A note on spurious model selection
指出，即使模型只涉及单个时间段，破坏时间结构也会导致模型性能被高估。
Breaking time structure leads to overestimating model performance, even if the model concerns only a single time period