CHAPTER 6 WEIGHTS AND THEIR SOURCES


INTRODUCTION

6.1 This chapter describes the procedures that are typically followed in updating the CPI weights.

6.2 Weighting practices vary at different levels of the CPI. At the published levels, weights are reviewed every five or six years. These reviews follow the release of data from the Household Expenditure Survey (HES). At the unpublished levels, the weights can be varied at any time, and often are. At the elementary aggregate or price-sample level, there is no explicit use of weights.


FIXED WEIGHTS

6.3 At the level of the index at which the weights are fixed, the ideal is to have a snapshot of all household expenditure. It is important that the data are consistent across the population group, that is expenditures can be added up without concerns about coverage and double counting, and that all the information is for the same period. For example, if expenditures are for different periods for different items, then these could be affected by changes in economic conditions, tax rates, and population growth. Any of these influences could easily distort the weights. In addition, although an information source might provide an estimate of total sales of an item, it would be necessary to know the proportion of those sales to households as consumers. For example, sales of whitegoods and browngoods(footnote 1) will include sales that are not in scope of the CPI, for example sales to businesses, and to households as owners of rental properties.

6.4 The HES is the only source of data that comes close to the ideal. The HES is a sample of approximately seven thousand households. Data are collected using a diary of personal expenditures in which residents aged 15 years and older record their expenditure over a two-week period. An interview questionnaire also collects information about each household's characteristics, expenditures common to all members of the household, and irregular or infrequent expenditures. The HES used in the 15th series CPI is for the reference year 2003-04.

6.5 Although the HES provides a comprehensive coverage of household expenditures, there are some weaknesses in the HES data for CPI purposes.

6.6 The HES is a survey, and so is subject to sampling error. It is possible for the selection of one or several households with exceptionally high expenditure on an item to significantly affect the expenditure estimate for that item in a region, especially in the smaller capital cities.

6.7 Some expenditures recorded in the HES are not fully contemporaneous. Estimates for some items which are more expensive or purchased infrequently are obtained or supplemented by recall, rather than relying on expenditures actually recorded in the diary during the two weeks that the household is included in the survey. Periods over which households are asked to recall expenditure vary depending on the item.

6.8 The HES records all data exactly as reported by the households with no adjustments for known cases of under-reporting, especially in expenditure on alcohol and tobacco.

6.9 There are some expenditures required by the CPI which cannot be obtained from households, such as interest margins on deposit and loan facilities.

6.10 Thus, various adjustments need to be made to the information reported in the HES before it can be used in the CPI, and for some items the HES data are supplemented or replaced by other data which are thought to be more accurate.


Adjustments for under reporting

6.11 There are data sources other than the HES that provide estimates of households expenditure on some commodities, but often only nationally. These alternative sources are used for validating the HES data. For most products included in the CPI, these alternative data are similar to the HES estimates. However, historically there have been significant differences between HES and the other data sources for households' expenditure on alcohol and tobacco. For both products, households are the major consumers, and the imposition of excise and other taxes on these products means that the alternative estimates are more accurate than the HES estimates which appear to understate household's expenditures.

6.12 For alcohol and tobacco, under-reporting factors were derived nationally using estimates of household final consumption expenditure from the Australian National Accounts. These factors were applied to the HES expenditure estimates for each capital city. The 2003-04 HES has expenditure on tobacco at a little under half, and expenditure on alcohol at a little over half, of the national accounts estimates. The same adjustment factor was applied across all expenditure classes for alcoholic drinks in each capital city.

6.13 The intention of the CPI classification is that all expenditure on alcohol should be in the subgroup Alcoholic drinks. As noted earlier, households do not always separately identify expenditures as fully as we would like. With restaurant meals, expenditure on alcohol is not always reported separately, and is sometimes reported simply as drinks. A sample of household records from the HES was examined, and an estimate was made of the proportion of reported expenditure on restaurant meals that was probably expenditure on alcohol. Expenditures on meals were then adjusted by applying the factor across all capital cities.


Recall adjustment

6.14 Some expenditures are collected in the HES as recalled items rather than as diary entries. The extent of the recall period varies. For some items, it is purchases in the last three months (most whitegoods and furniture, and repairs to houses); for general rates, electricity, and health services it is the last payment; for most vehicle-related expenditures, education, overseas travel, and alterations and additions to houses, the recall period is twelve months; and for house purchases it is the last three years. To the extent that prices for these items change between the time that the household last purchased them and their inclusion in the HES, expenditures will not accurately reflect the underlying quantities acquired during 2003-04. Accordingly, recall factors may be calculated to adjust the reported data for price changes.

6.15 No adjustments are made for any items where the recall period is the last three months since the adjustments would be small. Most items where the recall period is the last payment are typically those with a short billing cycle, often quarterly, or where there are options to pay periodically (e.g. local government rates). This leaves only a few items such as motor vehicle purchase, general insurance, overseas holiday travel, and education where expenditures could have been incurred up to twelve months before inclusion in the HES. Adjustments are made for these items.

6.16 The adjustment procedure for a recall period of twelve months is as follows. In the first quarter (Q1) of the HES reference year, the period potentially covered will be from the start of the corresponding quarter of the previous year (for those households selected in the first week of the HES reference year) through to the end of the first quarter of the reference year. Similarly, the pricing period for the households first included in the last week of Q1 in the reference year will commence in the last week of Q1 of the previous year. If expenditures are distributed uniformly over the period, then effectively expenditures in the reference quarter and the corresponding quarter of the previous year will only be half that for the intervening quarters. Thus, assigning weights of, say, one for the intervening quarters and 0.5 for the start and end quarters, we can derive a weighting pattern as shown below.

6.17 Thus the adjustment factor for items with a twelve-month recall for the 2003-04 HES is:



where I_S03 is the CPI index number for the expenditure class for the September quarter 2003 etc.

Using the following hypothetical index numbers:

;

The result of the above formula is:
=1.02

(which is the adjustment factor to be applied to the recalled price).

6.18 With a generally low rate of price change over 2002-03 and 2003-04, the adjustments made for recall were small. However, quite significant adjustments have been made in the past (e.g. for the 1984-85 HES).

6.19 In the 15th series CPI, a quantity adjustment was also applied to overseas holiday travel as the recall period included 2002-03 when the number of Australians travelling overseas fell temporarily. The adjustment factors were calculated using the same technique used in the price adjustment, with the number of short-term overseas departures replacing the price index in the formula above.


Salary sacrifice

6.20 Salary sacrifice is an arrangement between an employee and their employer whereby part of the employee's pre-tax cash salary is traded for non-cash benefits. Conceptually, these payments are part of both gross wages and salaries and household expenditure. Expenditures funded through salary sacrifice arrangements are generally not captured in the HES. Amounts that were reported to be salary sacrificed on the income side were used to adjust the average weekly household expenditures. In the 15th series, the items that have been adjusted for salary sacrifice are telecommunications, child care, computers, and motor vehicles. Salary sacrifice amounts reported against motor vehicles were allocated to purchases of vehicles, registration, insurance, repair and servicing, and automotive fuel.


Aberrant expenditures

6.21 Expenditure estimates for the CPI weights can be validated in time and spatial dimensions. Validation over time requires expenditure to be on a common pricing base (e.g. revaluing 1998-99 HES expenditures to 2003-04 prices, and comparing with the 2003-04 HES results will show changes in volume terms). Any large differences can then be investigated to see if they are valid. For example, there was a large rise in expenditures on telecommunications (both in nominal and constant prices) between the 1998-99 HES and 2003-04 HES, but this is broadly consistent with the growth in telephony services.

6.22 A few outlier adjustments were made in the smaller cities where sample sizes are smaller and, in general, the standard errors larger. In the 15th series CPI review, outliers were adjusted using the winsorisation technique of replacing the outlier expenditure with the expenditure of the next highest respondent. Where outliers could not be identified, differences were investigated to see if they were valid. A few volume changes could not be validated, resulting in adjustments using either alternative volume data or judgement.


Expenditures not sourced from HES

6.23 For the 15th series CPI, it was not possible to obtain suitable data from the HES for four items: purchases of new houses, financial services, general insurance services, and tertiary education.

New house purchases

6.24 The weight given to house purchase should reflect net acquisitions by households, and expansion of the housing volume by alterations and additions. Sales of houses that take place between households are excluded, so that the weights relate only to net additions to the housing stock arising from household purchases from other sectors (e.g. from businesses such as builders and developers).

6.25 An additional factor that arises with houses is that as well as providing shelter to their owners, they include an investment element because they typically appreciate over time. For the purposes of the CPI, the cost of housing should reflect only the shelter component. A common approach is to regard the cost of the land as the investment component, and the cost of the structure as the shelter (or consumption) component.

6.26 The net expenditure on house acquisition was estimated by applying the average value of private dwelling completions by capital city for 2003-04 published in Building Activity Australia (cat. no. 8752.0) to the net change in the number of owner-occupier households. The average value estimates for completions only include the cost of construction, and therefore enable the land component in house prices to be excluded from the weighting base. The net change in the number of owner-occupier households was derived by calculating the average annual net change in owner-occupier dwellings between the 1996 Census and the 2001 Census. This revised method of calculating the net change in the number of owner-occupier households was considered to be more accurate than the use of projected growth rates which was used in the 14th series CPI. Subsidies paid to first home buyers as part of the Commonwealth Government's First Home Owners' Scheme were treated as negative expenditure and subtracted from the expenditures recorded for house acquisition.

6.27 The alterations and additions component of house purchase includes any changes in the physical attributes of owner-occupied houses such as extensions, renovations, addition of insulation, and similar improvements to the structure. For this component, HES data are considered adequate.

Financial services

6.28 HES data are not useful for financial services because household expenditure on interest margins is a statistical construct; and it is too onerous to ask households to calculate their annual expenditures on the many fees and charges that they pay to financial intermediaries. Weights - as well as measures of price change - can only be calculated from information supplied by the financial intermediaries themselves. The weight of the Deposit and loan facilities expenditure class has two broad elements: implicit expenditure on interest margins, and fees and charges.

6.29 Interest margins has by far the biggest weight. To calculate household implicit expenditure on interest margins the ABS needs to know three things: the yields on individual consumer products, a suitable reference rate of interest, and the average balances of those products.

6.30 The sources of the data for these complex calculations are the financial intermediaries' monthly average balance sheet and related interest statements. They prepare these at different levels of consolidation. The ABS prefers to work at the lowest level that is available. This is often the product level. However, sometimes these statements are available for sub-products. The financial intermediaries prepare these statements at these low levels mainly for the purpose of yield management.

6.31 The financial intermediaries work out the monthly average balances by taking the balances at the close of business each day, adding them up for the month, and then dividing by the number of days. Interest is the flow during each of the months. For ease of recognition, the ABS converts the monthly interest flows into annual yields by dividing by the number of days in the month, multiplying by 365 and finally by 100.

6.32 The ABS has found that the data in these statements are sometimes not as smooth as we might expect. Therefore three-monthly moving-averages of both the balance and interest series are used to smooth any accounting anomalies in the data reported to us.

6.33 The reference rate of interest is specific to each bank. We use as the reference rate the mid-point of the average interest rate received on all loans (including business loans) and the average interest rate paid on all deposits (including deposits by businesses). This approach minimises the problem of negative margins encountered by other statistical agencies that have used bill or cash rates as their reference rates of interest.

6.34 The difference between the reference rate and the product yield is the margin rate on the product. Applying the margin rate to the average product balances provides the measure of households' implicit expenditure on interest margins (weight).

6.35 We calculate the weight for fees and charges using fee revenue data from the financial intermediaries. The ABS receives these data by product, and since we know which products are consumer ones, we add up the fees attributable to those products to get the estimate of weight.

6.36 Data about commissions on sales of residential properties and from conveyancing work come from Real Estate Services Australia, 2002-03 (cat. no. 8663.0). This information is used to derive expenditure on real estate agency services. Income data from Legal Practices, Australia 2001-02 (cat. no. 8667.0) are used to derive expenditure on legal and conveyancing services purchased by households. The national expenditures derived from these surveys were revalued to 2003-04 dollars, and the aggregate expenditures reported in the 2003-04 HES were used to allocate the expenditures to capital city. Household final consumption expenditure from the Australian National Accounts was used to derive an estimate of the expenditures on stockbroking services. Expenditures on taxes on transfers were compiled using data supplied by the state and territory revenue offices.

Insurance services

6.37 The weight for general insurance reflects the service provided by insurers (gross premiums less claims) rather than the gross premiums paid by policyholders. Expenditures which are funded by insurance claims are included. However, as it is only possible for the HES to collect gross premiums, these are adjusted by the proportion that represents the insurance service using information obtained from the Australian Prudential Regulation Authority (APRA) and insurance companies. For some items such as motor vehicle write-offs and vehicle smash repairs, the HES estimates exclude expenditure funded from insurance claims. Expenditures funded by claims were added back to these items by calculating the claims relative to premiums using APRA and insurance company data.

6.38 As it is impractical to measure price change in the insurance service charge, price change in gross gross premiums paid by households is used as a proxy measure.

6.39 The weight for medical insurance, on the other hand, is determined by the gross premiums paid by households. Expenditures on individual types of medical services are not priced; rather the gross premiums paid are considered a suitable measure of payments for all medical services covered (including the insurance service component) and no redistribution of expenditure to individual medical services is carried out.

Tertiary education

6.40 Another area where the HES expenditures are not appropriate for use in the CPI weights is expenditure on university education. The 2003-04 HES recorded Higher Education Contribution Scheme (HECS) payments made up-front plus any HECS repayments made through the taxation system during the reference period. This approach is not consistent with an acquisitions-based CPI where expenditures should reflect the cost to households of the educational service enjoyed during the reference period. The CPI scope includes the payments made during the period (up-front payments) plus fees for educational services acquired during the period but deferred to be paid at a later date.

6.41 As in previous series, in the 15th series an alternative measure of expenditures on HECS was calculated using other sources, including data from the Department of Education, Science and Training (DEST), on the total up-front and deferred expenditures, and the number of university students liable for HECS.


Revaluing expenditures to the link period

6.42 The expenditure weights derived from HES are based on expenditures in 2003-04. This new expenditure pattern was not introduced into the CPI until the June quarter 2005 (the link period). As the quantities underlying these expenditures had to be preserved, the expenditures were revalued to June quarter 2005 prices. The ABS does this by multiplying the 2003-04 expenditures by the ratio of its price index for the June quarter 2005 to the average of its quarterly indexes for 2003-04.


Adjustments for quantity shifts

6.43 Ideally, the CPI weights should be as up to date as possible and be broadly representative of the expenditure pattern that might be expected over the life of the index series. Thus, when the June quarter 2005 link was being introduced, it was necessary to consider whether any developments and policy changes since 2003-04 might have significantly affected the expenditure pattern and whether any revalued expenditures needed to be adjusted.

6.44 There were no major policy changes identified during this period that would have significantly changed volumes between 2003-04 and June 2005.

6.45 Other items where expenditures were likely to have changed between 2003-04 and June 2005 were also investigated. These included the purchase of computers and internet usage. Some adjustments were made, although the effect on the weights at the expenditure class level was minimal.

6.46 In principle, adjustments are appropriate for any significant change in expenditures between the time that the HES data are collected and inclusion of the weights into the CPI. However, this raises a methodological issue. The adjustments to expenditures are generally made without compensating adjustments to other expenditures in the CPI basket. In other words it is implicitly assumed that increased expenditure on internet services, for example, comes from savings and not from reductions in expenditure on other items.


LOWER LEVEL WEIGHTS

6.47 The weights and structure of the CPI below the level of expenditure class are continually reviewed and may be varied at any time. These changes are made through a process termed sample review. Essentially, a sample review involves selecting a component of the CPI and examining it closely. The review is generally performed for an expenditure class and determines what changes should be made to the items priced, the outlets they are sourced from, and the weights to be applied to the commodities and outlets within that expenditure class. For example, it might be determined from recent data that it is appropriate to introduce a price sample under the Fats and oils expenditure class for products composed of both butter and margarine, or to change the relative importance of large retail outlets and convenience stores in the price samples for soft drinks.

6.48 Information from any reliable source is used to assess the importance of one commodity relative to another. Sources include data collections of the ABS and other organisations, and publications by industrial organisations. Information from the HES is also considered but, for the main part, is not sufficiently detailed or reliable at the lower levels of the CPI structure. For example, the HES data for types of appliances purchased would not be as reliable as industrial sales data because of the small samples in the HES.

6.49 At the price sample or elementary aggregate level, there are no explicit weights. Rather, the price samples are constructed so they are self-weighting. For example, if there were a price sample for medium chocolate bars, and the major grocery outlets had 80 per cent of these sales and vending machines 20 per cent, then the price sample would be selected so that for every price from a vending machine there are four prices from the major grocery outlets.

1 So–called browngoods are household electrical items such as TVs, DVD players, sound systems, kettles, and toasters. <back