Files in this item
|(no description provided)|
|Title:||A Strategy for Monetary Policy Using a Multiple Time-Series Model|
|Author(s):||Fujihara, Roger Arnold|
|Department / Program:||Economics|
|Degree Granting Institution:||University of Illinois at Urbana-Champaign|
|Abstract:||The purpose of this dissertration is to explore ways of pooling various information sets and emphasizing the use of time series models in the process. While the combination of forecasts is not new, the combination of time series models and the more traditional structural econometric models has not been thoroughly evaluated.
The procedures used to combine forecasts are ad hoc, but serve a practical purpose by assisting the monetary authority policymakers who must evaluate information from various sources. However, the information generated by the weighting of forecasts may serve to produce a model which optimally exploits the various data sources. Several methods of combining forecasts are evaluated, including the use of the arithmetic mean.
Based on a root mean square error measure (RMSE) of performance, methods which employ the assumption that forecast errors are independent yield good results. In addition, when allowing the weights to vary with time, moving average estimation of the weights produce relatively stable results in terms of the RMSE, under the assumed independence of, as opposed to correlated, forecast errors. While this stability property of the independence assumption is desirable, there is no way to determine the rate at which the weights adapt over time. Attempts to try to model this evolution process of the weights were not promising, suggesting that additional attention has to be given to the prior use to characterize the adaptation of weights over time.
Time series models are gaining more attention by providing an additional means of uncovering how variables in the economy interact. Its usefulness in forecasting has been demonstrated, but mainly limited to univariate applications until recently. The development of multivariate forecasting models allow policymakers an additional source of information which can be exploited. Combining forecasts from a multivariate time series model can contribute to improving the forecast from a structural econometric model. Our results of pooling time series and structural econometric models are limited to short time horizons, one and two quarter forecasts. In these instances no one model dominated the other over forecast horizon or across variables.
Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1988.
|Date Available in IDEALS:||2014-12-16|