Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
It sometimes takes years to get “good”, “correct” figures for UK growth, so FT Alphaville is going to be gentle towards the soft animal of our body for being a week late to an interesting analysis by James Rossiter of TD Securities:
[S]peculation has risen that there is a growing issue with the [Office for National Statistics]’ seasonal factors in recent years, resulting in an over-stating of H1 GDP growth and an under-stating in H2 growth (seasonal factors don’t affect the year as a whole). This can be seen both in the pattern of quarterly growth rates and in the monthly GDP data, which seems to surge in H1 and flat-line in H2 (see 1st and 2nd charts).
As Rossiter nods towards, concerns over seasonality aren’t new. There’s a well-established concern that Britain’s ✨ post-pandemic ✨ growth tendency — quick out of the gate in the first quarter of the calendar year, then gradually slower each quarter thereafter — is, as the kids probably don’t say anymore, sus. As Reuters wrote in February:
While this pattern would not be unusual as a one-off due to a downturn in a particular year, the repeated pattern has raised concerns among economists that a problem with the ONS’ seasonal adjustment of data has developed since the COVID-19 pandemic…
[Some] economists say this process appears to have stopped working so well for British data since 2022, making it harder to forecast future trends and sometimes making it look as if the economy was close to recession when it was not.
Nethertheless, an analysis published by the ONS last September found “there is no residual seasonality in the main aggregate outputs for quarterly and monthly gross domestic product (GDP) estimates”. A blog post by James Benford, director-general for surveys and economic and social statistics, picked up the theme again last month. Benford wrote:
It is inherently difficult to judge in real time how much of a variation in quarterly growth reflects a new, regular seasonal pattern, as opposed to being a more idiosyncratic variation over the course of the year. The methods that we use to seasonally adjust GDP data need a run of 3 to 5 years of data — i.e. 3 to 5 consecutive observations on growth in a given quarter — to establish what the new regular season pattern is following a change. Given seasonal patterns have been changing, we are monitoring seasonality particularly closely and have expanded the breadth of tests that we run, including by testing whether there is evidence for residual seasonality over short-run periods.
[…]
There are a range of factors that may be causing the pattern of activity over the course of the year to change and it is difficult to judge what the lasting regular pattern will be.
Benford added that seasonality effects would be kept under “close review”.
Picking up the theme, TD’s Rossiter repurposed a double-seasonal adjustment technique, which was deployed by the San Francisco Fed about a decade ago amid “shockingly weak” US growth figures. The approach basically involves doing an adjustment again on the logic that if nothing is awry, nothing should change. It’s sort of like when Alphaville’s treasured colleague Nigella Lawson double-butters toast. Or maybe it’s nothing like that. Or, as the SF Fed wrote:
If there were no residual seasonality in the published real GDP, then these seasonal factors would all be essentially zero, and our procedure would be innocuous. However, a statistical test rejects that hypothesis. Instead, the evidence indicates the presence of residual seasonality in the published data and supports our double seasonal adjustment procedure.
Noting that “seasonal adjustment is an imprecise and uncertain statistical exercise, and our results could overstate or understate the true amount of residual seasonality”, they found “the published real GDP data still exhibit calendar-based fluctuations — that is, residual seasonality”.
So does the UK have the same issues? Yes, reckons Rossiter:
We take the approach the SanFran Fed used a decade ago on US GDP data to double-seasonally adjust the data, in essence seasonally adjusting the ONS’ already seasonally adjusted data to look for repeating patterns. Since 2023, the ONS’ seasonally-adjusted GDP series has been more volatile than our double-seasonally adjusted GDP
He continues:
Double-seasonally adjusting the data yields statistically significant results, and shows increasing positive bias in UK Q1 and Q2 GDP growth estimates, while the Q3 and Q4 growth figures are under-stated…
Our evidence suggests that reported Q1 GDP growth of 0.6% q/q might be overstated by as much as 0.25ppt. Of course, that means that second half growth might be understated by as much as 0.2ppt in each of Q3 and Q4 this year if seasonal factors aren’t adjusted.
Or, as rectangles:
As with pretty much every potential national statistics snafu, the implications of this are quite hard to disentangle. As we wrote last year, monthly GDP figures are almost always wrong in the long term, but that doesn’t stop people using them to build narratives. We (perhaps over-optimistically) assume that most big allocators already factor in potential weirdness of this variety, but it’s hard to deny that they can shape political narratives.
Precariat chancellor Rachel Reeves welcomed March’s bumper (0.3 per cent) growth figures. That may seem naive given the established growth pattern, but perhaps time isn’t on her side.
Further reading:
— Bye, bye, monthly GDP try?
