The need for a plain language description of the proposed GEKS-T methodology

Arthur Barnett
Arthur Barnett • 4 April 2025

Error message

Access denied. You must log in to view this page.

I was expecting the new version of SUN to include the posts below from the RPICPI User Group in the old SUN as the last post in the series was February 27, 2025 but for some reason they were not included.

The need for a plain language explanation of the GEKS-T methodology is important as the calculation is complex and implies that it represents a lagged indicator possibly centred a year or so in arrears.

All the best.

Arthur.

 

The RPICPI User Group posts are below starting with the earliest post

Arthur (Arthur Barnett) September 27, 2023, 3:11pm 1

Things have been a little quiet of late, but a couple of spreadsheets have been put together to calculate GEKS-T for the BEIS data on petrol and diesel forecourt sales. The accuracy of the spreadsheets has been confirmed with ONS colleagues familiar with the methodology as far as it is practical for ONS to do so.

Unsurprisingly the spreadsheets are complex and difficult to explain. This makes it difficult to post the spreadsheets on SUN without a clear explanation of what is being measured. Leaving it to SUN readers to interpret the spreadsheets is probably not really helpful for most. Thinking about how to explain the spreadsheets tends to lead to the issue of a lack of a plain language explanation of GEKS-T.

The question for ONS is what GEKS-T, or modified GEKS-T, is measuring which is not simple given the complexity of methodology.

The answer is presumably “inflation” but what is meant by inflation.

For the current indices used by ONS, the plain language description is that they measure –

“… the changing cost of a fixed basket of goods and services.”
Consumer Price Indices Technical Manual 2014 Edition
2.1.2 Basic principles

Of course, the real world practicalities of calculating the current price indices mean that the calculations are not as straightforward as the above principle might suggest.

In contrast the proposed GEKS-T is a complex methodology based on both monthly prices and quantities from a group of supermarkets, a 25 month window and a couple of sets of Tornquvist calculations. It is not clear what it is actually measuring and this is made more difficult with the additional methodologies likely to be used in modified GEKS-T.

To simplify the problem of comparing the above Technical Manual principle with GEKS-T it is helpful to consider the basic data for a single product and for each methodology ignoring the additional complexities of GEKS-T.

Consider a simple notation of months t and t+1, and base month b, with the respective monthly prices (P), quantities (Q) and expenditures (E = Q*P).

For the existing indices the change between the months is –

Q(b)*P(t+1)/Q(b)*P(t) = P(t+1)/ P(t)

The quantities cancel out.

Intuitively this makes sense if inflation is to be regarded as the change in prices.

In contrast GEKS-T appears to be a complex concept based around monthly expenditure on products. Simplifying this to a single product, ignoring the 25 month window and the sets of Tornquvist calculations etc, would suggest an outcome related to the following change between months –

Q(t+1)*P(t+1)/Q(t)*P(t) = E(t+1)/ E(t)

The quantities do not cancel out for a single product GEKS-T, and by implication modified GEKS-T, so the expression does not appear to reduce to the change in price between months.

The change would not appear to be solely related to the change in price in that single product but the change in expenditure which arguably would not intuitively be considered by the public as inflation – well, at least, not price inflation.

To repeat, there is an urgent need for a plain language description of what GEKS-T is measuring.

All the best.

Arthur

2 Likes

Arthur (Arthur Barnett) October 30, 2023, 12:46pm 2

On reflection, perhaps, it was too early to ask ONS for an official plain language description of the proposed GEKS-T methodology.

A more appropriate question for ONS may, perhaps, be along the lines of –

Do any plain language descriptions – in English or another language – already exist for the proposed GEKS-T methodology?

There has, apparently, been a lot of work carried out over a number of years both domestically and internationally on this methodology and as a statistician the hope would be that standard statistical practice would mean that there would be description of what the methodology measures.

One risk is that statistical offices leadership may have got carried away with how such complex methodology can be implemented and what it is actually measuring may have become lost along the way.

The above may seem a nerdy statistical point but, for example, there is arguably an important read across to the Code of Practice and whether the proposed new methodology meets the requirements of the code.

Now in practice price indices by their nature are subject to less than ideal methodologies for their more obscure aspects where the available data is problematic. But food prices at supermarkets is a far from obscure area. For the public, parliament and government they are front and centre for price indices particularly in the present cost of living crisis. Also the new data from participating supermarkets would appear to offer an excellent source for sampling food prices in a way that would be compatible with the existing standard methods for which there is a widely accepted plain language description.

A further question therefore arises of whether UK price indices should retain their National Statistics status – accredited official statistics status in the latest OSR terminology – if an established methodology for a key part of those indices is replaced by one that does not have an agreed plain language description that reflects the contemporary day to day reality of supermarket prices in a way the public can readily understand?

This latter question is arguably more for the OSR rather than ONS.

All the best.

Arthur

1 Like

Arthur (Arthur Barnett) July 9, 2024, 10:04am 3

Colleagues may be interested to see the link to the Eighteenth meeting hosted by Statistics Canada and Bank of Canada, Ottawa, 13-15 May 2024 – Ottawa group meeting - Meeting

It will not come as a surprise that my attention was immediately drawn to the Jacek Białek paper “Evolution of the GEKS index”.

It does not appear to address the issue of providing plain language descriptions of the methodologies for the average user.

The introduction the paper states –

“As it is commonly known, GEKS-type indices (e.g. the GEKS, CCDI, GEKS-W indices) have good axiomatic properties, are close to the economic approach since they are based on superlative indices”

So arguably the audience for the paper would appear to be clear.

Of course, the UKSA chair’s criticism during the general election of the PM for not providing an explanation of the statistics he used that would be suitable for the average person came too late for a paper at the Ottawa conference –  OSR highlights the importance of transparency in election claims – Office for Statistics Regulation

All the best.

Arthur

2 Likes

shaunrc999 (Shaun Richards) August 7, 2024, 2:11pm 4

Hi All

I would like to second Arthur’s request for a plain language version of GEKS-T methodology.

The reason why I am doing so today is that the ONS is asking for questions but apparently forgetting to provide answers when people do ask.

https://x.com/ONS/status/1820783474183152056

“We’re seeking views from users of our supplementary consumer prices item indices and price quotes microdata before grocery scanner data is introduced into headline consumer price inflation measures from March 2025.”

Regards

Shaun Richards

1 Like

Arthur (Arthur Barnett) September 19, 2024, 12:55pm 5

There may be progress of a sort on a plain language description of GEKS-T – see the minutes of the 26 April 2024 APCP-Stakeholder panel published 28 August 2024 https://uksa.statisticsauthority.gov.uk/wp-content/uploads/2024/08/APCP-S-Minutes-Apr-2024.pdf

The original post at the start of this topic was restricted to asking for a plain language description of GEKS-T so that the spreadsheet setting out that complex calculation could be meaningfully shared on SUN.

However the proposed methodology for use in price indices is not simply GEKS-T but GEKS-T with a “mean splice”. Whereas with significant effort the ONS GEKS-T methodology could be derived from the technical article “Introducing multilateral index methods into consumer price statistics” the explanation of a “mean splice” was confusing and not sufficiently clear to be included in a calculation in Excel or otherwise.

There is apparently no mention of the Excel spreadsheet calculation in the APCP-S April minutes or of the limited objective of the need for SUN colleagues to be able to more readily understand the spreadsheet. So it is perhaps surprising that the panel discussion appears not to have been extended to the full methodology of GEKS-T with a mean splice. The extended description is also required if an average member of the public is to understand the new price index methodology.

To keep this post reasonably short it will not go into detail about the entry in the minutes about GEKS-T but simply ask a few questions seeking clarification of those minutes.

Would ONS publish on SUN the slides, and all related material, from the presentation presented by ONS at the APCP-S April meeting?

Was the further paper referred to in the minutes of the April meeting provided at the APCP-S July meeting and, if so, would ONS publish this paper?

Would ONS extend the methodology paper to include a clear explanation of the “mean splice” to enable the Excel GEKS-T calculation, or similar, to be extended if necessary to cover this part of the methodology?

The minutes of the April meeting were published on 28 August 2024 apparently after the subsequent meeting took place in July. It would be helpful if ONS published the minutes of the July meeting in good time before the next panel meeting.

All the best.

Arthur

GEKS-T extract from APCP-S Minutes Apr-2024

2. Communicating the impact of GEKS-Törnqvist on ONS Consumer Price Statistics

2.1 Liam Greenhough (LG) gave a presentation that aimed to address some of the commonly-raised questions about the GEKS-Törnqvist multilateral index methodology. LG explained that GEKS-Törnqvist is not the simplest of methods used by Prices Division but neither is it the most complicated.
LG further explained some of the work that has been done to tailor communication of the method to different audiences.

2.2 The Panel welcomed the work that had already been done to explain the method to the public, noting that a detailed methodological paper had been published which had likely been of use to statisticians. The Panel also welcomed some of the less technical explanations of GEKS-Törnqvist that ONS have published.

2.3 The Panel then noted that a further paper could be produced that provides more detail for members of the public that have less of a mathematical background. Such a paper might include details on the reasons for using GEKS-Törnqvist and the advantages it brings when analysing large datasets but it should also aim to answer the questions people have already began to pose within the wider statistical community.

Action: ONS to review the Panel’s feedback on communicating the use of GEKS-Törnqvist and to bring this issue to the July Panel for further discussion

1 Like

Arthur (Arthur Barnett) February 27, 2025, 10:41am 6

Plain language descriptions of GEKS-T and also other lagged indicators like rents remain a yet to be resolved.

It is helpful that the move to groceries scanner data has been delayed until 2026 but there is apparently limited information on what analyses and other information will be provided what will allow the impact of using another lagged indicator to be assessed by users.

The notayesmaneconomics blog UK Inflation statistics are making a bad error about rents | Notayesmanseconomics's Blog pointed out that in December 2024 using a flow measure of rental inflation rather than the current lagged stock measure would likely have meant that the CPI was not above target but on target. In times of crisis lagged measures tend to lower the official estimate of a rapid increase of inflation and then raise that estimate when the crisis is abating which is not desirable for setting important policies like interest rates. An unnecessarily high interest rate will impact of home owner and buy to let mortgages and for a lot of people raise housing costs.

To quote from the ONS bulletin “Price Index of Private Rents” QMI  Index of Private Housing Rental Prices QMI - Office for National Statistics, page 9 –

“The monthly dataset of properties used in the calculation of the Price Index of Private Rents (PIPR) uses a stock measure of rental prices… although the Office for Statistics Regulation’s (OSR’s) Systemic review of Public Value: Statistics on Housing and Planning in the UK https://osr.statisticsauthority.gov.uk/wp-content/uploads/2017/11/Systemic-Review-of-Statistics-on-Housing-and-Planning-in-the-UK20171110-corrected.pdf points to recent research that suggests a flow measure may be worth considering…”

A flow measure doesn’t appear to have been considered by ONS.

A further lagged indicator of an important element of household budgets like groceries is likely to cause further problems for government and Bank economic policies in future crisis situations.

The question for ONS is to ask what analyses they will publish to support a move to GEKS-T for scanner data – historic series including December 2024 for the CPI comparing the proposed lagged indicators with unlagged flow indicators would be helpful.

All the best.

Arthur.

Be the first one to comment


Please log in or sign up to comment.