| Disaggregation is a topic which concerns both econometricians and statisticians. We discuss three existing techniques: the Wei and Stram method, Boot's First and Second Difference methods.;Wei and Stram (1990) proposed a model-based method, which used generalized least squares. The particular benefit of this technique is that the aggregate autocovariances are instrumental in calculating the disaggregate autocovariances.;Boot's First and Second Difference methods were proposed in 1967 to disaggregate annual data to quarterly data. Both are described in terms of constrained minimization expressions. These are non-model-based methods.;We then compare these three methods with a simulation study. We look at three criteria: the disaggregation mean square error (MSE), forecast MSE, and parameter estimation. We simulate various series, aggregate the series, and disaggregate the series, based on the existing methods. We find that the Wei and Stram method is the best by far for all three of these criteria.;If the aggregate model is ARIMA(p, d, r), where ;There has not been any published work on model-based disaggregation for vector series. We demonstrate aggregation and disaggregation properties for vector series. We propose a method for multivariate disaggregation: the Upper Bound method.;We simulated a bivariate AR(1) vector series. We compared the white noise, Boot, Wei and Stram, and the new method. The Upper Bound method does quite well in forecasting MSE, and does fairly well for estimating parameters. |
