| One of the most common data problems encountered by economic researchers is the lack of high-frequency data. The problem is especially disconcerting in dynamic time series models because estimation where the model's timing interval is smaller than the data-sampling interval leads to inconsistent parameter estimates.;I also apply the procedure to two macroeconomic issues. In the first, I forecast monthly data on GNP and hours worked and contrast the impulse responses from these data with data generated from a two-shock RBC model. The procedure allows me to investigate the role temporal aggregation plays in the discrepancies between output dynamics in quarterly RBC models and in the U.S. economy.;In the other, I explore whether a simple modified version of Taylor's overlapping contracts model is consistent with the Granger-causal orderings between income, money and prices in U.S. data. A quarterly version of the model does a remarkably good job of replicating the ordering in U.S. data; however, a semi-annual version of the model fails in several respects. In particular, although both versions of Taylor's model predict no feedback from output to prices, feedback appears in temporally aggregated, semi-annual U.S. data. Nevertheless, I show that one can disaggregate the data to a quarterly level and remove the feedback from income to prices induced by time aggregation.;My research addresses this problem by extending to vector autoregressions (VARs) an existing procedure by Telser (1967) that results in consistent estimates of the disaggregate (basic) autoregressive parameters. The procedure involves two parts: estimation and identification. First, the aggregate autoregressive parameters are estimated from the time-aggregated VAR by solving a system of Yule-Walker equations. Second, the basic parameters are identified by using information in the autocovariance matrices of the aggregate residuals. Once the basic parameters have been identified, observations on the basic series can be forecast via the Kalman filter. Furthermore, I perform a series of Monte Carlo experiments to investigate the small-sample properties of the estimates. |
