Consumption model calibration and related statistical problems

摘要

In this paper, we will discuss a problem pertinent to many situations in which a statistical model is developed on a sample of individuals (describing their trajectories), but then it is applied on a much larger population of interest. Typical examples occur in natural gas and other energy consumption contexts. Due to various deficiencies of the original sample-only based model, due to possible inconsistencies between sample and the population as a whole, and/or due to inherently different nature of the available sample and populational data (their measurement, error properties, etc.), calibration arises as a natural way to improve original model. We will introduce a formal and flexible, time-varying statistical model of state-space nature, from which such a calibration will come out as one of the products.