Reference Years in the Database
Several reference points will be needed in constructing historical time series, for instance, in computing various index numbers. For this purpose, it is suggested that the year 1960 be adopted as a post-WW?U reference year, as the project needs a reference point at the earliest possible date after WW?U to serve as a connecting point of pre-and post-WW?U series. The consensus judgement is that "normality" had been re-established by then.
If 1960 is too early for some countries (such as Indonesia, which was still struggling with political disorder then), however, a later year (say, 1970) may have to be used as an exceptional, unavoidable measure.
It is suggested, on the other hand, that a short term of 1934t36 be adopted as another reference point in the pre-WW?U period, since the world economy found itself in relative stability during these three years. Another, earlier reference year may be 1913, which was a good year in business shortly before WW?T.
Undoubtedly, however, the appropriate reference year may vary from one variable to another; e.g. peaks and troughs do not always coincide between them. The choice of the reference year is little more than a matter of convenience; one must have base year(s) in order to synchronise all the indices included in the project.
Missing Values
Some old documents are often marred with missing observations, due sometimes to measures to protect privacy, to the negligence of the surveyor, to the loss of original survey, or simply to deficient information. As a matter of basic principle, the COE Project endeavours to make all the time series continuous without missing data. In case observations for some years are lacking, the utmost effort should be made to remedy the deficiency, ideally by direct estimation of the corresponding value, but alternatively with the help of some working assumptions (or hypotheses), e.g. the variable in question is complementary to another commodity whose data are available, and demand for the former moves more or less parallel to that for the other. If the worst comes to the worst, the missing spot may have to be filled by simple interpolation. Whenever some such measures
