Only 5% of traders achieve alpha, or significant profits, while the other 95% finance those gains by losing money. Success in trading is not a matter of luck.
Just like in a Formula One race, having a skilled driver behind the wheel is not enough to win; you also need an impeccable car, fine-tuned down to the smallest detail. A vehicle that has been meticulously prepared by a team of expert mechanics, who follow the engineers’ instructions to the letter. In the world of trading, that team includes solid strategies, reliable platforms, and a deep understanding of the markets, while the trader is the driver, who must be ready to reach the checkered flag.
Trading, like Formula One, is a collective effort where every piece plays a crucial role.
Let’s start with clean and well-prepared data to work algorithmically. Now, every area needs to be optimized, and for that, nothing beats the power of AI and Machine Learning. But why each area? The answer is simple. Just like in a Formula One car, every aspect is much more complex than it seems at first glance. Attempting a model that solves everything without compartmentalizing would only push us further from success. On the contrary, let’s break down each step.
Primary AI model with raw Profit Factor:
Let’s remember the well-known saying in Data Science: “Garbage in, garbage out” (GIGO). In a field like trading, which is closely related to statistical martingales, noise far outweighs the scarce signal or useful information for prediction. One must be extremely careful, or at any moment, you’ll find yourself modeling noise. And noise is unpredictable by definition. That’s why the first machine learning model focuses on extracting the signal, discarding the noise—in other words, obtaining the profit factor.
Metalabeled AI model with net Profit Factor:
However, raw profit factor requires fine-tuning strategies with optimized, not fixed, deadlines, and dynamic stop loss and take profit levels. Of course, all of this must be done in real-time for intraday data and for each individual asset. Averages can’t be used here, because what works in EUR/USD might not work in CHF/JPY, much less in Tesla or Bitcoin. Therefore, the second step is an AI model that optimizes these profit factors, turning them into useful strategies.
Where does the profit factor come from? Machine learning can be supervised or unsupervised, classic learning or deep learning, but ultimately, it seeks the features in each asset that make one situation successful and another not. In this way, primary AI models must be filtered again by a different AI. To explain, let’s assume we have a primary model that doesn’t use AI but instead relies on classic technical analysis, like a candlestick hammer pattern model. The second model, known as a metalabeler, would detect the differences between hammers, identifying the features that make some reliable and others not.
Reinforcement Learning Agent:
These strategies, however, are highly susceptible to the ever-evolving nature of the market. Market shifts present one of the greatest challenges in trading — unlike the constants of physical laws, what proves effective today may no longer be viable tomorrow. This highlights the importance of machine learning, which adapts in real-time, evolving alongside the market to stay ahead of these changes.
Even then, we’ve managed to build a competitive car, but it still needs to be driven to cross the finish line among that 5% of successful traders. For that, it’s important to have AI working at another level, AI capable of making decisions and learning from its trading. AI that accumulates “flight hours”. That’s where agents and reinforcement learning come into play — this is the final part, the driving.
At this point, we can talk about possibilities of success in trading, but never certainties. There are processes that are not entirely modelable and cannot be predicted with certainty. What can be guaranteed is to be as prepared as possible, maximizing our chances of success through analysis, strategy, and risk control. In trading, just like in meteorology, we must get as close as possible to the correct prediction. About 5% of traders succeed.
AIMetrics Signals:
In our case, after 3 years training models, we launched our signals in the real market. In the five weeks of live production of AImetrics signals, the hourly Forex market has generated positive profit factors in all of them, ranging from 1.39 to 2.10. After accounting for commissions, the range is between 1.33 and 1.95. There are no guarantees of consistently achieving these figures, but we will continue working on the models to make them increasingly effective and stable.
