Impact of COVID-19 on seasonally adjusted and trend series
This page describes how COVID-19 has impacted seasonal adjustment, and why Stats NZ has temporarily suppressed publication of selected time series.
What are seasonally adjusted and trend time series?
In the seasonal adjustment process, the actual value is broken down into three components: trend for medium to long-term movements, seasonal for systematic annual effects, and irregular for short-term unsystematic fluctuations.
The seasonally adjusted values allow us to compare different periods without seasonality contributing to the differences. Trend time series are useful to compare data points against previous peaks and troughs, as well as indicate the direction of the series.
How has COVID-19 impacted seasonal adjustment and trend estimation?
The COVID-19 lockdown and restrictions have caused abrupt changes to actual values for many time series. Our seasonal adjustment calculation initially attributed too much of the unusual values to the seasonal component, and not trend components, and not enough to the irregular component. This would have significantly altered the seasonal pattern and caused large revisions to the seasonally adjusted time series going back years. Hence, we specially treated unusual data points for many time series, so that they were attributed to the irregular component. Special treatment also excludes unusual data from trend estimation.
For series that have been impacted by COVID-19, it is impossible to know how much of this impact will be long-term and part of the trend, or short-term and part of the irregular. Calculating the trend in the usual way could be misleading, and so Stats NZ has suspended publication of the trend for many time series.
Specially treated monthly time series example
The actual value of retail card spending on fuel, from monthly electronic card transactions data, started to fall in late March 2020 as COVID-19 border restrictions and the level 4 lockdown came into effect. Fuel spending plummeted in April 2020, and has partially recovered since, though the actual value in August 2020 was down 19 percent compared with August 2019. The latest fall was partly due to non-essential travel restrictions in Auckland with a regional level 3 lockdown. We suspended publication of the trend series from March 2020 as it does not meet Stats NZ’s usual quality standard.
The seasonal and irregular factors for retail card spending on fuel were showing typical patterns up until February 2020. (See graph below – these are easier to see when ‘irregular factor to August 2020’ is deselected.) The seasonal factor normally ranges between about 0.95 and 1.07, while the irregular factor is usually even closer to 1, except in periods with significant fuel price or volume changes.
We specially treated actual values from March 2020 – reflected and to account for the sudden drop in the irregular component in April 2020 to just 0.3, and partial recovery since.
How long will Stats NZ specially treat some time series?
For series that are only temporarily impacted by COVID-19, these special treatments will be left in place, so that the short-term but large effect of COVID-19 will not have a long-term impact on the trend or seasonal pattern, and post-COVID-19 values can be treated the same as normal.
For series that have permanent or long-term alterations to the trend or seasonal components, such as for travel and migration, seasonal adjustment may be temporarily ceased, until there is enough information to accurately calculate the new trend behaviour or seasonal pattern.
We produce the seasonally adjusted and trend series using the X13-ARIMA-SEATS package developed by the U.S. Census Bureau. See Seasonal adjustment in Stats NZ for more information.
Enquiries
Dave Adair
03 964 8976
info@stats.govt.nz
