Abstract
One way to evaluate the importance of particular assumptions for different parts of the REMI model is to run the same simulation but with different suppressions in each instance. Another way is to postulate alternative pure versions or closures of the model, and then to test them with the same policy change. In addition to these two ways of seeing how sensitive the model results are to the model assumptions, we can also carry out ex post sample period forecasts to test which formulation of the model produces a model that would have best explained the 1980s. In this chapter we first consider how the predicted policy effect changes as the model suppressions are released. Next we look at the standard REMI model and four special cases of the model. Finally, we conclude with statistics comparing the five versions in post sample period forecasts.
This chapter draws heavily on Dan Rickman and George Treyz, “Alternative Labor Market Closures in a Regional Forecasting and Simulation Model” in Growth and Change, Winter 1993, and a memo by Gang Shao and George Treyz.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Notes on Chapter 9
This chapter draws heavily on Dan Rickman and George Treyz, “Alternative Labor Market Closures in a Regional Forecasting and Simulation Model” in Growth and Change, Winter 1993, and a memo by Gang Shao and George Treyz.
Note that government employment is not included in the REMI multipliers.
This section is based on “Alternative Labor Market Closures in a Regional Forecasting and Simulation Model” by Dan S. Rickman and George I. Treyz in Growth and Change, Winter 1993.
For examples of analytical regional general equilibrium models and comparisons to the economic base model, see Merrifield (1987; 1990) and Mutti (1981).
Economic base and input-output models are types of regional Keynesian models because of the implied assumption of perfectly elastic supply.
The demand and supply curves in the model are general equilibrium, because each reflects all endogenous responses in the system of equations. Thus, an exogenous change in export demand shifts the labor demand curve by an amount that includes the indirect and induced effects. Any subsequent change in the wage rates to obtain the new intersection of demand and supply also involves endogenous induced, indirect, and competitive effects. For the remainder of the article, the general equilibrium demand curve is simply referred to as the labor demand curve.
Though specification of input-output models varies in practice, the common elements among the models are the vertical demand curves and horizontal supply curves. This specification implicitly assumes sufficient unemployment to meet demand rather than perfect labor mobility and is often used by practitioners.
The population predictions of this version equal those of an input-output model that specifies migration responses that maintain a constant relative employment rate.
The last year of history of the model is 1988, with forecasts that run from 1989 to 1993.
The choice of region and industry is not critical for the comparison of the relative responses of employment and population across the versions. The same qualitative differences have been observed in other REMI models.
At existing tax rates, tax collections increase as real income increases (the tax base), financing the increase in government spending. If tax rates equaled zero, the induced consumption effects would be greater and there would not be an induced government spending effect. However, because of a balanced budget multiplier effect in the model, the existence of induced taxes and government spending increases net induced expenditures. The negative effects of income taxes occur through migration if tax rates are increased because potential migrants respond to real after-tax wage rates.
Though a Cobb-Douglas production function gives the substitution between labor and other primary factors, the substitution occurs at .0625 per year, assuming an average thirteen-year lifetime of equipment, which yields limited substitution in the short run (Treyz, Rickman and Shao, 1992, p. 231).
Again, since a horizontal supply curve makes the slope of the demand curve irrelevant, the GD/HSM version produces multipliers and population projections that also would be obtained from an input-output model with perfectly mobile labor and induced state and local government expenditures.
The labor demand curve in VD/GS is vertical with respect to the market shares and labor intensity suppression, but it is slightly upward sloping because of the induced effect of real wages on consumption, which produces slightly larger multipliers than in GD/HSM.
Since we are testing differences of MAPEs, the variances of the APEs are used to perform the F-tests, not the variances of the percent errors. The Levene and Welch tests are calculated using the BMDP analysis of variance procedure (BMDP Statistical Software, Inc., 1990).
The nonparametric Kruskal-Wallis test confirms all the parametric hypothesis tests of differences in means for first- and last-year forecasts of employment, population,
The nonparametric Kruskal-Wallis test confirms all the parametric hypothesis tests of differences in means for first- and last-year forecasts of employment, population, and wage rates. Thus, the hypothesis tests are robust with respect to the assumption of normality and the remaining tests performed are parametric.
The signs of the t-statistics for the MAPEs can be inferred from the relative values of the MAPEs in table 2. Also, the signs of the t-statistics for the variances all happen to correspond to the signs of the t-statistics for the MAPEs.
The similarity in t-statistics of the differences in MAPEs with those of the variances suggests that the differences in MAPEs are due to differences in the variances of the percent errors, not differences in the means of the percent differences. This is confirmed by the bias proportions of Theil’s U-statistics. The bias proportion is near zero in most cases and never exceeds fifteen percent. The results are available from the author upon request.
A two-way analysis of variance of APEs by model by major Census Bureau region is run to explore possible regional differences in relative model performance. Only two interactions of model with region have any statistical significance based on F-tests: 1988 population forecasts (p = .053) and 1988 wage rate forecasts (p =.078). The immobile labor version gives comparatively more accurate population forecasts for states in the Northeast, and the perfectly mobile labor version gives comparatively more accurate population forecasts for states in the West. During the 1980s the Northeast experienced dramatic increases in housing prices (not predicted by the REMI model) that inhibited in-migration and induced greater labor force participation rates to satisfy the employment growth. For 1988 wage rate forecasts, the market-clearing version is comparatively more accurate for the Northeast, reinforcing the argument above that wage rates rose increasing labor force participation instead of in-migration.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 1993 Springer Science+Business Media New York
About this chapter
Cite this chapter
Treyz, G.I. (1993). Forecasting and Simulation Sensitivity to Alternative Model Specifications. In: Regional Economic Modeling: A Systematic Approach to Economic Forecasting and Policy Analysis. Springer, Dordrecht. https://doi.org/10.1007/978-94-017-2874-4_9
Download citation
DOI: https://doi.org/10.1007/978-94-017-2874-4_9
Publisher Name: Springer, Dordrecht
Print ISBN: 978-94-017-2876-8
Online ISBN: 978-94-017-2874-4
eBook Packages: Springer Book Archive