Nowcasting with Backcalculated Short Time Series Simulations & Empirical Evidence

This paper is concerned with the backcalculation of short explanatory time series when the dependent variable is longer. In particular, we consider two competing linear regression models: (i) the main model which is estimated in the overlapping period when all time series are available, and (ii) the suggested model which produces the estimates in two steps; first, we create backcalculated values of the explanatory time series using some auxiliary variables (which are observed at the same -longer- time history as the target variable) in the overlapping time period and, using these coefficient estimates, we extend the short explanatory time series and estimate a model which regresses the dependent variable on this new set of variables. This research provides both simulations and empirical evidence in favour of the suggested method.