⁽¹⁾ Board of Trade Journal. The figures used include Orders VII, VIII, and IX of the Standard Industrial Classification, i.e. Mechanical and Electrical Engineering, plus certain heavy vehicles (part of XI).

⁽²⁾ An earlier study published in this Review was confined to an examination of turning points only. (See E. A. Shirley, ‘Engineering orders : their use in forecasting’, National Institute Economic Review no. 21, August 1962.)

⁽³⁾ Since the work on this relationship was completed a further study has been published by P. K. Trivedi [9] using essentially the same method as that adopted here. A short note at the end of this section reviews the main differences and similarities between these two independent pieces of research. (⁴)i.e. net of cancellations.

⁽⁵⁾ The i subscript begins at 1 rather than zero because, as net new orders are calculated from current deliveries and orders on hand at the end of the period, any goods which were ordered and delivered in the current period do not appear in the order series. In practice, this seems to be a very small proportion of total deliveries.

⁽¹⁾ Data sources and definitions are given in Appendix II. All results are for the case in which a third order polynomial is assumed for both fixed and variable weight distributions (see Appendix I, page 51).

⁽²⁾ The results of these experiments are indicated in the text, but not presented in full. They may be obtained, together with a fuller version of this paper, on request to the National Institute of Economic and Social Research.

⁽³⁾ ‘t’ values are given only for the two points on each distribution which were explicitly estimated in the regression, as the restrictions serve to fix some of the remaining points completely and hence give them zero standard errors. The necessary equality of ‘t’ values for the variable weight coefficients is proved by Tinsley [8] —footnote page 1286.

⁽¹⁾ The variable weight predictions appear to be worse than those from the simple lag for late 1962—early 1963. How ever, as no adjustment was made to the delivery series for the effects of the severe winter, this anomaly can probably be discounted.

⁽¹⁾ The values reported for R² are relatively low compared with those for most regressions using economic time series data. However there is a large random component in the new order series caused presumably by the erratic timing of large individual orders. Several regressions were run using a 3 quarter moving average of the new order series as the dependent variable; R² moved up to the 0.85 range, but the relative performance of the explanatory variables remained as reported here. It should also be noticed that the order figures presumably contain orders from certain nationalised industries (e.g. electricity) whose output is not included in the index of manufacturing production.

⁽¹⁾ The particular series used includes United Kingdom production : this is open to objection.

⁽²⁾ The world price again includes the United Kingdom. This indicator is obviously a very crude measure of the relative price of machinery, but as it is only used to pick up the effect of the large change associated with devaluation it is probably satisfactory.

⁽¹⁾ This hypothesis is consistent with the failure to find any negative and significant price coefficients in equations estimated on pre-devaluation data.