Using AI in Trading: How Computer Programs Decide When to Buy and Sell - page 2

[Dr Matthias Hammelsbeck]

In my opinion, the price data is far too random to make reliable predictions. No matter which method is used. Neither with the help of "AI" nor neural networks nor with other methods.

[Dominik Egert]

Dr Matthias Hammelsbeck #:
In my opinion, the price data is far too random to make reliable predictions. No matter which method is used. Neither with the help of "AI" nor neural networks nor with other methods.

 I disagree...

Price tells you everything you need to know. - My personal opinion, though.

But I would take an educated guess when stating successful (manual) traders would agree with me.

[Rajesh Kumar Nait]

Dominik Egert #:

 I disagree...

Price tells you everything you need to know. - My personal opinion, though.

But I would take an educated guess when stating successful (manual) traders would agree with me.

I agree with you, every science of trading hints do exists in price, but it may not work as you wish to make it work

for example majority of traders looks for 1 risk : 2 reward but who knows 5 risk : 1 reward may work with 99% accuracy. In other way managing risk with large SL hits and less winning but still winning in the end.

Its just traders lacking to look the other side of part which requires deep thinking, only hardcore researchers could be able to find it. All info is hidden within the price, nowhere else to look. Its necessary to have data accuracy in order to make it work. Some exchange may give broken data which makes market feel random

[Rafael Grecco]

I think we should see AI as just a really advanced programming tool. It's not gonna magically create a winning strategy just because you throw market data at it and ask it to come up with something (at least not yet... maybe in a few years?).

The first step is always having a solid strategy. Then, through code, we kinda have to show the AI what we want it to do, what’s right or wrong. How you do that depends on the type of learning and the kind of AI you’re using.

For example, with supervised learning, you give it labels or rules, basically telling it “this is a good result” and “this is a bad one.” Then the AI looks at the input data (like price, volume, indicators, etc.) and starts spotting patterns that lead to good or bad outcomes.

With reinforcement learning, it’s more about giving it rewards, like a positive reward when there's profit, and a negative one for losses. The AI starts trying out different entry and exit points on its own, and over time it learns what works. The downside is it becomes a bit of a black box, since the rules are created internally by the AI itself.

Sounds simple, but the truth is: it’s hard. There are thousands of possible model types, algorithms, networks, and decisions around what data to feed it, and that data has to be relevant, clean, and not too repetitive. Too much info can actually make it harder for the model to learn properly. The real challenge is finding something that works well, consistently, and predictably.

And to be honest, that takes a solid mix of three things: programming skills, AI architecture knowledge, and deep understanding of financial markets. Each one of those is already complex on its own, and when you put them together, it’s a lot to master. But if you can bring those pieces together, then yeah, AI can definitely become a powerful tool.

[Sergey Golubev]

Forum on trading, automated trading systems and testing trading strategies

The Ultimate AI EA Project

Sergey Golubev, 2025.10.07 15:20

Neural Networks in Trading: Models Using Wavelet Transform and Multi-Task Attention

Today, we introduce the Multitask-Stockformer framework, presented in the paper "Stockformer: A Price-Volume Factor Stock Selection Model Based on Wavelet Transform and Multi-Task Self-Attention Networks". Despite the similarity in name to the previously discussed StockFormer framework, these two models are unrelated - except for their shared objective: generating a profitable stock portfolio for trading in financial markets.

[Conor Mcnamara]

The reason why I haven't decided to trust AI in trading is because

1. Models that are trained on pattern recognition engines form their own random assumptions based on historical data (which are not a traders affirmations, but an AIs affirmations)

2. It's slow and behind the fact with fundamental analysis, just as the news release can sometimes be too slow.

3. It has proven to fail in trading already and not perform well live, while also blowing up accounts.

[Sergey Golubev]

Developing Trend Trading Strategies Using Machine Learning

Several types of trading strategies have proven their effectiveness in trading. One such strategy-the mean reversion strategy was covered in a previous article. In this article, I decided to share with the reader some ideas on how machine learning can be used to create trend-based or trend-following strategies.

This article will use a similar approach based on data clustering to identify market regimes. However, the actual trade labelers will differ significantly. Therefore, I recommend first reviewing the first article,and then proceeding to this one as a logical continuation. This will allow you to see the difference between the first and second types of strategies, as well as the differences in labeling training examples. Well then, let's get started!