What are you actually predicting?

Observe the past, predict the future

At time t, past prices and information are already known. The next price has not happened yet: history is an observed path, while the future has multiple possible outcomes.

From prices to returns

Suppose a price moves from 4000 to 4010. Its log return is:

r = ln(4010 / 4000) ≈ 0.00249688 ≈ 0.249688%

Predicting a price and predicting a return are different tasks. Define the target before interpreting a model’s output.

Prediction is not foreknowledge

A regression model may estimate the conditional mean of a future return. A classification model may estimate the probability that the return exceeds a threshold. Neither reveals the actual future return or implies a single possible future path.

Take a clear question into your research

Write down the prediction time, future horizon and target variable. Check whether each input was already observable at the prediction time. Clarify the question before discussing the model.

All price paths, distributions and animations in this lesson are synthetic teaching illustrations, not real market data, model outputs or investment advice.