Artificial Intelligence (AI) is a vague concept. In our case, "AI Strategy" is an umbrella term we use to name a set of underlying tools (like machine learning, LLM-based feature engineering, model crossbreeding) which work together so traders could benefit from advances in the AI/ML domain.
On TrendSpider, traders can build AI strategies which generate predictive trading signals. AI strategies are focused on fixed R/R signals. Once you design an AI strategy, it will give you signals any time when over the course of the next few candles, the price is more likely to hit a take profit of X% than a stop loss of Y% (in opinion of a given strategy).
You can think of an AI strategy as of a black box (which you personally have designed) which consumes input data (like indicator values) and spits out signals. AI strategies can and should be backtested, which helps with the fact that it's a black box. An AI strategy backtested on the data it has not previously seen is a legitimate equivalent of backtesting a condition-based strategy (algorithmic strategy). If it's good, then you should forward test.
What are AI Strategies
AI strategies are a direct product of machine learning. Machine Learning is a process when computers figure our certain things without being explicitly programmed to do so. Recognizing human faces on photos is a product of machine learning. Self-driving cars are. Predicting structure of proteins is. Large language models are. There are many kinds of machine learning, but the overall idea of computers "figuring things out" is the foundation.
The history of using machine learning in trading is vast. There are numerous whitepapers published on the topic. There are books. There are a lot of smart people who just figure things out, without being loud about it. Machine learning can do a lot of things, but for the purpose of trading, it always boils down to whether the outcome of machine learning is exploitable (for trading) or not. In example, predicting whether the next candle will have its Close higher than the current candle Close is not as exploitable as it might look like at a first glance.
We have made a few opinionated decisions and intentionally framed the process by them. This way, we focus the power of machine learning (as a tool) on the goal of discovering exploitable signals. That allows you to leverage decades of progress in machine learning.
A trained AI Strategy on TrendSpider is a trained machine learning model. We support a few different types of models (like Random Forest, K-Nearest Neighbor and such) and we make it easy to create new models, experiment and build up on results of your experiments.
What AI Strategies are not
An AI strategy are is powerful tool. Tools have their strengths and weaknesses, and this tool is no different in that sense. It's still not a silver bullet. It's not like you plug all the data in the world into this black box and get awesome signals as an output. You should build your models, assess their quality and backtest them prior to actually using the signals.
Again — AI Strategies will not print you money directly. There is still work to be done by you.
How to create an AI strategy
Designing AI strategies has its challenges. However, these challenges are materially different from challenges of designing algorithmic strategies. These challenges are different to a degree when the goal of discovering a strategy becomes potentially reachable for people who could never design a working algorithmic strategy before. Depending on your knowledge in ML and trading, you could pick one of the two directions:
- If you know what you're doing then devote significant effort to feature engineering for your models and go with the simplest model types.
- If you don't really know what you're doing, then use medium-to-complex models and prepare to do a lot of grinding (train-crossbreed-backtest-discard) to compensate for the lack of knowledge.
Point 2, while not looking very attractive, is doable, and that's the key.
In general, creating an AI strategy consists of the following steps:
- Define the learning data set.
- Define what your model would have to predict.
- Define what type of a model do you want to use.
- Define what inputs your model will have to derive its predictions from.
- Assess the quality of a trained model
While 5 steps is still more complex than one click everyone craves, an example of the first 4 steps can be packed into a sentence like "SPY,60 (2022 to 2024), predict signals for TP 1.5% and SL 0.5% on a horizon of 30 candles, use KNN model, use RSI to derive predictions from".
Please refer to a dedicated article to see detailed information about every step of creating an AI strategy.