Temporal Disaggregation by State Space Methods: Dynamic Regression Methods Revisited

The paper documents and illustrates state space methods that implement time series disaggregation by regression methods, with dynamics that depend on a single autoregressive parameter. The most popular techniques for the distribution of economic flow variables, such as Chow-Lin, Fernandez and Litterman, are encompassed by this unifying framework. The state space methodology offers the generality that is required to address a variety of inferential issues, such as the role of initial conditions, which are relevant for the properties of the maximum likelihood estimates and for the the derivation of encompassing representations that nest exactly the traditional disaggregation models, and the definition of a suitable set of real time diagnostics on the quality of the disaggregation and revision histories that support model selection. The exact treatment of temporal disaggregation by dynamic regression models, when the latter are formulated in the logarithms, rather than the levels, of an economic variable, is also provided. The properties of the profile and marginal likelihood are investigated and the problems with estimating the Litterman model are illustrated. In the light of the nonstationary nature of the economic time series usually entertained in practice, the suggested strategy is to fit an autoregressive distribute lag model, which, under a reparameterisation and suitable initial conditions, nests both the Chow-Lin and the Fernandez model, thereby incorporating our uncertainty about the presence of cointegration between the aggregated series and the indicators.