Monthly Household Spending Indicator methodology

The experimental Monthly Household Spending Indicator is derived using banks transactions, supermarket transactions and VFACTS data. As this data is not designed for statistical purposes, its scope varies from Australian National Accounts concept of Household Final Consumption Expenditure (HFCE) and the Retail Trade turnover estimates for retail businesses. Coverage adjustments are subsequently made to align the MHSI with the HFCE concept.

The indicator should be considered experimental at this stage, as further enhancement to the transformation processes and methodology are expected in the future.

HFCE consists of expenditure by Australian resident households on goods and services for own non-commercial use, whether they are offered within the domestic territory or abroad. It is referenced to the period where the value of the good/service is consumed, rather than when it was paid for, in cases where these differ.

The MHSI consumption concept is aligned as closely as possible with the National Accounts' HFCE concept. However, MHSI data sources do not capture spending by Australian residents on goods and services offered outside Australia. The card transactions data received from the banks excludes transactions if they are in a foreign currency and the location of the card terminal is outside Australia.

The bank transactions data is provided to the ABS by participating banks for statistical production purposes. The data is provided to the ABS following the end of each calendar month. Once all data is received it is validated and collated.

The data is received in different formats and classifications from the participating banks. Some banks provide data at a low level of categorisation e.g., at a bank allocated merchant category level, and others provide categorised data at the Australia and New Zealand Standard Industry Classification (ANZSIC) or retail or non-retail level.

Data is received at different frequencies e.g. daily, weekly or monthly. All participating banks provide data at an identifiable state and territory level.

The aggregated bank transaction data does not contain information about individuals or households. Banks transactions data is only accessed by ABS staff required to produce relevant statistics, including the indicator. The ABS is committed to upholding the privacy, confidentiality and security of the information it collects.

The ABS receives weekly transactions data aggregates from selected supermarket chains. Transactions (Scanner) data captures point-of-sale purchases from supermarkets and contains detailed information about transactions, dates, quantities, product descriptions, and values of products sold.

The data is received in different formats and classifications from the participating supermarkets and is standardised before being processed for MHSI estimation. Each supermarket provides item-level sales data by week, at varying levels of geography. 

The weekly aggregates received do not contain information about individuals purchasing the products, and are a business-side view of total product sales.

The ABS receives monthly vehicle sales data from FCAI via the VFACTS datasets. These datasets capture total sales from FCAI-associated dealers (>99% of all dealers) within the reference month and contains detailed information about the make, class and engine type of the vehicles. Sales aggregates are delivered at the state and part-of-state level of geography (e.g. Victoria metro or Victoria rural).

The monthly aggregates received do not contain information about individuals purchasing the products, and are a business-side view of total product sales.

The indicator is derived from aggregated, de-identified bank card transactions data for resident households, as well as supermarket and VFACTS car sale data. The data is provided to the ABS from participating banks, selected supermarkets and the FCAI respectively. 

The source data does not cover the whole population and in the case of bank data, does not include all payment types. While cash payments, direct transfers outside the banks, cryptocurrency and BPAY transactions are all forms of household spending, these are not currently included due to lack of available data sources or ability to identify such transactions in the current data set.

Estimation methods are used to adjust for under coverage of the population, under coverage within each payment type and over coverage of institutional sectors. For more information see Estimation.

Furthermore, the following assumptions have been used in producing the indicator:

1. All transactions represent final consumption expenditure for own non-commercial use. The data includes household transactions that may feed into the production process (e.g., a household purchase of seeds to grow vegetables at home). As it is not possible to identify or adjust for these transactions, it is assumed any ‘intermediate consumption’ transactions in the data have an immaterial impact on the final estimates.
2. Non-resident transactions (net expenditure overseas) captured within the bank transaction data have an immaterial impact on the estimates. Lacking any means to identify and adjust non-resident transactions, it is assumed that expenditure patterns between resident and non-residents exhibit similar trends on consumption goods and services.

Outputs are produced for nine out of the thirteen major Classification of Individual Consumption According to Purpose (COICOP) Divisions. Outputs are produced at the Australia, and state and territory level. For more information see Outputs.

In addition to normal editing procedures, a number of data transformation steps and adjustments are made to derive the Monthly Household Spending Indicator. These include:

• Classification and concordance mapping
• Apportioning industry sales totals to individual consumer products
• Estimation, including coverage adjustments and benchmarking
• Seasonal adjustment and Calendar adjusted estimates

The bank data has varying classifications. To produce aggregate COICOP outputs, a concordance structure was developed to map the transactions data from each participating bank. This maps bank data, by merchant category or ANZSIC class, to the COICOP Division classification and creates a consistent data set for further transformation. 

For retail industry categories, bank data is mapped to the COICOP output classification using a "industry-product matrix" derived from the 2012/13 Retail and Wholesale Industries (RISWIS) survey. This matrix allows industry sales to be apportioned to individual products by the use of survey-derived weights. 

As VFACTS and supermarket transactions data already contains product data, no separate industry-product mapping is required to produce COICOP outputs. The supermarket transactions data is used to map bank card spending at supermarkets to the Food or Cigarettes and tobacco categories.

There are potential areas of both under and over coverage. For example, not all Australian resident households are represented within the supplied bank data resulting in under coverage of the population. Additionally, not all payment types are captured in the bank data, such as cash payments. There is also potential over coverage as individuals may use their personal bank cards for business-related purchases. Instances of over or under coverage can introduce bias and variance in the estimates. 

To rectify coverage issues, data-driven adjustments are made to the estimates. An adjustment is made to correct for under-coverage of people who are not customers of data providing banks, using published RBA statistics totalling all card spending across all financial institutions as a "control level". An additional adjustment is made to correct cash spending under-coverage, by combining ATM withdrawal data with RBA survey insights into household payment behaviour from different types of retailers. 

An additional layer of quality assurance includes benchmarking to annual state-level HFCE, which is done for the primary reasons of consistency with published National Accounts' HFCE figures and to ensure quality by using the annual HFCE as a quality control total. MHSI estimates are currently benchmarked up to 2021/22 HFCE.

When the next annual benchmark becomes available, the growth rates will be adjusted to meet the benchmark series. This will result in revisions. Revisions from the initial to the benchmarked growth rates can be used to provide an indication of the accuracy of the initial estimates of the MHSI in terms of size (variance) and direction (bias).

The indicator exhibits systematic seasonal and calendar related effects. Producing seasonally adjusted series improves the quality of the output in these ways:

• Interpretability – predictable seasonal effects will be removed to better identify short-term and long-term behavioural trends
• Relevance – increased suite of statistics available for users
• Coherence – comparability with existing outputs including retail trade and HFCE

MHSI produces seasonally adjusted series for national total, total goods, total services, total discretionary and total non-discretionary consumption. Individual COICOPs and state level series are at this point not seasonally adjusted, but these will be added in the future and replace the calendar adjusted series.

Monthly and quarterly series are produced for total household spending and nine of the major COICOP Divisions:

• Food
• Alcoholic beverages and tobacco
• Clothing and footwear
• Furnishing and household equipment
• Health
• Transport
• Recreation and culture
• Hotels, cafes and restaurants
• Miscellaneous goods and services (excluding Insurance and Other Financial Services)

Published outputs also include analytical series such as total consumption of goods/services, and discretionary/non-discretionary products. The table below shows how the COICOP categories are mapped to these particular groups

Current price estimates reflect both price and volume changes, and are published on a monthly basis. However, chain volume estimates, which are published on a quarterly basis (i.e. March, June, September and December publications), measure changes in value after the direct effects of price changes have been eliminated, and hence only reflect volume changes. The chain volume measures of the MHSI are annually reweighted chain Laspeyres indexes referenced to current price values in the chosen reference year. The indexes are linked using the one-quarter overlap method, and the reference year is currently 2021-22. Each financial year's quarterly data in the MHSI chain volume series are based on the prices of the previous financial year, except for those quarters of the latest incomplete or just completed (i.e. in June quarter) financial year which are based upon the prices of the reference year. Comparability with previous years is achieved by linking (or chaining) the series together to form a continuous time series. 

With each release of the September quarter issue of the MHSI publication, a new base year is introduced, and the reference year is advanced one year. This means that with the release of the September quarter 2024 issue of the MHSI publication, the chain volume measures for 2023-24 will have 2022-23 (the previous financial year) as their base year rather than 2021-22, and the reference year will be 2022-23. A change in the reference year changes levels but not growth rates for all periods. A change in the base year can result in revisions to growth rates for the last financial year.

Chain volume measures are not generally additive. In other words, component chain volume measures do not, in general, sum to a total in the way original current price components do. This means, for example, that the chain volume estimates for COICOP Divisions will not add to the total for Australia. In order to minimise the impact of this, the ABS uses the latest base year as the reference year. By adopting this approach, additivity does exist for the quarters following the reference year and non-additivity is relatively small for the quarters in the reference year and those immediately preceding it. Further information on the nature and concepts of chain volume measures is contained in the ABS publication Information Paper: Introduction of Chain Volume Measures in the Australian National Accounts (cat. no. 5248.0).

A deflator seeks to measure price change between two time periods on a like-for-like product basis. A deflator is often a price index, and is used to compile chain volume estimates.

Price indexes sourced from the Consumer Price Index (CPI) are mapped to relevant products at the Input-Output Product Classification (IOPC) level. These price indexes are aggregated to COICOP categories by using current price expenditure data from the Input-Output tables as weights to form MHSI deflators.

MHSI and HFCE deflators comparison

MHSI and HFCE deflators use the same source data – predominantly CPI (for COICOPs in scope for both estimates). However, there are two notable differences in methodology and scope:

• Methodology: MHSI deflators are created by mapping price indices at the IOPC level, while HFCE deflators are mapped at varying COICOP levels.
  • The IOPC system is a lower and more detailed classification system, and multiple products constitute each COICOP division.
• Scope: MHSI deflators exclude expenditure by Australian residents overseas (imports/debits) because it is assumed to have an immaterial impact on MHSI estimates (due to the coverage of MHSI). However, this spending is in scope for HFCE deflators.

Legislative requirements to ensure privacy and secrecy of this data have been adhered to. The source data for MHSI has been aggregated and de-identified by the data providers. Therefore, in accordance with the Census and Statistics Act 1905, the outputs are unlikely to enable the identification of a particular person or organisation.

All personal information is handled in accordance with the Australia Privacy Principles contained in the Privacy Act 1988. For more information, see ABS privacy.

The data provided by the participating banks does not include all domestic resident households nor does it capture all household spending across the COICOP Divisions. Bank card transactions are an appropriate and preferred mode of payment for most of the published COICOP Divisions, however, it does not represent all spending due to the absence of cash-based spending and other payment modes. Coverage adjustments are made to improve coverage and fill data gaps arising from transactions for non-participating financial institutions, and cash spending.

Due to the varying payment modes, high concentration of non-card bank transaction payments for some COICOP Divisions, and lack of available monthly data sources, a household spending indicator could not be produced for all major COICOP Divisions.

The appropriateness of using available data sources to measure household spending across each COICOP Division was assessed, this resulted in some Divisions being excluded from the Monthly Household Spending Indicator.

COICOPs that are currently excluded from the indicator are:

• Rent and other dwelling services
• Electricity, gas and other fuels
• Communication Services
• Education Services
• Insurance and other financial services

Revisions are a change in the values of published data and may arise due to a variety of reasons.

Revisions may be applied to the Monthly Household Spending Indicator due to:

• Revisions to the source data
• Refinements to decisions made around the treatment of data anomalies in the series
• Implementation of methodological and process improvements
• Application of the annual benchmark
• Revisions as a standard feature of seasonal adjustment and trend estimation
• Application of a new chain volume reference year

The methods and data sources used will be subject to ongoing review to improve outputs and maintain the relevance of this indicator.

Indicator estimates are produced in: 

• current price original terms 
• chain volume original terms
• current price seasonally adjusted terms for selected series, and in current price calendar adjusted terms for the remaining series.

Calendar adjusted estimates account for trading day impacts and length of month, and seasonally adjusted estimates removes the time-of-year related regular variation in consumption, so that movements represent meaningful deviations from seasonal patterns.

Published dollar levels are converted to millions and rounded to one decimal place. Published percentage changes are calculated on the underlying dollar levels and rounded to one decimal place. Any discrepancies between percentage changes published and percentage changes derived from published dollar levels are due to rounding.