To speak of artificial intelligence applied to the financial markets, it is first necessary to speak of what makes it possible at all: the data. We are living through a moment in which data has ceased to be a mere record of the past and has become the raw material of our decisions, and it is precisely that transformation —quiet, yet profound— which it is worth retracing before we move forward. Within the reflection I have been developing around technology placed at the service of people, it is time to dwell on a piece of work I have had the privilege of co-signing with Laura Sanz-Martín, within the BISITE Research Group of the University of Salamanca, published in 2024 in the journal Mathematics under the title ‘Artificial Intelligence in the New Era of Decision-Making: A Case Study of the Euro Stoxx 50’.

The underlying question is as old as the markets themselves: can we anticipate how prices will behave? What is genuinely new, however, does not lie in the question, but in the tool with which we now attempt to answer it. Artificial intelligence does not promise a perfect crystal ball —that will never exist— yet it does offer an instrument capable of detecting patterns that entirely escape the human eye, and it is in that capacity that much of its disruptive character resides.

Why does this study matter now?

It is important to note that the article does not settle for a general approach; rather, it offers a rigorous, comparative evaluation of three neural network architectures applied to the EURO STOXX 50 index —the barometer of the fifty largest companies in the eurozone—, with the aim of discerning which of them best suits each type of question. Specifically, the architectures under study are the following:

• Artificial neural networks (ANNs), which provide a solid and efficient benchmark when the available resources are limited.
• Convolutional neural networks (CNNs), particularly adept at capturing local patterns and microtrends.
• Long short-term memory networks (LSTMs), well suited to identifying long-range temporal dependencies and, therefore, to understanding volatility.

Nonetheless, what is truly significant is that we did not stop at raw prices. On the basis of the above, we integrated a set of technical indicators widely used by analysts, in order to endow the models with a real sensitivity to trends and to shifts in market momentum:

• Simple and exponential moving averages (SMA and EMA).
• Moving average convergence/divergence (MACD).
• The stochastic oscillator.
• The relative strength index (RSI).
• Accumulation/distribution (A/D).

Results that speak for themselves

Without a doubt, the most notable finding is that the combination of the convolutional network with the relative strength index (CNN with RSI) delivered the best performance, with an RMSE of 0.0263, an MAE of 0.0186 and an R² of 0.9825. Put in plain terms: the model explains more than 98% of the price variation in the test data, a truly remarkable accuracy in a field as slippery as the stock market.

In fact, each architecture revealed a character of its own: CNNs excel at capturing local patterns and microtrends; LSTMs shine at identifying long-range temporal dependencies, so useful for understanding volatility; and ANNs offer a solid, efficient point of reference when resources are limited. There is, therefore, no single universal winner: the challenge lies in choosing the right tool for each question.

Essential reading for decision-making

I champion the reading of this work for a reason that transcends finance: it teaches us to decide better with data. In an age in which executives, analysts and public officials face ever more uncertain environments, understanding what artificial intelligence can and cannot do has become, without a doubt, an essential competence.

In this respect, it is necessary to stress that the study is honest about its limits —it does not incorporate social media sentiment, economic news or geopolitical events— and it is precisely for that reason that it proves valuable: it points the way towards hybrid models that are more transparent, fair and accountable. Artificial intelligence in finance must not be a black box, but an explainable ally for those who decide, for the technique must always remain at the service of human judgement and never the other way around.

The corner of little, big, curious minds

Let us imagine for a moment the school playground at break time. Every day one notices the same things: who runs towards the swings, when the slide fills up, what everyone plays when it rains. At first it is no more than what one sees on a single day; but if one observes many break times in a row, things that repeat begin to stand out. That, no more and no less, is what discovering a pattern means.

Artificial intelligence is like a friend endowed with an enormous memory and an almost infinite patience. It has watched so many break times that it learns to guess what will happen tomorrow, not because it is magical, but because it has lived the same scene thousands of times. In our study, that friend does not watch the playground: it watches the prices of Europe’s largest companies and tries to anticipate whether they will rise or fall.

And the important thing is this: our clever friend gets it wrong at times, and that is why it never decides alone. It helps us to think, but the final word always belongs to people. Artificial intelligence is a very powerful magnifying glass for seeing the world better; the eyes that decide what to do are still ours.