What is the main limitation of conventional temporal mean squared error objectives in multi-step forecasting, and how does FreDF address it?

The main limitation is that conventional temporal mean squared error objectives carry an inherent bias caused by label autocorrelation when used for multi-step forecasting. FreDF addresses this by introducing a frequency-domain learning objective that decorrelates the labels, providing a statistically sound and empirically superior alternative.

The source explains that standard temporal mean squared error objectives are biased in multi-step forecasting because of label autocorrelation, meaning the errors are not independent across time steps. FreDF, the Frequency enhanced Direct Forecasting framework, replaces such objectives with a frequency-domain learning objective. This transformation decorrelates the labels, removing the bias introduced by their temporal dependence and yielding a more reliable training signal for multi-step forecasting.