Calendarization with interpolating splines and state space models

摘要

We consider the problem of transforming values from a flow time series observed over varying time intervals into values that cover calendar intervals such as day, week, month, quarter and year. We call this process calendarization. We propose simple methods based on interpolating the cumulated flows with natural spline interpolations. Alternatively, we provide state space models with missing observations to obtain smoothed values of the level of the cumulated flows. The state space models are the underlying statistical models behind Denton's benchmarking methods modified by Cholette. We therefore provide efficient alternative methods for benchmarking and temporal distribution. We show the theoretical properties of our methods, compare them and illustrate them with various examples.