#forecasting

Question 1: Single choice

A model predicts a cryptocurrency's price one day ahead from the current price, the price change over the last 24 hours and the last hour, the volume traded in 24 hours, and the number of recent tweets. Over the past year the price moved by more than $100, yet the model's average error is under $1. What is the main problem with buying whenever it predicts a rise?

Tomorrow's price is close to today's, so a small error proves little1
Target leakage, because the current price is not known when the prediction is made2
Train-test contamination, because the tweet counts mix training and test periods3
An error that small cannot be real, so the price data must be corrupted4

Timed test