A quick note on seasonally adjusted car sales
Fri 4 Aug 2017 by Mieke Welvaert in Transport

Infometrics uses seasonally adjusted data to gauge month-to-month trends in car registrations.  The purpose of doing so is to track whether there are any changes in momentum in car sales growth.  It also helps us to look through the seasonal patterns that normally drive sales up or down in any given month.  Examples of a regular seasonal pattern include strong growth in new car sales due to rental car purchases in October and November, or the lift in sales around the time of Fieldays in June.

There are a few different ways in which we can track trends in car sales data:

1.         Using annual running totals

2.       Comparing months or quarters to the same period a year earlier

3.       Comparing seasonally adjusted data

All these processes allow us to “look through” seasonal trends and track the overall growth in the data, but each have their limitations.

The limitation to the first option is that it is slow moving – it takes several months for any change in the trend to become apparent in the data.  In other words, because annual data looks at the whole year, it might not pick up emerging weakness in car sales if the first 6-9 months of the year were strong.

Comparing months or quarters to the same period from a previous year is an effective way to gauge growth.  But volatility or aberrations in the data a year ago can greatly alter the annual percentage changes that are calculated.  For example, in June next year, we’re unlikely to see a similar surge in rental cars during that month and, as a result, the annual percentage change is likely to be weaker for June 2018 than the months around it.  Even though the measured annual percentage change is likely to be weaker, the underlying trend in sales activity will probably not be as soft.

Using seasonally adjusted data as in option three, we remove the seasonal effects to allow us to compare between any two periods of time.  Perhaps we want to compare the June quarter to the March quarter.  With seasonally adjusted data we can do this without having to try and make a subjective allowance for the lift in sales that normally happens in the June quarter due to sales at Fieldays.

How do we calculate the seasonally adjusted data?

Infometrics uses the statistical standard X-12 ARIMA program to generate seasonally adjusted series.  Essentially, this program estimates the historic seasonal pattern (putting more weight on earlier years) and then removes the seasonally-related growth from the observed series.

Ideally, this process would result in the trend line as depicted in Figure 1, from which we can measure growth over any period we choose and know that it represented the true (seasonally adjusted) change in underlying sales activity.  However, in reality the random factors are still included in the seasonally adjusted series.  As a result, we must exercise caution when interpreting data on a month-to-month basis.

Nevertheless, the benefit of seasonally adjusted data is that it often fills a gap in our knowledge regarding recent trends in sales activity.  In seasonally adjusted terms, car sales were softening over the entire six-month period from November 2016 to April 2017, even though the message from the unadjusted data was that activity in each of these months was higher than a year earlier (due to growth in the earlier part of 2016).  Weakness in seasonally adjusted data gives us an early indicator of emerging trends and indicates that activity might be cooling down before we see it appear in annual growth measures. 

Figure 1

Related Articles