

Step 1: Identify an edge

Perhaps after spending many hours observing price action in crypto, you notice that the markets tend to behave in a repeatable pattern. In order to test this theory, you backtest this pattern over a few years of historical price data to see if this amounts to an edge. There are many online digital packages available that can assist you with this process. Alternatively, you can manually log every instance where the edge being examined works or not, using some charting software and a spreadsheet. If after extensive backtesting, the pattern in question shows merit, you may decide to proceed to the next step. 

Step 2: Program your algorithm

Assuming the edge you've tested can be converted into an algorithm (some highly discretionary strategies cannot be easily automated), you may wish to start the process of creating or finding software that can:

• Identify the setup you are looking to trade.

• Execute the trade exactly as desired, including the entry, stop, and take-profit levels.

Consider the following:

• What programming language to use.

• Which trading platform to use.

• What type of orders will be required.

• How to handle errors, etc.

This step is complex and cannot be fully covered here, but there are ways to expedite the process:

• TradingView allows you to code your own indicators, which, when paired with off-the-shelf execution software, can form a complete automated trading system.
• ProfitView is an application that enables traders to use TradingView signals to trade automated strategies on crypto exchanges. Note that Kraken does not endorse these services.

Step 3: Forward test

Once you have: a) A backtested strategy and b) A way of executing that strategy algorithmically, you are now ready to test it live in the market. This is known as forward testing. Using a very small amount of capital to prevent unnecessary losses, you can deploy the strategy and track the results. The amount of time you will need to forward test a strategy will depend on how much data you collect, or how often it makes trades. 

Step 4: Review

After deploying the strategy in the market, it’s time to review the results. Assess whether it performed as expected and consider adding additional filters or variables to improve the overall strategy.

Step 5: Monitoring

Once you are happy that the algorithm is performing within a range of expected performance, you may feel it’s time to deploy more capital. You will still need to monitor its performance over time and potentially make iterative improvements.

Note, some algorithms will stop working over time for no obvious reason. If it continues to lose funds over an extended period, you may have to reassess whether you want to continue running it.

Whether you are a complete beginner or an experienced trader, integrating AI into your trading regime may offer numerous advantages. Here are several ways to incorporate AI into your trading strategies:

1. Integrate Data from Generative AI Models: Utilize generative AI language models, such as GPT-4, to monitor news and market data. These models can make predictions based on this information, which can then be integrated into your trading algorithms.

2. Conduct Sentiment Analysis via Natural Language Processing (NLP): Track the language people use in relation to crypto markets using NLP. Analyze historical patterns to see if there are correlations with price action, which can help in making informed trading decisions.

3. Scan for Potential Setups: Use AI to identify common technical price patterns and potential trade setups. AI can quickly analyze vast amounts of data to spot opportunities that may not be immediately apparent to human traders.

4. Generate Market Reports: Employ AI to generate reports on key market events or summarize large datasets. These reports can provide insights into how markets responded to similar events in the past, aiding in future trading decisions.

5. Backtest Strategies: Leverage AI to backtest trading strategies by examining specific sequences of price action over large datasets. AI can efficiently process historical data to evaluate how well a strategy would have performed, helping refine and optimize trading approaches.

• Ensure that good risk management is employed at all times and where it matters. Noting the risks above, trading bots can have catastrophic failures and this should be taken into consideration when deciding how much capital to deploy. You can reduce counterparty risk by spreading your capital across multiple trading platforms.

• Consider using an equity-curve based stop loss. More simply, if the bot continues to lose funds over an extended period beyond what you would expect from your backtesting, consider turning it off to review its performance. In truth, you can never know whether an edge has fully decayed, but you can put measures in place to cap your losses.

• Equally, don’t turn off an algorithm before it’s had a chance to perform. It’s normal for any strategy to lose for a certain amount of time (this is known as drawdown). Your backtesting should inform you as to how large and how long these periods of poorer performance tend to last. If you frequently stop a strategy from running when it is losing capital, you may be simply doing so during a period of expected drawdown, and the system may recover immediately after.

• Just as investors use diversification, you may do the same as an algorithmic trader. Some bots will perform well when the market is ranging, and some bots will perform well when the market is trending, but few will perform well in both phases. By having a range of bots that complement each other, this can create a smoother growth of your capital.

• Monitor developments in AI-driven bots to see how you might incorporate the latest advancements into your own trading.

• Be wary of black box products, with no evidence of profitability and expensive subscriptions. Trading edges in markets are hard to find and extremely valuable. Not only that, edges can decay if they are exploited by too many people. With this in mind, question why any service would be willing to open up a strategy to thousands of people.

• Shop around for the best fees and the fastest execution across various trading platforms. Trading fees can make the difference between a strategy being profitable or not, as can execution and liquidity.

AI-driven crypto bots represent an exciting development in the world of algorithmic trading, and research shows that machine learning could be used to successfully predict crypto markets.

Advancements in the way in which bots can monitor new information and learn adaptively based on previous performance, may enable crypto traders to deploy highly sophisticated algorithms that autonomously adjust their approach over time.

Interested in diving deeper into AI and crypto? Kraken makes it easy to buy and trade AI-based crypto tokens and more!

Sign up for your free account today.