Misspecified Predictor Bias Comparison

Background: In the setting of a misspecified linear system, where partial observability is present but the model assumes full observability (Markovian), the asymptotic prediction errors become irreducible biases that depend on the prediction horizon $H$.

Question / Future Work: Rigorously characterize the irreducible prediction bias for single-step, intermediate, and multi-step predictors in the misspecified setting (partial observability) and determine the resulting hierarchy of bias values, particularly how the bias scales with the prediction horizon $H$ due to information loss.