Simulation-based designs for multiperiod control

In multiperiod control one sequentially chooses the value of an input variable to keep an output variable close to a target value. At any stage the input variable has to be such that the output is close to target while at the same time it provides as much information as possible about the model that relates input and output, to help keep the output close to target in the future. Certainty equivalence policies set the input variable at the value under which the output is estimated to be equal to target. Three alternative approaches that improve on certainty equivalence by searching for the best policy among families of policies that embed them are proposed, and their use is exemplified both on static as well as on time-varying dynamic models. Two approaches observe on credible regions for the root, while the third one is an adaptation of stochastic approximation sequences.