I found the same thing in national accounts data revisions. Stock markets go jittery when GDP rises or falls by a fraction of a percent, but the size of the figures are rarely final or fully accurate, and can be revised up or down by a margin of 1%, 2% or sometimes 5% or more (for example, with inventory changes). The reason is that the aggregates are based on a very large number of data sources, and data for the current period is not available until later (the missing values are estimated with mathematical models). Usually it does not alter the main trend, but it does change the year-to-year fluctuations.
The data revisions usually occur, because it is decided to include some activity that was previously disregarded, or because the survey instrument is changed. The difference which the change makes is then projected back into the past. Usually the numbers get bigger, not smaller. However, that backward projection is often not based on real data, but on a mathematical model estimating what the measure would have been, "if" classifications had been different, or "if" a different survey instrument had been used. The most honest thing to do, would be end one time series, and start another one, but in that case, the two series would not be comparable, so they try to revise the old series, to make them comparable. Yet, if there is no way to base the backward revisions on actual observational data, and if it is only based on a mathematical extrapolation from the present, then the truth is, that the two data sets will never be genuinely comparable - the retrospective "top up" according to some algorithm is really spurious, and you may be better off using the unrevised old data, depending on your purpose. Of course, most times the revised old data series are only used to show a long-term trend leading up to the present, and that is really what the data revisions are aimed at. An alternative route is to use the observational data from the past to construct wholly new estimates consistent with the concepts used today. In fact people like Angus Maddison and Jan Luiten van Zanden used to do that a lot. They estimated GDP series going back for centuries using benchmarks and leading indicators. But it remains a rather tenuous activity and it is doubtful whether the series can indicate much more than whether the trend was up or down, or a rough idea of what national income would have been, given the number of qualitative changes occurring across long intervals of time. When I say "observational data", I mean data directly created out of current survey observations, in contrast to data that is extrapolated, indirectly derived or estimated from incomplete information. J.
_______________________________________________ pen-l mailing list [email protected] https://lists.csuchico.edu/mailman/listinfo/pen-l
