In this paper, we describe our investigation into whether using the trend rather than the seasonally adjusted series could improve the stability of the figures in Statistics New Zealand publications.
Summary of our methodology and results
- We analysed 15 Statistics NZ time series from the business and social areas, using absolute total revisions and Levene’s test for equality of variances.
- We calculated and analysed the absolute total revisions for the seasonally adjusted and trend estimators using various statistical tests, including Levene’s test.
- We used Levene’s test to identify if there is a statistically significant difference between the variances of the seasonally adjusted and trend series in terms of total revisions.
- Levene’s test for equality of variances plus other tests confirmed that the difference between the variances of the seasonally adjusted and trend series was statistically significant for all 15 time series.
- The absolute total revision graphs showed that the trend series had larger revisions than the seasonally adjusted series in all 15 series we analysed.
- The plots of the seasonally adjusted revisions and trend revisions showed the trend series to be less stable than the seasonally adjusted series. This was more pronounced towards the recent end of the series due to the end-point problem.
- The seasonally adjusted series showed consistent significantly smaller variance than the trend cycle series.
- We found the seasonally adjusted series consistently outperformed the trend series in the relative size of revisions.
- For greater stability in revisions in Statistics NZ’s official estimates, we recommend leading with the seasonally adjusted series.
To read this publication, download or print the PDF from 'Available files' above. If you have problems viewing the files, see opening files and PDFs.
Buchanan, R, and MacCormick, C (2014). Comparing revisions in time series data: A report on seasonally adjusted versus trend series. Available from www.stats.govt.nz.
Published 7 July 2014