WHICH INDICATOR SHOULD I USE?
WHICH SERIES SHOULD I USE?
The original, seasonally adjusted and trend estimates are three separate series describing different aspects of the same data. These series are useful for different purposes.
The original is the best estimate we can make of the level of activity at any particular point in time. These are the estimates you’d use when focusing on history - e.g. auditing, comparing different data sources, assessing market share. Note that different months are not directly comparable.
SEASONALLY ADJUSTED ESTIMATES
Seasonally adjusted estimates are produced by removing seasonal patterns from the original estimates. They are good for performance measures and comparisons - was that advertising campaign effective, how much tourist traffic did we lose because of the Tsunami, is unemployment doing better in QLD than in TAS. Again, they’re very history focused. Different months are comparable but month-to-month movements are usually dominated by irregular effects.
Trend estimates are produced by smoothing noise from the seasonally adjusted estimates. This is the best source of information for forecasting or making decisions about what to do in the future. It is directly comparable at different points in time. Trend estimates are revised as new original estimates become available. This makes sure we have the most up-to-date and best possible estimate.
For some questions you need to consider more than one series. Few retailers would hire many staff in late December, because sales consistently drop in January - something you can see from the seasonal pattern, the original minus the seasonally adjusted.
The original estimates, seasonally adjusted estimates and trend estimates are series that present different but complementary information.
WHY SHOULDN’T I COMPARE ORIGINAL DATA FROM ONE PERIOD TO THE NEXT?
The original data contains the seasonal patterns, residual noise and irregular influences. A comparison of original data from consecutive periods may lead to misleading conclusions if there is a strong seasonal pattern in the data.
For example, if you are looking at the unemployment rate, the original value for December 2009 is lower than January 2010; you might conclude that the underlying unemployment rate is going up. But January unemployment is seasonally high compared to December, and the underlying unemployment rate was actually falling; the Dec-Jan increase was not as big as usual.
HOW ABOUT COMPARING SEASONALLY ADJUSTED DATA FROM ONE PERIOD TO THE NEXT?
Seasonal patterns have been removed from the seasonally adjusted data, however the residual noise and irregular influences are still present. If the residual noise is high it may distort a comparison of seasonally adjusted estimates. Similarly, an unusual event or irregular influence may affect a comparison. For example, consider a series which has an underlying increasing trend. An unusual event such as a strike may lead to a "once-off" low value for one period.
WHICH INDICATOR SHOULD I USE TO COMPARE MONTH-TO-MONTH OR QUARTER-TO-QUARTER PERCENTAGE CHANGES?
Trend estimates are usually preferred to compare data at different points in time as potentially misleading seasonal patterns, residual noise and irregular influences have been removed.
WHY SHOULDN’T I JUST COMPARE ORIGINAL DATA FROM THE SAME PERIOD IN EACH YEAR?
|Series||Recommendation||Benefits and disadvantages|
|Original estimates||Do not use||Usually dominated by seasonal effects; also residual noise and irregular influences.|
|Seasonally adjusted estimates||Use with caution||Provides useful information on the effects of short term, major events. Dominated by irregular and noise, except for series with very little volatility.|
|Trend estimates||Preferred option||The best indicator of underlying behaviour for month-to-month or quarter-to-quarter changes. Recent estimates, usually the last 3 or 4, may be revised.|
Comparing original data from the same period in each year is a crude form of seasonal adjustment which assumes that the seasonal patterns do not change. It does not completely remove all seasonal effects.
Certain holidays such as Easter and Chinese New Year fall in different periods in each year and their effects may distort the observed values. This comparison also ignores trading day effects caused by different day type compositions of the month in each year.
Each month consists of 4 complete weeks plus an extra 1, 2 or 3 days. Different levels of daily activity on the extra days may cause differences in the original estimates for the same month in consecutive years, even though the underlying level of activity is unchanged. Similarly, this type of comparison will ignore any changes to the seasonal pattern over time.
Since the original estimates also contain the influence of the irregular component, a comparison of original estimates may also be distorted if the magnitude of the irregular component is strong when compared with the magnitude of the trend.
However, the major disadvantage of comparing year to year data, whether original, seasonally adjusted, or trend, is lack of precision and time delays in the identification of turning points in a series.
Turning points occur when the direction of underlying level of the series changes, for example, when a consistently decreasing series begins to rise steadily. Using year to year changes in original data may cause delays of up to 6 months in the identification of turning points.
WHICH INDICATOR SHOULD I USE TO COMPARE YEAR APART CHANGES?
|Series||Recommendation||Benefits and disadvantages|
|Original estimates||Do not use||Crude form of seasonal adjustment assuming seasonal patterns do not change. May be misleading as it ignores evolving seasonal patterns, trading day and moving holiday effects. May contain high contribution from residual noise.|
|Seasonally adjusted estimates||Use with caution||May be misleading, because year apart percentage changes in the seasonally adjusted estimates usually contain a high contribution from the residual noise.|
|Trend estimates||Preferred option||Stable measure indicating average trend movement over the year. May not reflect current direction of the trend if there has been a change in the direction of the trend during the year.|
This page last updated 10 December 2012