Improving seasonal adjustment by accounting for sample error correlation using state space models¹

Article type: Research Article

Authors: Mayer, Andreas

Affiliations: Australian Bureau of Statistics, ABS House, 45 Benjamin Way, Belconnen, ACT, 2617, Australia | Tel.: +61 2 6252 7140; E-mail: andreas.mayer@abs.gov.au

Note: [1] Views expressed in this paper are those of the author and do not necessarily represent those of the Australian Bureau of Statistics. Where quoted or used, they should be attributed clearly to the author.

Abstract: National statistical organisations seek to publish seasonally adjusted time series in which measurement errors have been minimised and systematic and calendar-related effects are removed. Time series derived from survey data contain sample error, and for rotating panel surveys such errors are correlated over time. Standard seasonal adjustment processes do not account for this, leaving sample error spread across the trend, seasonal and irregular components of the time series. This paper proposes an improvement: modelling sample error as a component of a structural time series model, and removing modelled estimates of sample error before applying existing seasonal adjustment processes. This results in improved seasonally adjusted and trend estimates which better reflect underlying movements and real world phenomena. We discuss several practical considerations for this method: revision properties, estimating sample error for aggregate series, prior corrections, and model maintenance. We demonstrate the potential of this approach using the example of employment and unemployment series from the Australia Labour Force Survey. Simulations show that compared with the current seasonal adjustment method, the proposed method produces estimates of month-to-month movement of seasonally adjusted and trend series which are consistently closer to the series that would arise if there was no sample error.

Keywords: Seasonal adjustment, sample error, state space modelling, rotating panel survey

DOI: 10.3233/SJI-180432

Journal: Statistical Journal of the IAOS, vol. 34, no. 3, pp. 409-423, 2018