17.18. The Australian Bureau of Statistics (ABS) maintains long time series for national and international accounts. A large proportion of those series are maintained in original, seasonally adjusted and trend variations. The introduction of changes to compilation methods and treatments can result in shifts to the levels of component and total series. If the shift in level is sufficient to distort the seasonally adjusted time series, ABS revises the historical series to make the time series as continuous as possible.
17.19. In some cases, there is not sufficient detail available to adjust the historical series directly. In such cases, ABS estimates the shift in level of the series by comparing estimates at one point in time for both the current and new basis (although comparison for additional periods is desirable).
17.20. Ideally, any change in the level of a series would be measured over a sufficient time period to enable seasonal patterns to be observed. This is possible for some modelled estimates, but for estimates based on surveys, the cost of producing two estimates for one or more time periods is expensive, both in terms of processing costs and provider burden. In such cases, alternative methods may be needed.
17.21. To ensure the consistent treatment of time series, the ABS has established a standard approach to measuring shifts in the level of series. The size of the level shift induced by a methodological or measurement change is assessed using regression analysis techniques on ratios between the current published estimates and actual or simulated estimates produced by the revised methodology. In cases in which the level shift is found to be significant in the seasonally adjusted series, the historical series is back-casted to make the time series as continuous as possible while maintaining, as far as possible, the integrity of the period-to-period seasonally adjusted movements, taking into account real-world changes. For a small number of lower level series, it may not be possible to create a valid time series and such series are marked “not available” for periods prior to the start of data collection.
17.22. Where it is not possible or necessary to maintain a long time series, an approach of “bridging” the current published estimates and the estimates produced by the revised methodology is used. Bridging means that estimates on both the current and new basis are produced for one point in time and both sets of estimates are released along with analysis to help users understand the differences between the series. The technique is particularly relevant for series where modelling beyond a certain time may not be appropriate.
Back to C. Forecasting, back-casting and revising time series