5609.0 - Housing Finance for Owner Occupation, Australia, Sep 2003  
Previous ISSUE Released at 11:30 AM (CANBERRA TIME) 10/11/2003   
   Page tools: Print Print Page Print all pages in this productPrint All RSS Feed RSS Bookmark and Share Search this Product

Feature Article - Introduction of Concurrent Seasonal Adjustment


The ABS will introduce concurrent seasonal adjustment into the monthly Housing Finance for Owner Occupation, Australia (cat. no. 5609.0) and Lending Finance, Australia (cat. no. 5671.0) publications from October 2003. The purpose of this note is to demonstrate the impacts of this change in methodology on the Housing Finance series which is published in both publications.


The seasonal adjustment process uses estimated factors to remove the effects of normal seasonal variation, moving holiday and trading day effects from the original estimates. These factors are estimated by analysing historical patterns to determine the significance and direction of these systematic influences. There are two approaches used to do this, known as forward factor and concurrent adjustment.

Forward factor seasonal adjustment relies on an annual analysis of the latest available data to determine factors (known as forward factors) that will be applied in the forthcoming 12 months. The advantage of this approach is that estimates of seasonal factors will only be revised on a yearly basis. However, the seasonal factors produced by this method can become out of date with the addition of further original estimates. Under forward factor adjustment, revisions to the seasonally adjusted estimates are hidden essentially as they occur only at the time of annual re-analysis. The method is not responsive to current significant, or unexpected, movement of the seasonal component.

Concurrent seasonal adjustment uses the data available at the current reference period (month for Housing Finance) to estimate seasonal factors for the current and previous months. Under this method, the estimates of the seasonal factors are fine tuned as new original estimates become available each period. The seasonally adjusted estimates are subject to revisions at each reference month as the estimates of seasonal factors are improved. This method eliminates the need for projecting forward factors, and can result in significant improvements in the accuracy and consistency of the seasonally adjusted series.


In general, concurrent seasonal adjustment produces more accurate initial seasonally adjusted and trend estimates with less expected future revisions. In addition, the seasonally adjusted and trend estimates converge to the final estimates more quickly. The analysis below supports these assertions.

The expected level of revision for a time series can be examined using multiple simulations to estimate an average revision graph. Comparison of the average revision paths under concurrent adjustment and forward factors are given for the number Housing Finance by banks for established housing in Figure 1, the number of Housing Finance by banks for construction in Figure 2 and the number of Housing Finance by all lenders for refinancing in Figure 3. The lines on the graph represent the average of the absolute percentage revision to the level estimates at specified lags from both the concurrent method (lower line) and forward factor method (higher line). The average of the absolute percentage revision against lagged estimates is a measure of the revision size and convergence speed of the seasonally adjusted estimates. It is calculated using simulations of all estimates at a specific lag over the period of study (in the case of Housing Finance, this period is from October 1980 to August 2000).

For example, the average at lag zero (0) of the absolute percentage revisions uses the initial estimates over the whole study period. Lagged estimates are obtained by using available data at successive time points. Similarly, the lag zero July 2000 estimate is the initial estimate using data up to July 2000 and the lag 1 estimate is the second estimate of July 2000 using data up to August 2000.

Figures 1 to 3 show that on average, the concurrent seasonally adjusted estimates exhibit less revision on each lagged estimate than the forward factor estimates and converge to the final estimate quicker. The average revision path generally decreases at each lag, indicating that as additional original estimates become available for use in estimating the seasonal factor, the average amount of revision required to achieve the final estimate decreases.

Figure 1: Average Revision on level of seasonally adjusted estimates Housing Finance by Banks for Established Dwellings - Number

Graph - Average Revision on level of seasonally adjusted estimates Housing Finance by Banks for Established Dwellings - Number

Figure 2: Average Revision on level of seasonally adjusted estimates Housing Finance by Banks for Construction - Number

Graph - Average Revision on level of seasonally adjusted estimates Housing Finance by Banks for Construction - Number

Figure 3: Average Revision on level of seasonally adjusted estimates Housing Finance for All Lenders for Refinancing - Number
Graph - Average Revision on level of seasonally adjusted estimates Housing Finance for All Lenders for Refinancing - Number

Simulations were conducted for selected components of Housing Finance, comparing the initial seasonally adjusted estimates to their 'final' seasonally adjusted estimate. The results of these simulations are shown in Table 1 below, and summarise the improvements highlighted in the charts above. The average percentage revision between the initial and final seasonally adjusted estimates, under both concurrent and forward factor methods are given in columns 2 and 3 respectively. Column 4 shows that the concurrent method produces seasonally adjusted estimates that are between 17.7% to 18.1% closer to the final estimate.

Table 1. Average revision of the initial seasonally adjusted estimates compared for concurrent adjustment and forward factor method

Housing Finance series
Average revision (%)
Percentage improvement
Factor (ff)
Banks for Established - Number
Banks for Construction - Number
Refinancing, All Lenders - Number


While the simulation analysis above provides evidence of the benefits of introducing concurrent adjustment for the Housing Finance series, there are a number of issues that will impact on users of this series.

The estimates of combined seasonal adjustment factors are currently only amended once a year, in the October reference month following the annual seasonal reanalysis. Concurrent adjustment could result in the combined seasonal adjustment factors and the seasonally adjusted estimates being revised each month rather than only once a year.

It will still be necessary to undertake an annual seasonal reanalysis of the Housing Finance series. The annual reanalysis for the concurrent method is used to review details of seasonally adjusted methods used, examine time series for outliers or unusual data, changing seasonality and structural breaks. It is possible that some issues will only gradually emerge over a number of months and may not be as obvious when the focus is on concurrent adjustment. The annual seasonal reanalysis should therefore not result in significant revisions to combined seasonal adjustment factors, as combined seasonal adjustment factors and therefore seasonally adjusted estimates will be revised each month as a result of concurrent adjustment.

Provision of Forward Factors

The ABS currently provides forward factors to ABS clients wishing to undertake their own seasonal adjustment and trend process. Since the concurrent adjustment process will revise seasonal factors at each period, forward factors will no longer be required in the seasonal adjustment process undertaken by the ABS. The concurrent adjustment process will produce forward factors for the forthcoming year, but these will be revised each month after a concurrent adjustment takes place. Forward factors will be available to ABS clients. Clients who require annual forward factors from the ABS should contact Darren Page on Canberra 02 6252 6731 or by email at <darren.page@abs.gov.au>.

Sensitivity Analysis

A sensitivity analysis is presented in page 18 of the 5609.0 publication that looks at how the trend estimates of recent months would be revised if the next month's seasonally adjusted estimates move by a specified amount. This ‘What if?’ sensitivity analysis assumes there will be will be no change to the combined seasonal adjustment factors or the seasonally adjusted series. If there are other revisions affecting the trend data, such as the revision of previous original estimates or of seasonally adjusted estimates under concurrent adjustment, the outcome will be different from that shown by the sensitivity analysis. Typically, since the movement in next month's seasonally adjusted estimate will have a far greater impact than the revision to other months' estimates as part of concurrent adjustment, the results of this analysis still provide a reasonable indication of trend revision behaviour. This analysis will continue to be published under concurrent adjustment.


The ABS is in the process of introducing concurrent adjustment for all seasonally adjusted indicators. The ABS has also demonstrated the advantages of concurrent seasonal adjustment methodology and implemented this method on several main ABS economic indicators such as the monthly Retail Trade (cat. no. 8501.0), the quarterly Business Indicators (cat. no. 5676.0), the monthly Building Approvals (cat. no. 8731.0) and the quarterly Building Activity (cat. no. 8752.0). Users' responses to these changes have been positive.

For further information on concurrent adjustment, please contact the Assistant Director, Time Series Analysis on Canberra 02 6252 6345 or by email at timeseries@abs.gov.au.