Introduction -
At present, the deflators used for computers in the Australian System of National Accounts are based on hedonic indexes compiled by the U.S. Bureau of Economic Analysis [BEA]. During the past year, the Australian Bureau of Statistics [ABS] has experimented with estimating hedonic functions and compiling price indexes based on Australian data. Although some investigations remain to be done, the work is showing promise.
Current Practice -
The ABS, like many other national statistical agencies, uses the U.S. computer price index for deflation purposes. The ABS lags the U.S. index by three months (to take account of delays in shipping new models to Australia and other delays in price transmission) and adjusts it for changes in $A/$US exchange rate. This has been ABS practice for the past thirteen years — it is appealing because of its simplicity, and because Australia imports most of its computers and the worldwide market (for personal computers, especially) is very competitive.
-
The question arises, however, whether this practice reflects conditions in the Australian computer market. There is some evidence, for example, that (at least in the short to medium term) Australian prices do not track U.S. prices exactly — wholesale and retail margins appear sometimes to fluctuate to maintain a $A price, even when U.S. prices or exchange rates display large changes. Also, large proportions of PCs and computer components are imported from Asia rather than from the U.S. (see Table 1), so the U.S. price and the $A/$US exchange rate may not be the sole or most direct influences on local prices.
Table 1: Sources of Computer Imports into Australia, 1999-2000
-
Source country1
|
Percentage of total imports
|
Singapore
|
24
|
Malaysia
|
20
|
USA
|
17
|
Taiwan
|
14
|
Japan
|
12
|
Other countries
|
13
|
TOTAL
|
100
|
(1) Imports recorded against some countries (especially Singapore) may have been ultimately sourced from the USA or other countries.
Estimating an Australian Hedonic Function -
The ABS has estimated hedonic functions for desktop and notebook personal computers. Key methods and results are summarised below; a more comprehensive report is available on request.
-
The data. The datasets used in the study were obtained from the International Data Corporation (IDC) which tracks PC prices and specifications for the Australian market. The primary source of the data is electronic price lists from the major vendors of PCs in Australia (namely IBM, Compaq, Hewlett-Packard, Dell and Toshiba). The price lists are updated when new models are introduced, old models become obsolete or existing models undergo price change. These four vendors are estimated to have a combined market share of approximately 40 per cent for desktop PCs and 75 per cent for notebook PCs. The data used for our initial experimental work extends from April 2000 to June 2001.
-
The variables. The IDC dataset contains a wide variety of variables that might be suitable for hedonic modelling. For the dependent variable, the dataset provides a $A wholesale price. For explanatory variables, the dataset provides such items as: CPU speed; hard disk size; amount of RAM, video RAM and cache memory; size of monitor and screen resolution; presence of a network card, audio card and speakers; number of expansion slots; manufacturer; length of warranty; etc.
-
Hedonic function. A variety of functional forms was tested; the preferred forms are summarised in Attachment 1. The regressions fit fairly well, and variables were generally of the expected sign in nearly all regressions tested.
Constructing an Experimental Australian Computer Price Index -
A variety of experimental price indexes has been constructed using IDC data for the period April 2000 through June 2001.
-
The experimental indexes have been constructed using the following four methods:
-
Using the entire IDC dataset across all time periods, to estimate a hedonic price index directly from the coefficients of time dummy variables in the pooled regression. This is called the "time dummy variable method".
-
Drawing a sample from the IDC dataset, tracking that sample over time, and using the hedonic functions to adjust prices for quality change when models become obsolete and are replaced in the sample. This is called the "hedonic price imputation method".
-
Matching IDC datasets between periods and calculating the average price change (weighted by vendors' market shares) across all models which match between the periods.
-
Drawing a sample from the IDC dataset, tracking that sample over time, and using the "overlap method" for quality adjustment when models become obsolete and are replaced in the sample.
-
Figure 1 compares the indexes compiled for PCs with the index used at present in the Australian National Accounts (which is derived from the U.S. index as outlined above).
Figure 1
-
All experimental indexes indicate a more rapid fall in Australian PC prices than is indicated by the exchange-rate-adjusted U.S. index. It is interesting to note that there was a substantial depreciation in the $A during this period - and, if the experimental indexes are accurate, this may indicate squeezing of computer sellers' margins.
Next Steps -
The ABS must undertake a good deal of further investigation to refine and validate this work (and certainly before the experimental indexes might be used for national accounts deflation or other purposes). The issues to be investigated include:
-
what form of index construction based on the Australian data is most appropriate (and in particular whether it is better to construct the index directly from a pooled regression or to use the hedonic function indirectly to quality-adjust the index);
-
whether the implied changes in sellers' margins (in the presence of recent large exchange rate fluctuations) are plausible and sustainable;
-
how often the parameters of the hedonic function should be re-estimated and how often the specification of the function should be reviewed; and
-
the implications of basing official price and national accounting statistics on business by-product data such as those available from IDC.
-
The ABS is also considering the application of hedonic methods to other goods, such as other consumer electronics, motor vehicles and so on.
For More Information -
More details of this work can be obtained from richard.mckenzie@abs.gov.au or shiji.zhao@abs.gov.au
Issues for Discussion -
What experiences have other countries had with hedonic computer price indexes based on local data? Did the price indexes differ appreciably from the U.S. indexes?
-
What methods of constructing a computer price index have proven most successful? Are direct methods more suitable than indirect methods? How often should the hedonic models be re-estimated?
-
Are other countries applying hedonic methods to goods apart from computers?
Attachment 1 Preferred Form of Hedonic Regression for Personal Computers
Table A.1 : Hedonic Regressions for Desktop PCs
Table A.2 : Hedonic Regressions for Notebook PCs
The dependent variable (price) is logged. Among the explanatory variables, only RAM and VRAM are logged.
The explanatory variables include:
-
Cpu_score - measure of processor speed
-
HD (GB) - size of hard disk drive
-
RAM - amount of memory (logged to base 2 - somewhat more convenient, as RAM size is commonly expressed as powers of 2)
-
Cache - size of cache (logged)
-
VRAM - size of video RAM
-
D_FormFactor - dummy for case type
-
D_NtwCrd - dummy for presence of network card
-
D_Expand - dummy for large number of expansion slots
-
D_SCSICntr - dummy for presence of SCSI controller
-
D_Compaq; D-Dell - dummies for particular brands
-
Weight - weight of PC (notebooks only)
-
Battlife - battery life (notebooks only)
-
Display_size - size of screen (notebooks only)
-
D_DVD— dummy for presence of DVD drive
-
dummies for time period
Share with your friends: |
