Figure 3
Panel A shows a time-based holdout split. It consists of a series of 13 blocks labeled from 1 to 13. The first 10 blocks are colored blue and labeled as Train (W1-10), while the last 3 blocks are colored red and labeled as Test (W11-13). Panel B shows a walk-forward backtest. It consists of a grid of blocks arranged in a diagonal pattern. The blocks are labeled from 1 to 13. The blue blocks represent the training periods, and the red blocks represent the testing periods. The process starts with a training period of six weeks, followed by a testing period of one week. This process repeats, expanding the training period by one week each time and testing on the next single week.Evaluation protocols. Panel (a) shows the time-based holdout split: the model trains on Weeks 1–10 and is tested on Weeks 11–13. Panel (b) shows the walk-forward backtest: the model is retrained at each step on an expanding window and tested on the next single week, starting from a minimum training window of six weeks