No, the model is not REALLY predicting the stock price. This is what it's doing instead.

Ever seen those fancy graphs claiming a Machine Learning model can predict stock prices? Next time you do, take a closer look.

🔎 You'll sometimes find that tomorrow's predictions look pretty much like the price today (except for some random error). But why?

👉 Because if a model has no idea whether the stock's price is going up or down, today's price is the best prediction it can make for the price tomorrow.

And guess what? Metrics such as the Mean Squared Error (MSE) will look pretty nice, especially when tomorrow's price is close to today's.😲
(this will even work for a random walk, so you can get good metrics even without capturing any signal)

Let's be honest here: do you really need an ML model whose prediction for tomorrow's price is simply the price today? 🤷‍♀️

So next time you see a ML project claiming it can predict the stock market, have a healthy dose os skepticism 🍹 and check that the curve of predicted values isn't simply the curve of actual prices slightly shifted to the right... 👯