Why Predicting Markets Is Different

The advert that follows you around

You have seen it. A calm voice over a rising green line, a laptop on a kitchen table, someone explaining that the machines have worked out what the market is going to do next and that you can be on the right side of it for a small monthly sum. Perhaps it was a video. Perhaps it was a friend of a friend in a group chat who seemed to be doing rather well.

If you have wondered whether there is something in it, that is a perfectly reasonable thing to wonder. Prices are numbers, computers are good at numbers, and AI has genuinely done startling things in the last few years. The question deserves a straight answer rather than a lecture, so here it is: the thing AI is good at and the thing markets require are not the same thing, and the gap between them is not a matter of waiting for better technology.

Before anything else, the honest framing for this whole course. This is educational material about how these claims work. It is not financial advice, it does not recommend or discourage any investment, and nothing in it is a suggestion to buy, sell or hold anything at all. If you are making real decisions with real money, speak to a qualified adviser who is regulated in your own country.

What AI is genuinely good at

Modern AI is extraordinary at a particular shape of problem: one where there is a large pile of examples, where the rules producing those examples stay still, and where being right is defined the same way tomorrow as it was yesterday.

Reading a chest scan works like that. Human anatomy does not reorganise itself because a model got good at spotting something. Translating between languages works like that. Recognising a face, transcribing speech, predicting the next word in an ordinary English sentence: all of these are pattern recognition over a world that holds reasonably still while you learn it.

That stillness is doing an enormous amount of quiet work. Take it away and the whole approach changes character.

Markets do not hold still

Two things make prices different, and they compound.

The first is that prices move on information that does not exist yet. A price today already reflects what is currently known. What moves it next is the thing nobody has yet: an announcement, a resignation, a war, a weather event, a decision made in a room this afternoon. No amount of history contains tomorrow's news, and a system trained on the past is not so much pessimistic about the future as blind to it.

The second is stranger and more important. In a market, finding a pattern changes the pattern.

Suppose a genuinely reliable rule existed, something that told you a price was about to rise. The moment anyone acts on it, they buy, which pushes the price up before the rise they were expecting. As more people act, the edge gets smaller, until acting on it is no longer worth the cost of doing so. The pattern does not merely fade. It is consumed by the act of using it. Nothing in medical imaging behaves like this. A tumour does not move because you found it.

The plain logic question

Which brings us to the question worth keeping in your pocket for the rest of your life.

If a system reliably turned money into more money, why would anyone sell it to you for a monthly fee?

Sit with that for a moment. Selling access is a strange business decision for someone holding an actual money machine. Running it themselves would pay better, take less effort, and involve no customers, no refunds, no advertising and no regulator. Every subscription sold makes the edge smaller for the seller, if the edge is real. The business model quietly tells you what the seller actually believes.

There is a version of the pitch that pretends to answer this: they only want a few members, they are giving back, they want to democratise something. Notice that the answer is always about the seller's character, never about the mechanism. That is the tell.

Why the charts in the advert look so good

Almost every impressive result you are shown is a backtest: the strategy applied to price history that had already happened when the strategy was designed.

The problem is not only dishonesty, though there is plenty of that about. The deeper problem is that with enough attempts you will always find rules that fit the past beautifully. Try a thousand variations on old data and some will look spectacular by coincidence alone. That is overfitting, and it is the easiest mistake in the field to make, including for people acting entirely in good faith. A backtest is a rehearsal in an empty theatre where you already know every line. It tells you very little about opening night.

Survivorship bias finishes the job. The strategies that lost money are not screenshotted, not posted and not discussed. You see the ones that happened to work, which makes luck look exactly like skill.

What this course is for

Not to teach you to trade with AI. That is the opposite of the point. This course exists so that the next time an advert promises you something, you can tell what kind of claim it is, why it usually does not survive contact with reality, and where AI genuinely does earn its keep in ordinary financial life. That last part is real, and it arrives in lesson four. It is simply far duller than the adverts, which is rather the theme of everything that follows.