The new arrangements will replace the Consumer Prices Advisory Committee (CPAC). In future there will be two advisory panels – a stakeholder panel and a technical panel. The stakeholder panel will advise on which price measures should be used for what purposes. It will have an independent chair, unlike the CPAC, which was chaired by the National Statistician. Likely issues for the stakeholder panel might include the measure of inflation which should be used by the Bank of England to set interest rates and the index or indices used to uprate things like pensions, investments, contracts and the minimum wage.
In the past, such matters have tended to be glossed over by CPAC in favour of more technical issues. The technical panel will advise on all the technical aspects of price index construction, including practical issues such as data sources. The stakeholder panel will have wider user representation than was achieved under CPAC and the technical panel will have more external input. How well these two panels work together will determine how successful the new arrangements will be.
The key to this is clarity of role. The technical panel appears to have the clear role of seeking to produce the best quality statistics to meet the requirements of the stakeholder panel. But what will the role of the stakeholder panel be? It is not good enough for users to simply demand a particular specification without providing the rationale for it. There is a need for user 'requirements' to be challenged, and I hope the new independent chair will see this as a key part of his or her role. Perhaps also, the technical panel will sometimes challenge the stakeholder panel requirements. An interesting decision for the National Statistician and Stakeholder Chair will be whether to include a Eurostat representative on the stakeholder panel, which would allow EU requirements to be challenged also.
A few examples illustrate the general issue. Those with responsibility for national budgetary and monetary policy require a macroeconomic measure of inflation. However, the Bank of England has been instructed to use the Consumer Price Index (CPI) as its inflation target measure when setting interest rates. Yet the CPI is not a macroeconomic measure, it only measures consumer prices and excludes the effect of inflation on the public and business sectors of the economy. Also, it does not normally permit revisions when better or more data becomes available.
When the available evidence changes one would hope that national policy makers would be open to changing their policy, so shouldn't they be using a measure which allows for revisions? There is one such index available 'off the shelf' - the Gross Domestic Product (GDP) deflator - which is part of the National Accounting Framework (and which covers the public and business sectors as well the consumer sector). The target inflation measure for the Bank of England has already been changed once and can be changed again.
Then there is the tendency for stakeholders to try to dictate the technical solution, rather than simply state their requirement and its rationale. There are two prime examples of this – the choice of formula for combining price changes at the lowest level of aggregation of a price index and the methodology to be used to represent owner occupiers' housing costs.
First, let's take the choice of formula, there has been considerable pressure on the Office for National Statistics from some users to apply the Jevons formula (geometric mean). This is one technical possibility among many others, only a few of which have been tried in practice. The need for a formula of this nature arises from the inadequacy of data sources, so that expenditure weights at very detailed levels have not been available. So some kind of formula has had to be used to approximate the index that would have been used if the relevant weighting data were available. What is needed, is for stakeholders to specify the index ideally required (the target index) and for the technical panel to provide the formula which best approximates that, or alternatively to develop better data sources so that the need for such approximation is less necessary.
The word 'approximate' also needs some consideration. Statistically, it implies the adoption of an 'error function' or 'loss function'. In this instance there are two main candidates for a loss function – the mean squared error and the bias. Mean squared error is equal to the sum of the variance and the square of the bias, and so encompasses both variance and bias. But is this what users really want? Many users who rely on indexation for different purposes are mainly concerned with long term accuracy and may not be bothered much about short term variance. For example, a member of one of the modern average salary pension schemes will rely on indexation of their pension rights from when they join the scheme until they die. An extreme case perhaps, but long time period horizons are a feature of indexation provisions. This suggests that the bias, undiluted by considerations of variance should be the loss function.
The treatment of owner occupiers' housing costs in price indices in littered with different opinions. If the GDP deflator is used as an inflation measure, the issue is forced by the National Accounts which require the use of imputed rental value as the measure of owner occupiers' housing costs. But in a measure of inflation experienced by consumers, and which will probably be used in indexation arrangements, the answer is less obvious. So we need to check what underlies requirements. There are two sides to indexation – those who receive and those who pay.
Generally, the government is opposed to measures which fluctuate a lot between years, especially when the fluctuations result from policy changes, notably changes in interest rates. House prices also cause a lot of volatility because they include the value of land which is priced in a speculative market. So the government opposes the main existing treatment of owner occupiers' housing costs as represented in the Retail Price Index (RPI). However, consumers suffer exposure to variations in both house prices (including land value) and interest rates. A possible compromise between different users is to apply smoothing to such components of the index. There are several other methodologies available, so it is clear that the treatment of owner occupiers' housing costs will test the new governance arrangements for price statistics to the limit.
The views expressed in the Opinion section of StatsLife are solely those of the original authors and other contributors. These views and opinions do not necessarily represent those of The Royal Statistical Society.