Hi Yacoub, well written article, but I believe your model sufferes from a common pitfall where it simply learns that “tomorrow’s price will be similar to todays price”. This can be seen in your last graph, where the predicted line is very similar to the real line shifted one day over. To mitigate this, try expressing the price as a “change in price” instead, making it more difficult for the model to learn price(t+1) = price(t)