Improving baseline forecasts in a 500-industry dynamic CGE model of the USA

2017-02-28T03:21:04Z (GMT) by Mavromatis, Peter George
MONASH-style CGE models have been used to generate baseline forecasts illustrating how an economy is likely to evolve through time. One application of such forecasts is to examine how exogenous shocks (such as a proposed policy) cause changes away from the forecast path. Since the future state of the economy can be critical to the impact of such shocks, an accurate forecast is likely to enhance the reliability of policy analysis. This thesis examines methods for improving baseline industry/commodity projections by conducting forecasting-performance validation of a 500-industry MONASH-style recursive-dynamic CGE model of the U.S., known as USAGE. USAGE generates baseline forecasts by incorporating expert projections for certain macro and energy variables and extrapolating historical trends in tastes and technology. A previously-produced USAGE forecast for 1998 to 2005 of commodity outputs using information available up to 1998 is explored. When compared to actual outcomes, USAGE comfortably outperformed a non-model extrapolation-based trend forecast. However, there were numerous large errors prompting the question of whether USAGE should have performed even better. This thesis seeks to answer this question and closely examines the twenty largest USAGE errors. The thesis extends forecasting-performance model-validation methodology. It shows how CGE forecasting techniques can be improved by: obtaining expert industry-specific projections; carefully assessing on a case-by-case basis whether it is reasonable to project forward changes in preferences and technologies; and, changing the model’s implementation to better reflect historically observed policy behaviour. It is found that in instances where important trends either dissipate or reverse, large forecast errors can arise. For some commodities, had all publicly available information in 1998 been appropriately utilised, some trends should not have been expected to continue and hence a better forecast could have been generated. Furthermore, the nature of some forecast errors suggests that projecting forward large values for preference variables relating to import penetration might best be avoided. In some instances, changes to regulatory regimes that were put in place by 1998 suggested that affected industries had highly constrained growth prospects. These regulatory changes should be taken into account in forecasting exercises. For commodities in the trade-exposed textile, clothing and footwear (TCF) industries, moderately better results could have been produced by implementing import-price forecasts in a way that is more aligned with outcomes that are consistent with the historical operation of U.S. trade policy. Moreover, the key drivers behind USAGE errors in the TCF industries were usually the significant underestimation of the impact of domestic-import preference twists, as well as the overestimation of factor-input cost savings. Upon implementation of improved methodology, vast improvements in forecast accuracy for some industries were obtained. However, the average forecast error across industries did not greatly improve due to the sheer volume of commodities. While it is disappointing that the average error is not very reducible, it is also reassuring because it implies that the default implementation of the model is quite powerful. A large reduction in the forecast error—and hence improvement in model performance—would probably necessitate the input of numerous industry specialists.