RPI/CPI User Group

last person joined: 15 days ago 

To foster co-operation and the exchange of information between the ONS, its advisory bodies and other users. Please see 'Aims of the RPI/CPI User Group' in the library for further details.

1.  Technical APCP Publications

Posted 14-02-2017 12:38
Dear All, 

This morning we published the agenda, minutes and papers from the Technical Advisory Panel on Consumer Prices meeting, held on Friday 20 January 2017. These can be found on the Technical APCP landing page on the UKSA website. 

Kind regards,

Helen (Prices Development, ONS)

2.  RE: Technical APCP Publications

Posted 15-02-2017 03:25
Dear Helen Sands:

Thank you for these documents. A quick observation on the formula effect: the use of January links seems to be making an important contribution to the size of the effect. Simply replacing the January links with December links would probably reduce it considerably. My own preference would be for annual links (i.e. calculating index numbers for 2017 with annual 2016 values as the base, but this would be against the current rules for calculating an HICP. There is no reason one could not calculate a household inflation index in this way, though, and in the long run I believe the rules for calculating an HICP should change. Annual links are more stable than monthly links whatever calendar month is chosen and are particularly advantageous for seasonally disappearing items that may not be available in January (Christmas trees) or in December (in my country, tickets to Canadian Football League games; I am sure there are similar examples for the UK).

Whatever its scope, some study of the choice of link period as it affects the formula effect should be considered. 

Best regards,


PS I won't have the use of my own computer again until February 23. I find what I can do on my son's is pretty restricted or this would be a longer and better researched e-mail.

3.  RE: Technical APCP Publications

Posted 16-02-2017 11:05
Hi Andrew

We have had some discussion of chain linking before.  I posted something around 26th Nov 2016 following up your suggestion around that time.  There are different options for chain linking as the discussion showed and although the CPI is constrained for the time being by EU legislation, CPIH and RPI are not.  So I agree that a change is desirable.

Having said that I would not expect such a change to make much difference to the formula effect.  What now seems ages ago, I showed, by comparing the Jan formula effect averaged over many years with the formula effect for other months of the year averaged over the same years, that the difference in chain drift between the RPI and CPI was less than 0.02% per annum compared with a formula effect of about 1.0% per annum.   Changing the chain linking arrangements will therefore have a very small impact on the formula effect.

The formula effect is caused almost entirely by using a non-linear transformation in taking averages in the CPI.    I don't know of anywhere else in statistics where this is done, except perhaps in outlier treatment.


4.  RE: Technical APCP Publications

Posted 18-02-2017 17:33
Dear Gareth:

Thank you for your reply. I went back to look at the 1999 paper by David Fenwick on the subject, "The impact of choice of base month and other factors on the relative performance of
different formulae used for aggregation of Consumer Price Index data at an elementary level". Unlike you, he doesn't actually provide a number for the reduction in the formula effect that would result from using a December link period for the CPI and the RPI. He does, however, say:"the difference between the index computed on arithmetic means and one computed in a
geometric mean reduces with a move to a December base month". Also, he notes that: "the depth of January sales has increased overtime particularly for those
goods such as clothing and footwear and household furnishings traditionally offered at knockdown prices in sales." Clothing prices is, of course, where the big increase in the formula effect has occurred. So your mutli-year estimate, while it may be accurate for that period, may underestimate the impact of a change in base period going forward. Also, the overall impact is not the only consideration. For some people the differences for particular categories like clothing are of interest in themselves. I still think a study would be warranted.

Yes, the geometric mean is not a linear transformation of the price relatives and that has its inconveniences. For one thing, one can't calculate contributions to percent change that are additive to the percent change of the aggregate. However, I don't see this as something that disqualifies it from consideration as a formula for elementary aggregrates. In most cases, it is still the preferred choice. Where prices can go to zero from positive and back again, the ratio of mean prices (Dutot) performs better than either the arithmetic or the geometric mean of price relatives.

Best regards,