SUMMARY AND FUTURE
6.1 The ABS has produced official statistics for Australia for over 100 years to inform decisions on matters of importance. However, the environment in which NSOs operate is changing. Dynamic information is now available, creating new opportunities, and also challenges, for statistical offices. To capitalise on these opportunities, and better meet current and future statistical requirements of the Australian community, the ABS is undertaking a research program to enhance the Australian CPI. This publication focuses on one component of that program, making greater use of transactions data to compile the Australian CPI.
6.2 The Australian CPI currently uses transactions data by calculating an average unit value by product, by taking the quantity and expenditure information over the period of interest and replacing the directly observed price with this unit value. While this has enhanced the Australian CPI, it is recognised that more can be done with big data.
6.3 The availability of timely expenditure data in transactions datasets allows weighted bilateral indexes to be calculated, accounting for consumer substitution. However traditional bilateral methods break down when using transactions data. New methods and processes are required. Temporal multilateral methods have been proposed as they preserve transitivity and make greater use of price and expenditure information.
6.4 Multilateral methods allow NSOs to use the dynamic universe of transactions data to enhance the accuracy of their price indexes. They make greater use of automated processes, providing NSOs with an opportunity to reduce costs across the statistical cycle. Automated processes also provide an opportunity for more timely output - less manual intervention facilitates completion sooner after the reference period, rendering higher frequency price indexes more feasible.
6.5 While having strong theoretical and practical properties, multilateral methods also present NSOs with challenges. Recalculating the index as new periods become available revises the historical series, which is unacceptable for the Australian CPI. To overcome this challenge each new multilateral index must be spliced on to the previously published index levels. Extending the time series in this way impairs transitivity, however this is less severe when using multilateral methods than bilateral methods.
6.6 Another challenge lies in the fact that multilateral methods use prices and expenditure weights for several periods to calculate a transitive index, having potential to affect the relevance of the index. The choice to find a suitable balance for this depends more on the choice of window length and extension method than the multilateral method itself. ABS research has determined that if there is sufficient data available then the estimation window should be a reasonable length, and recommends a 25 month, or 9 quarter window.
6.7 Over the years a range of multilateral methods have been proposed for use in CPI aggregation when using transactions data, while only a handful of NSOs have actually implemented them in their CPI, each using different methods and practices. This stems from a lack of consensus on the best method, as well as the circumstances of each NSO.
6.8 This publication presents a selection of well-known matched-model multilateral methods for producing temporal indexes from transactions data - TPD, GK, QAUV_TPD and GEKS - as well as four different methods for extending the index to create a continuous non-revised time series. With no consensus amongst researchers or leading NSOs on the best method, the ABS used a framework to help guide the choice of method. This framework was assessed empirically.
6.9 The ABS DQF provides an opportunity to guide the choice of multilateral method by weighting the level of statistical quality at a broad level. Doing so revealed all methods facilitate a better use of resources, as discussed above. From a theoretical perspective, no method satisfied all of the Tests proposed in the literature. The GEKS method has the best established economic properties, but the ABS has not found evidence of substitution bias for the other methods in this context. All methods were found to be adaptable for different data sources, and although they are more complex than bilateral indexes, none is prohibitively difficult to explain.
6.10 Empirical analysis reveals the choice of both multilateral and extension method impacts the estimated price index series. In general, the multilateral methods produced similar trends over time, with short-term departures primarily due to the different use of expenditure shares to weight individual products. Analysis of the extension methods lended support for the HS method, while the direct extension method appeared to be impacted by the choice of link month for certain commodities. Findings also reinforce traditional ABS practices, as aggregate results for the multilateral methods follow similar trends to the published CPI. Where there are divergences they are driven by multilateral methods using contemporary data for weighting purposes, more accurately reflecting changing consumer preferences.
6.11 This publication demonstrates support for the use of multilateral methods as the most opportunistic way to make greater use of transactions data to enhance the CPI. An assessment of the theoretical and practical aspects of the four multilateral methods presented in this publication reveals they are all suitable for use in a CPI context. Whilst having many benefits, multilateral methods present NSOs with challenges which must be considered in a local context. The ABS will continue to assess aspects of multilateral methods as outlined in this publication. The ABS will also consult with peers and experts in order to develop a best practice approach.
6.12 User and stakeholder input is sought to contribute to the resolution of outstanding methodological challenges. A final ABS publication will be released on this topic in mid-2017, which will articulate the precise methods to be implemented in the CPI. These methods will be finalised following extensive consultation with international experts, academics and users.
6.13 For further information relating to making greater use of transactions data in the CPI, readers should write to:
Mr Andrew Tomadini
Consumer Price Index Section
Australian Bureau of Statistics
PO Box 10
Belconnen ACT 2617