Reflections on `Adaptive prediction' (Q1804257)

Adaptive prediction is discussed. A quantity to measure the degradation of the accuracy of the performance of a general nonlinear predictor due to the presence of uncertainty is introduced. Especially the use of time invariant predictors is discussed. The paper focuses on two issues: (i) Adaptive prediction is a good design methodology to cope with uncertainty. (ii) Adaptive prediction does not always lead to satisfactory performance, if compared with the performance achievable with other kinds of predictors.