No CrossRef data available.
Article contents
The Accuracy of Canadian Forecasts of Manufacturers' Capital and Repair Expenditures*
Published online by Cambridge University Press: 07 November 2014
Extract
Canadian anticipations data for manufacturers' capital and repair expenditures are now extensive enough to allow statistical estimates of their accuracy to be made. The data are based on mail questionnaire surveys conducted and analysed by the Dominion Bureau of Statistics and the Department of Trade and Commerce. Their year-end “cut-off” surveys include manufacturing establishments having gross sales in excess of $200,000.00 annually, and are used to estimate intentions for the year ahead, as well as to provide actual expenditures for the current year and the previous year. The results are published the following spring, and the forecasts are revised at mid-year on the basis of smaller “follow-up” surveys which include only the larger establishments.
The accuracy of these forecasts is measured in this study by comparing them with the estimates of actual expenditures which are published two years later. Objections to such comparisons on the grounds that the forecasts could be used as an instrument of control, designed to influence the subject of the forecast, can be safely ignored, since the published results are simply summations of the estimates of individual establishments, with a mechanical allowance for non-surveyed expenditures. Our measure of accuracy might also be criticized on the grounds that the estimates of actual expenditures may be as inaccurate as the forecasts themselves, so that the comparison of forecast to “actual” is no measure of the accuracy of the former. This is unlikely to be the case, however, since the accounting data of most firms provide accurate and readily available statements of actual capital outlays.
- Type
- Research Article
- Information
- Canadian Journal of Economics and Political Science/Revue canadienne de economiques et science politique , Volume 31 , Issue 2 , May 1965 , pp. 242 - 255
- Copyright
- Copyright © Canadian Political Science Association 1965
Footnotes
*
The author is indebted to A. D. Scott and P. H. Pearse for valuable comments and to C. A. Prentice, the Committee on Research, and the Computing Centre of the University of British Columbia for generous assistance provided.
References
1 See Firestone, O. J., “Investment Forecasting in Canada” in National Bureau of Economic Research, Short Term Forecasting, Studies in Income and Wealth, vol. 17 (Princeton, 1955)Google Scholar, for previous estimates.
2 See Department of Trade and Commerce, Private and Public Investment in Canada.
3 In 1960 for example, the estimates were published at the following times: spring, 1960, original forecast expenditures; mid-year, 1960, revised forecast expenditures; spring, 1962, final actual expenditures. Preliminary estimates of actual expenditures are also published, but they are not used here.
4 First differences are of interest in their own right, and ako provide a means of avoiding auto-correlation in regression residuals. See Cochrane, D. and Orcutt, G. N., “Application of Least Squares Regression to Relationships Containing Auto-correlated Error Terms,” Journal of the American Statistical Association, 03 1949.Google Scholar In this study, the use of first differences reduced the number of auto-correlated error terms from 7 to 3 of 204 cases tested.
5 The equations are generally based on 8 or 9 paired observations depending on whether first differences or original values are used. Because the data for 1952 are unavailable, the regressions on the revised forecasts are based on 7 and 8 paired observations respectively. More recent years are not included because of the change in industry classification occurring in 1961.
6 See n. 5.
7 With regressions run on totals by industry, type of expenditure, or over-all, misleading results may be obtained because of a cancellation of errors in the various forecasts.
8 In this study, these indexes were calculated both ways for a final check on the computations.
9 As one might expect, none of the 72 tests with repair expenditures shows a significant difference at the 0.05 level (Tables VII and VIII).
10 One should bear in mind that because of the skewed sampling distributions of the correlation coefficient, a given absolute difference is less likely to be significant for low values of the statistics sudi as are obtained with first differences, than for higher values.
11 See n. 5.
12 Repair expenditures are ignored since these forecasts are no longer revised.