Did the Bank of England base its hiking pause on bad data?

From mainFT:

UK services activity contracted by less than first estimated in September, according to a closely watched survey published on Wednesday.

The final S&P Global/Cips UK services PMI business activity index, a measure of the health of the sector, was 49.3 last month — down slightly from 49.5 in August, but well above the flash estimate of 47.2.

It contrasts quite sharply with another mainFT report, written September 22nd:

UK economic activity has fallen at the fastest pace since January 2021, according to a closely watched survey that suggests the chances of a recession have increased.

That data release coincided with the pound reaching a six-month low.

We’ve written a lot recently about economic statistics and narratives so, uh, let’s continue to do that.

The well-worn cliché (itself a well-worn cliché) about PMIs (purchasing managers’ indices), in evidence above, is that they are closely watched. And it’s true, because the surveys — based on businesses’ responses to a number of questions about conditions — have tended to do a pretty good job of acting as faster indicators for growth.

They are not, however, without issues. The onset of the pandemic exposed some of the problems PMIs have with handling acute movements within some sub-components. Former Alphavillain Claire Jones wrote about this a lot.

Many PMIs arrive in two phases: the “flash” reading, released before the end of the month being measured and based on the responses received at that point, and the “final” reading, released early in the subsequent month. The new beef is over the gap between these readings.

For September, the gap between flash (46.8) and final (48.5) readings in the “composite” PMI, which combines the services and manufacturing surveys, was the third biggest on records going back to 2006 (a PMI score below 50 indicates a contraction compared with the previous month, if you didn’t already know that please scroll up to enjoy the cover image joke):

Commenting on the flash report, S&P Global economist Chris Williamson said:

The disappointing PMI survey results for September mean a recession is looking increasingly likely in the UK.

While in today’s update, economic director Tim Moore says:

Although only modest and slower than indicated by the earlier ‘flash’ PMI reading, the downturn in UK service sector output was the greatest seen since the beginning of this year and stood in contrast to solid growth during the spring months.

The size of the miss has prompted some raised eyebrows in the City. Allan Monks, from JPMorgan, wrote in a note today:

Revisions to growth have been in the spotlight recently, with the ONS having taken up its estimate of GDP during the pandemic by a couple of points. But the revisions to the PMI deserve a lot more attention. The average revision in the final survey has been a positive 0.2%-pt since 2006. But since 2021, that upward bias has grown to 0.4%-pts. And in 2023, it stands at 0.5%-pts. This should be borne in mind in future months.

So, currently flash PMIs look about two-and-a-half times as inaccurate as the long-run average. Not very helpful.

Alphaville chucked the numbers into a chart, which came out like this:

How consequential could this be? Notably, the Bank of England’s Monetary Policy Committee was given advanced access to the (misleading) flash data ahead of its knife-edge decision to hold rates on September 21st. From the minutes of that meeting:

Ahead of its final meeting, the Committee was made aware of the flash S&P Global/CIPS UK composite PMI for September that would be released publicly on Friday 22 September.

For JPMorgan’s Monks, that could have been enough to shift the narrative:

The upward revision to the September PMI feels especially consequential . . . Had the initial print been closer to the latest data, it is possible rates would have instead risen 25 bps in September, as had been expected at the time. That said, a big downside surprise in inflation on the eve of the decision had in any case made the outcome look very close.

JPMorgan is forecasting slow annualised growth of 0.8 per cent, despite the negative reading. Monks, who thinks the real slowdown point is at 46, told FTAV:

It tells you that a net balance of respondents are indicating lower output over the month, but not by how much. Some firms output will matter more for GDP than others. So a level below 50 doesn’t automatically mean falling GDP.

What went wrong? We reached out to S&P Global’s Moore, who kindly answered our questions and provided us with the data we’ve used in this article.

He told us there were some “strong late responses on the UK Services PMI survey” after the flash cut-off, meaning those gathered from September 21st to 27th, and than there had been “a slightly higher than usual number of final responses relative to flash” (we’ve chart this below). There was no unusual variation in total response numbers, he added.

This is interesting, but doesn’t really solve the mystery.

One person who saw this coming is Pantheon Macroeconomics’ Samuel Tombs, who had a very sceptical response to the flash readings. From his note last month:

The composite PMI has incorrectly signalled falling GDP many times before, and likely is too downbeat again now.

We can’t be sure until the final report, but seasonal adjustment likely is depressing September data too much.

It makes little sense for the economy to be struggling more now, given the emerging recovery in real wages.

He continued:

One plausible theory is that big swings in activity during the pandemic might have distorted the seasonal factors. A large increase in activity is now ‘expected’ by the adjustment process, given the strong post-lockdown growth in September 2020 and 2021.

S&P’s bespoke seasonal adjustment procedure also might be throwing a spanner in the works. It uses the outdated X-12 method, rather than the X-13ARIMASEATS package now used by most statistical authorities; and it has a “proprietary method” for removing the impact of outliers . . . 

August’s services PMI would have been 0.6pp higher than reported if S&P had adjusted the raw data using X-13 and made no discretionary tweaks.

Vindicated by the change today, Tombs said:

The composite PMI is no longer an implausible outlier, after the sharp upward revision to the flash estimate. At the time, we suggested that September’s first estimate was too low due to erroneous seasonal adjustment, but we’ll never know for sure, given that S&P only publishes the unadjusted data with the final estimate.

Where does that leave us? Alphaville has pretty much limitless patience for the data wonks who are trying to get this stuff right, but, once again, the forces of Big Macro Statistics seem to have to brought Britain down.

And in the case of the BoE — obviously they will never disclose the impact one bit of data had on their decision making, but citing a survey that turned out to be wrong is not a great look.

Source: Economy - ft.com