Statistics New Zealand, our under-resourced national statistics agency, has just taken another small but useful step forward. As from the June quarter CPI this week they will be publishing additional series seasonally adjusting the CPI data. Most other major SNZ series are published, and analysed, in seasonally adjusted terms and there is clearly significant seasonality in some of the components of the CPI. Analysts could previously do their own seasonal adjustment but few did, and none of those estimates got any media coverage. In time, we might hope that the seasonally adjusted series will become the ones analysts and media focus on.
Last week, SNZ released the seasonally adjusted historical data. True to my hope that people would focus on the seasonally adjusted series, I went to the table which showed quarterly non-tradables inflation in seasonally adjusted terms. As one who has been sceptical for a long time of the Reserve Bank’s story that core inflation is just about to turn up, I was somewhat taken aback to find SNZ reporting that seasonally adjusted non-tradables inflation had been 0.8 per cent for the March quarter, up from 0.5 per cent in each of the previous three quarters.
But then I looked at the rest of the series, and was disconcerted to find that in each of the previous three years, the March quarter inflation rate had been higher than the quarterly inflation rates for each of the other three quarters. That looks a lot like some residual seasonality that hasn’t been picked up in the seasonal adjustment. I took comfort from the fact that the March quarter increase was the (equal) lowest March quarter increase over 2011-2015. But something about the seasonally adjusted series doesn’t look quite right.
(If anyone from SNZ has an explanation, I’ll happily report it.)
In terms of future innovations, backdating the tradables and non-tradables breakdown, as official (even if “experimental” or “analytical”) SNZ series, would be helpful. The Reserve Bank has estimates going back to the early 1990s that we derived from detailed SNZ data (not initially very scientifically – I and one of my staff went through the components with a pen, labelling each T or NT), but it would be useful to have semi-official series going further back than 2006. Producing good linked estimates of trimmed mean and weighted median inflation further back would also be helpful.
And then, one day, perhaps we might dare to hope that SNZ might be funded to allow the publication of a monthly CPI, joining the rest of the advanced world (other than Australia, the only other country producing a quarterly CPI).
UPDATE: A commenter reminds me of the role of the tobacco tax increases. Here is the SNZ comment
Non-tradables are goods and services that do not face foreign competition. They include rentals for housing, and services such as hairdressing. The non-tradables series is seasonally adjusted. The algorithm suggested the series probably wasn’t seasonal, but that there was emerging seasonality in the data over the past three years. Figure 5 below shows this, where the pattern from 2011 in the actual series appears to be more consistent than previously. We see the biggest peak occurs in the March quarters in the actual series, but this effect is lessened in the seasonally adjusted series.
The multiple-point method detects seasonality in the full time series when we include the more recent quarters. We investigated to see if any one contributing series was influencing the pattern and found that cigarettes and tobacco did increase the size of the peak in March quarters since 2011. Checking the diagnostics, we found it was a marginally seasonal series and decided to seasonally adjust.
Sadly, this suggests that the seasonally adjusted series is not very helpful.