20 Seasonally adjusted estimates are derived by estimating and removing systematic calendar related effects from the original series. In the Retail trade series, these calendar related effects are known as:
- seasonal e.g. annual patterns in sales, such as increased spending in December as a result of Christmas
- trading day influences arising from weekly patterns in sales and the varying length of each month and the varying number of Sundays, Mondays, Tuesdays, etc. in each month
- an Easter proximity effect, which is caused when Easter, a moveable holiday, falls late in March or early in April
- a Father's Day effect, which is caused when the first Sunday in September falls in the first few days of the month and Father's Day shopping occurs in August.
21 Each of these influences is estimated by separate factors which, when combined, are referred to as the combined adjustment factors. The combined adjustment factors are based on observed patterns in the historical data. It is possible that with the introduction of ANZSIC 2006 from July 2009 the historical patterns may not be as relevant to some series. For example Watch and jewellery retailing moved from the Other retailing n.e.c industry subgroup to the Footwear and other personal accessory retailing industry subgroup under ANZSIC 2006. The seasonal patterns for other businesses in the Footwear and other personal accessory retailing industry subgroup appear to differ from watch and jewellery retailers. The combined adjustment factors will evolve over time to reflect any new seasonal or trading day patterns, although in this example, an estimate for this impact (seasonal break) has been implemented in the combined adjustment factors.
22 The following Retail trade series are directly seasonally adjusted:
- Australian turnover
- each state total
- each Australian industry subgroup total
- each state by industry subgroup.
23 A "two-dimensional reconciliation" methodology is used on the seasonally adjusted time series to force additivity - that is, to force the sum of fine-level (state by industry subgroup) estimates to equal the Australian, state and industry subgroup totals. The industry group totals are derived from the lower level estimates.
24 Quarterly seasonally adjusted series used in the compilation of the chain volume measures are the sum of their applicable monthly series.
25 Autoregressive integrated moving average (ARIMA) modelling can improve the revision properties of the seasonally adjusted and trend estimates. ARIMA modelling relies on the characteristics of the series being analysed to project future period data. The projected values are temporary, intermediate values, that are only used internally to improve the estimation of the seasonal factors. The projected data do not affect the original estimates and are discarded at the end of the seasonal adjustment process. The retail collection uses an individual ARIMA model for each of the industry totals and state totals. The ARIMA model is assessed as part of the annual reanalysis.
26 In the seasonal adjustment process, both the seasonal and trading day factors evolve over time to reflect changes in spending and trading patterns. Examples of this evolution include the slow move in spending from December to January; and, increased trading activity on weekends and public holidays. The Retail series uses a concurrent seasonal adjustment methodology to derive the combined adjustment factors. This means that data from the current month are used in estimating seasonal and trading day factors for the current and previous months. For more information see Information paper: Introduction of Concurrent Seasonal Adjustment into the Retail Trade Series (cat. no. 8514.0).
27 The seasonal and trading day factors are reviewed annually at a more detailed level than possible in the monthly processing cycle. The annual reanalysis can result in relatively higher revisions to the seasonally adjusted series than during normal monthly processing.
28 The seasonally adjusted estimates still reflect the sampling and non-sampling errors to which the original estimates are subject. This is why it is recommended that trend series be used with the seasonally adjusted series to analyse underlying month-to-month movements.
29 The trend estimates are derived by applying a 13-term Henderson moving average to the seasonally adjusted monthly series and a 7-term Henderson moving average to the seasonally adjusted quarterly series. The Henderson moving average is symmetric, but as the end of a time series is approached, asymmetric forms of the moving average have to be applied. The asymmetric moving averages have been tailored to suit the particular characteristics of individual series and enable trend estimates for recent periods to be produced. An end-weight parameter 2.0 of the asymmetric moving average is used to produce trend estimates for the Australia, State and Australian industry group totals. For the other series a standard end-weight parameter 3.5 of the asymmetric moving average is used. Estimates of the trend will be improved at the current end of the time series as additional observations become available. This improvement is due to the application of different asymmetric moving averages for the most recent six months for monthly series and three quarters for quarterly series. As a result of the improvement, most revisions to the trend estimates will be observed in the most recent six months or three quarters.
30 Trend estimates are used to analyse the underlying behaviour of the series over time. Trend estimates from March 2020 to June 2022 are not available due to the degree of disruption and volatility caused by COVID-19. Trend estimates throughout the pandemic period are likely to be unhelpful and potentially misleading for users in interpreting underlying trend in retail activity. For further information on seasonally adjusted and trend estimates, see: