Can’t weight anymore!
South Africa’s use of national accounts and point of sale data to update
the CPI basket and weights
Patrick Kelly and Matlhatsi Mogalanyane
Statistics South Africa
Meeting of the Group of Experts on Consumer Price Indices
7-9 June 2023, Geneva, Switzerland
Abstract
Funding constraints and COVID-19 lockdowns prevented Statistics South Africa (Stats SA)
from surveying household expenditure data in time for a five yearly update of the consumer
price index (CPI) basket and weights. This despite the importance of a relevant CPI for
economic policy and the urgency of updated poverty information. In 2022, Stats SA instead
used recently benchmarked national accounts estimates together with detailed sales data
provided by retailers to update the weights and basket. The paper explains the technical
methods used to determine changes to the basket and weights and highlights the
importance of user consultations during the project. An assessment of the results lays the
methodological groundwork for future weight updates. Key features of the update included
using a pre-COVID year as the weights reference period, abandoning price updating and an
uneven focus on retail goods at the expense of services.
Contact
[email protected]
[email protected]
1
1. Introduction
According to international standards, the weights and basket of the consumer price index
(CPI) should be updated at least every five years. This is to ensure that changes in
consumer expenditure are reflected in the weighted aggregates of the measure of inflation.
South Africa subscribes to the International Monetary Fund’s (IMF) Special Data
Dissemination System (SDDS) and considers its CPI to comply with key international
methods.
What happens when the country’s fiscal framework does not fund a household expenditure
survey (HES)? Faced with the possibility of failing on a key feature of international best
practice, as well as missing changes in consumer behaviour, Stats SA adopted a new
approach to update the CPI basket and weights.
The paper explains how the CPI benefitted from a timely revamp and benchmarking of the
national accounts, and the methods used to adjust the weights. Detailed point of sale (POS)
data from retail chains provided the basis for changes to selected categories of the basket.
2. Background
2.1 Household expenditure survey
The South African CPI basket and weights were previously updated in January 2017 based
on a household expenditure survey – the Living Conditions Survey (LCS) – conducted over
12 months during 2014 and 2015. No additional funding was allocated by National Treasury
for this survey, and it only took place due to skimming of budgets from other areas of Stats
SA.
Obtaining full funding for the next expenditure survey was seen as essential because the
survey is designed to be representative of the entire population with the ability to provide
geographic and population group breakdowns. Importantly, the survey captures expenditure
on all products in a particular period, providing for calculation of detailed expenditure
proportions. The revised CPI manual (IMF 2020, p108) affirms that a household expenditure
survey is “the primary data source for deriving expenditure shares for the goods and
services covered by the CPI.”
Following the 2014/15 LCS, Stats SA made annual requests for a permanent budget for an
expenditure survey. Different options, including an annual, multi-module household survey,
to a stand-alone expenditure survey were offered. Due to the requirement that an
expenditure survey cover 12 months and the detail of the data required from households, the
survey is more expensive than most household surveys. The approximate budget was
R300 million spread over two financial years. This was more than 10% of the entire annual
Stats SA budget of about R2,5 billion. Unfortunately, no additional funds were allocated, and
it became clear that no fresh expenditure data would be collected before 2019/20 to allow an
update by January 2022 – five years after the previous basket and weights were updated.
Approximately 1/3 of the required funds were provided in the 2020/21 financial year.
Although this was inadequate to run the survey, the COVID pandemic set in and scuppered
any possible alternative uses of the money.
2
2.2 The structure of the South African CPI
The structure of the product and geographic classification of the CPI is important to
understanding the limitations of not having a comprehensive expenditure survey to update
the weights.
The South African CPI closely follows the international Classification of individual
consumption by purpose (COICOP) structure as illustrated in Table 1. This structure forms
the basis of what data sources are most appropriate for weights at the different levels of the
classification.
Table 1. Classification hierarchy of the CPI
COICOP level Name Example
2-digit Division Food and non-alcoholic beverages
3-digit Group Food
4-digit Class Bread and cereals
5-digit Sub-class Bread
8-digit Indicator product White bread
12-digit Sampled product Albany 700g loaf of white bread in store xxx
The CPI divides the country into three categories: large or metropolitan urban areas
(metros), other (non-metro) urban areas and rural areas. The headline CPI covers all urban
areas. In most provinces there is one weight for each category, however, Gauteng province
has separate weights for three metros (Johannesburg, Tshwane and Ekurhuleni); Eastern
Cape province has separate weights for two metros (Nelson Mandela Bay and Buffalo City)
and Kwa-Zulu Natal province has weights for two municipalities (eThekwini and Msunduzi).
Weights are allocated to each indicator product in each of these areas which then comprise
the elementary indices of the CPI at an 8-digit level.
Indices are published in aggregated format at national and provincial levels. The statistical
release contains information at the 3 and 4-digit level while more detailed indices are
available in an Excel format on the Stats SA website.
In addition to the classification-based indices, Stats SA computes a number of analytical
indices, aimed at shedding light on the inflation experienced by specific groups. These
include geographic-based indices for rural areas and the total country, indices for
expenditure deciles, a CPI for pensioners as well as core measures of inflation.
3. The solution
3.1 National accounts – international perspectives
National accounts data are used by a number of statistical offices to update CPI weights.
Specifically, the household final consumption expenditure (HFCE) component of the
accounts measures the expenditure by households using the same concepts and
classification as the CPI. The CPI manual emphasises that national accounts should be
seen as “an alternative source for deriving CPI expenditure weights” (IMF 2020, p110) and
should not replace an expenditure survey. However, “national accounts data may be used
when the household expenditure survey is conducted too infrequently to ensure the reliability
3
of the CPI” (p33). The manual also acknowledges that the value of national accounts data as
an alternative to a HES is that it is updated regularly.
Advantages of using national accounts data are that they are updated frequently, are
nationally comprehensive and that the expenditure on certain products (e.g. insurance and
gambling) is treated in the same way as required for the CPI.
However, there are some drawbacks to using the national accounts data including the
limited level of product detail, some data sources may include business and non profi
expenditure which is difficult to remove, and estimates are frequently revised as new data
becomes available.
Increasingly, more countries are using HFCE data to update the CPI weights. The European
Harmonised index of consumer prices (HICP) requires that countries update their weights on
an annual basis and are derived from national accounts. The HICP is deeply rooted in the
national accounts with all “concepts, definitions and conventions adopted in the HICP
...consistent with those used in the European System of Accounts” (Eurostat 2018). A similar
practice is applied in the United Kingdom’s CPIH (ONS 2017).
Since 2018, the Australian CPI weights have been updated annually using HFCE data (ABS
2017). Here the growth rates in the matched elementary aggregates are used to update the
proportions of the existing CPI weights. For lower-level geographic estimates and indices for
special populations, it is assumed that their change in expenditure is the same as the total
aggregate.
The Australian initiative is pertinent to South Africa due to the long time period (six years) in
between household expenditure surveys, especially when compared to some of the
European countries and the UK which have annual or biannual surveys.
In Brazil, the Institute of Geography and Statistics (IBGE) has done substantial research
work to identify possibilities and constraints of using national accounts data. They have not
yet implemented this approach and there is no timeline to do so. Similar to the ABS, they
intend to maintain the existing lower-level proportions of the CPI and to apply the national
growth rates reported in the GDP (Ventura 2019).
3.2 National accounts – local developments
Prior to 2016, the HFCE and other GDP by expenditure accounts, were compiled by the
South African Reserve Bank (SARB). Following a lengthy capacity building and transition
process, Stats SA began publishing these accounts from the first quarter of 2016. The
changeover in compilation responsibility was accompanied by a major review and update of
the source data and methodologies used for compiling the accounts (Stats SA 2016).
In 2021, Stats SA published benchmarked and rebased national accounts. This led to a 16%
upward revision in the size of household consumption in the 2015 base year compared to
9,2% for the economy as a whole. This change resulted from significant improvements in
coverage, source data, and classification (Stats SA 2021). Importantly, HFCE was based on
the 2014/15 LCS, meaning CPI and HFCE structures shared a common data framework.
Improvements were made in the measurement of owner occupied housing, gambling and
insurance services. In some of these cases, methods previously used in developing CPI
weights were employed.
4
There is no perfect fit between the data available in the national accounts and that required
for the CPI weights. While a common classification is used, the CPI weights are determined
with finer granularity than the national accounts. The 2016 CPI basket comprised 404
products (at a national level) while the HFCE reports expenditure for 79 categories at the 4-
digit level and 15 where the 3-digit is the lowest level of detail. The national accounts report
for the whole territory of South Africa whereas the CPI publishes provincial indices and is
weighted at lower levels as explained in Section 2.2.
Table 2. Shares (percentage) of consumer expenditure – 2016
COICOP HFCE
CPI total
country
Food and non-alcoholic beverages 14,6 19,1
Alcoholic beverages and tobacco 4,8 6,0
Clothing and footwear 5,2 4,1
Housing and utilities 15,1 22,5
Household contents and equipment 5,8 4,3
Health 6,9 1,3
Transport 15,7 14,7
Communication 4,2 2,6
Recreation and culture 7,2 4,8
Education 3,2 2,3
Restaurants and hotels 5,1 3,4
Miscellaneous goods and services 12,2 14,9
Table 2 shows the difference in expenditure shares between the 2016 CPI weights and the
equivalent HFCE. Despite the methodological improvements in the national accounts, there
are large differences with the CPI weights. Significant differences also exist at lower levels of
detail. Simply adopting the HFCE structure for the CPI was not feasible as it would would
have disrupted the time series.
The benchmarked HFCE was rooted in the 2015 reference year, and a nominal time-series
is available for subsequent years. The 2019 nominal values are therefore updated with a
more limited set of data than the reference period.
3.3 Other data sources
Stats SA has previously made use of alternative data sources in order to supplement the
household expenditure survey (Stats SA 2017).
Extensive use of business survey data assisted with previous basket updates including large
sample surveys (LSS) of the retail, accommodation and food and beverages sectors.
However, without the detailed HES data, it proved difficult to link the more aggregated LSS
results and the CPI basket. The LSS surveys the formal sector businesses and so excludes
sales in the informal sector – however this is believed to be a small percentage of overall
sales. The different classifications used – LSS uses Central Product Classification (CPC) –
is a further obstacle.
For selected product groups, Stats SA has also previously made use of administrative
records such as tax revenues on alcohol and tobacco, and motor vehicle registration data for
5
purchase of vehicles. These data have proven most useful in obtaining a proper level of
expenditure as these categories tend to be under reported in the expenditure survey.
However, these are already used to compile the HFCE estimates. Some other administrative
data and industry reports were sourced to verify the national accounts information.
Summarised point of sale data was provided, in the past and for this exercise, by a number
of retail chains covering all product categories where expenditure is primarily through retail
outlets. Millions of rows of data contain annual sales values for each unique stock keeping
unit (SKU). This data required classification into COICOP, which was achieved by coding the
product descriptions. This information was previously used for detailed level basket selection
and calculation of weights.
Importantly, all these complementary data sources have previously been used in conjunction
with HES data.
3.4 Weights reference period
COVID-19 had a devastating economic impact, starting in 2020 and continuing into 2021.
Overall, the South African economy only reached pre-COVID levels in the first quarter of
2022 (Stats SA, 2022). COVID-19 forced changes to consumer spending patterns with big
shifts in restaurants and hotels, clothing and footwear, and alcohol and tobacco (see Table
3).
Table 3: Change in HFCE 2019 and 2020
2 Digit COICOP % change
2019-2020
Food and non-Alcoholic beverages 4.0
Alcoholic beverages, tobacco and narcotics -13.8
Clothing and footwear -20.3
Housing, water, electricity, gas and other fuels 5.7
Furnishings, household equipment, maintenance -3.0
Health -1.4
Transport -9.9
Communication 5.1
Recreation and culture -13.3
Education 7.5
Restaurants and hotels -40.0
Miscellaneous goods and services 5.8
Total Expenditure -3.9
Source: Stats SA 2023
When planning the weights update during 2020 and 2021, it was not known how long the
pandemic would continue. As the weights remain fixed for a number of years into the future,
the CPI manual guides that the weights should reflect a ‘normal’ consumption period and
that weights that are not likely to change much in the future should be adopted.
In the past, Stats SA has used growth factors sourced from specific short-term industry
surveys as well as price updating to adjust the weights to refer to the same period as the
index reference month – that being the December before implementation. The revised CPI
6
manual advises that price updating may exaggerate the upward substitution bias inherent in
a Laspeyres-type index and cautions against its use.
The year of 2019 was selected as the weights reference period in order to satisfy the
considerations outlined in the preceding paragraphs. However, this meant that there was
actually only a two year period between the current weights reference period (December
2016) and the new one (2019).
3.5 Calculating the weights
As a result of the differences in the expenditure shares between the CPI basket and the
national accounts data, the changes to the weights were derived from the HFCE growth
rates between annual estimates for 2016 and 2017 (providing a proxy for December 2016)
and 2019.
This method assumes the structural validity of the original CPI weights. It also retains the
existing proportions for elementary aggregates, regional baskets and special aggregations
for population groups.
Table 4: Changes in HFCE and CPI weights 2016/17-2019
COICOP HFCE
growth
% change to
CPI weights
values
TOTAL 15,2 15,0
Food and non-alcohol beverages 14,5 14,6
Alcohol beverages, tobacco and narcotics 21,8 22,1
Clothing and footwear 9,1 8,6
Housing, water, electricity, gas and other fuels 14,7 15,0
Furnishings, household equipment, maintenance 15,2 16,3
Health 18,2 17,8
Transport 13,2 15,1
Communication 6,1 6,1
Recreation and culture 13,2 16,7
Education 19,8 19,8
Restaurants and hotels 17,6 17,7
Miscellaneous goods and services 19,3 13,4
As seen in Table 4, HFCE growth and high-level changes in the CPI weights are not a direct
match. HFCE growth rates were applied to the lowest detail available (3 or 4-digit) from
HFCE to the 8-digit CPI values in a bottom up approach. The new 8-digit estimates were
summed up to obtain aggregates at next highest level. Below (Table 5) is an example of the
bottom up approach, using transport services as an example. Here the growth rate obtained
from the national accounts at the 4-digit level (passenger transport by road) is 18,6%. This
factor is applied to the expenditure values for the lower level detail. The total of R68,129
billion is the sum of the 8-digit expenditure values.
7
Table 5. Applying 4 digit national accounts growth rates to 8-digit CPI values (R’000)
COCIOP 2016 CPI
expenditure
HFCE
growth
rate
2019 CPI expenditure
2 digit Transport
3 digit Public transport
4 digit Passenger transport by road 57 422 923 18,6% 68 129 673
8 digit Local bus fares 48 806 307 57 906 453
8 digit Car rental 6 191 030 7 345 374
8 digit Minibus taxi fares 1 553 499 1 843 155
8 digit Long distance bus fares 872 087 1 034 691
Table 6 shows the changes to the headline CPI weights at a 2-digit COICOP level. The
changes are all less than one percentage point. The category with the largest increase is
alcoholic beverages and tobacco and with the biggest decline is miscellaneous goods and
services.
Table 6. CPI headline weights 2016 and 2019
Category 2016 2019 Difference
Food and non-alcoholic beverages 17,2 17,1 -0,1
Alcoholic beverages and tobacco 5,8 6,3 0,5
Clothing and footwear 3,8 3,7 -0,1
Housing and utilities 24,6 24,5 -0,1
Household contents 4,4 4,4 0,0
Health 1,4 1,4 0,0
Transport 14,3 14,4 0,1
Communication 2,6 2,4 -0,2
Recreation and culture 5,2 5,2 0,0
Education 2,5 2,6 0,1
Restaurants and hotels 3,1 3,3 0,2
Miscellaneous goods and services 15,1 14,8 -0,3
Total 100 100
3.6 Updating the basket
The inherent disadvantage of a Laspeyeres-type price index is that it does not account for
changing consumer behaviour, and cannot accommodate new products. As a result a key
component of updating the weights must be to review and update the basket of goods and
services.
Point of sale data sourced from retail chains provided the basis for making changes to the
CPI basket. After classifying and aggregating the sales information into discrete elementary
level product groups (indicator products), the top 80% of products are considered for
selection. Any new products within the cut-off threshold that were not part of the existing
basket are considered for inclusion.
In cases where a new 8-digit product was added to an existing 4 or 5-digit COICOP group,
the expenditure share of the higher level aggregate was kept constant and new products
were incorporated by distributing expenditure among the selected items.
8
In some cases a new 4 or 5-digit COICOP group was created. In these cases, the 3-digit
COICOP aggregate expenditure was fixed and new expenditure was added.
Based on the availability of products experienced during data collection, some products were
grouped together or split to improve homogeneity, sample size and collection rates.
The number of products in the basket increased to 415 from 404 in the 2016 basket. A total
of 14 products entered the basket and two were removed. One product was split into two; in
two cases, two products were combined into one; and one product was renamed.
A major limitation was that this update was only possible for goods sold through retail
channels. No low-level data on services was available. However, industry reports did not
indicate significant increases in expenditure on new services products.
3.7 Communicating with stakeholders
Stats SA had widely publicised the lack of funding for the HES and the associated risk that
South Africa’s CPI may become non-compliant with international best practice. The methods
used to update the basket and weights without the HES represented a significant change in
methodology for the CPI. Good practice and Stats SA standards require timely and
transparent communication to users of the changes.
A key feature of the communication strategy was the establishment of an advisory
committee consisting of representatives of the SARB, National Treasury, economists from
several commercial banks and economic research institutions. The aim of the committee
was twofold. Firstly to obtain inputs and advice from expert users. Secondly and perhaps
more importantly, it was to obtain the buy in and understanding of key industry role players.
The committee met twice, first to discuss the proposed methods, and then to share the
preliminary results.
Separate meetings were held with a broader group of SARB and National Treasury officials,
as well as a subcommittee of the Statistics Council prior to the release of the results.
Documents were placed on the Stats SA website together with a note in the monthly CPI
release to advise users of the proposed changes. At the time of releasing the new basket
and weights, Stats SA held a public briefing and conducted numerous media interviews.
4. Looking forward
Stats SA eventually received funding for an Income and Expenditure Survey which kicked off
field work in November 2022. This is a permanent budget allocation which should yield
expenditure estimates every three years. Stats SA plans to introduce an update to the CPI
basket and weights in January 2025.
The experience of using the national accounts and retailer POS data has pointed to lessons
to be considered in the run up to this next update.
• Working with the national accounts data has improved the understanding amongst
the price statistics team of how the national accounts are compiled. Much of the
adjustment work that was done in previous reweighting exercises (e.g. to raise the
9
share of alcohol and tobacco) is now done by the national accountants. Using this
work will improve the alignment of national accounts and CPI expenditure shares.
• Stats SA will investigate using the national accounts data as the basis for higher-level
weights and updating these on an annual basis. This is already the practice in the
producer price index and the experienced gained on the CPI illustrated its feasibility.
• Classifying and aggregating the retail POS data is time consuming and technically
demanding. Obtaining and processing the data each year, even if the results are not
applied to the CPI annually, will enable the development of capability and experience
within the price statistics CPI team. This experience can be applied to other projects
such as research into the use of scanner data for CPI compilation. The development
of a time series will allow proactive analysis of changes in consumer behaviour to
guide basket updates. It is possible that this information may assist in reducing
respondent burden for large sample business surveys.
• This experience reinforced that there is no proper replacement for a household
expenditure survey. Stats SA did not obtain data to update the services components
of the basket. Consumer preferences for services may change more rapidly than
goods and this basket update missed any of these shifts. The fact that the HES
provides detailed information on all COICOP groups is essential for creating a
balanced set of weights and a basket that reflects all the shifts in consumer
spending.
10
References
ABS, 2017. An implementation plan to annually reweight the Australian CPI.
https://www.abs.gov.au/statistics/research/implementation-plan-annually-re-weight-
australian-cpi
Eurostat, 2018. Harmonised Index of Consumer Prices (HICP) Methodological manual.
https://ec.europa.eu/eurostat/web/products-manuals-and-guidelines/-/ks-gq-17-015
IMF, 2020. Consumer Price Index Manual: Concepts and Methods.
https://www.imf.org/en/Data/Statistics/cpi-manual
ONS, 2017. Investigating the impact of different weighting methods on CPIH.
https://www.ons.gov.uk/economy/inflationandpriceindices/methodologies/investigatingtheimp
actofdifferentweightingmethodsoncpih
Stats SA, 2016. Expenditure on GDP: Information Note.
https://www.statssa.gov.za/publications/P0441/Information_note_on_Expenditure_on_GDP.
pdf
Stats SA, 2017. Introduction of new weights and basket for the Consumer Price Index.
https://www.statssa.gov.za/cpi/documents/Introduction_of_2016_CPI_weights_and_basket.p
df
Stats SA, 2021. National Accounts: Source and methods. Report 04.04.04.
https://www.statssa.gov.za/publications/Report%2004-04-04/Report%2004-04-042021.pdf
Stats SA, 2022. Economic recovery from COVID-19: Not all countries are equal.
https://www.statssa.gov.za/?p=15690
Stats SA, 2023. GDP time series. https://www.statssa.gov.za/?page_id=1854&PPN=P0441
Ventura, 2019. Challenges of Using National Accounts for a More Frequent Update of the
Brazilian CPI Weights. Paper presented at the 16th Meeting of the Ottawa Group on Price
Indices. https://www.ottawagroup.org/ottawa/ottawagroup.nsf/home/Meeting+16+-
+Rio+de+Janeiro,+2019