RDP 2002-06: Output Gaps in Real Time: are they Reliable Enough to use for Monetary Policy? Appendix D: Data Sources and Definitions

Real-time Real GDP

The Australian real-time GDP data are described in Stone and Wardrop (2002). For each vintage we use the hybrid series, which Stone and Wardrop (2002) consider represents what analysts at the time would have considered the best measure of GDP. Specifically, this corresponds to: GDP(I) for each vintage from 1971:Q4–1991:Q3; GDP(A) for each vintage from 1991:Q4–1998:Q2; and chain-volume GDP for each vintage from 1998:Q3–2001:Q4. The bulk of our real-time output data set was supplied to us as unofficial data by Peter Rossiter from the ABS. This data set was then supplemented by the manual entry of additional information from historical tables contained in original hard-copy National Accounts releases, where available.

For vintages where the data did not go back to 1959:Q3, these series were back-cast wherever possible based on the growth rates from the most recent preceding vintage for which these data were available.

Consumer Prices

The price series we use is a measure of core consumer prices, the weighted-median CPI, which is calculated by the RBA and is available back to 1976:Q3 from Bulletin (RBA) Table G.1. The series is adjusted for the introduction of a Goods and Services Tax in 2000:Q3, assuming that the tax led to a 2.95 per cent rise in the weighted-median CPI in that quarter.

The weighted-median CPI is extended to cover the period from 1966:Q3 to 1976:Q2 by direct construction using data from Consumer Price Index (ABS Cat No 6401.0) [2001:Q4], Consumer Price Index Particulars for Sub-groups and Special Groupings (Commonwealth Bureau of Census and Statistics (CBS) Ref No 9.7) [1966:Q3–1971:Q2], Labour Report No 52 (CBS Ref No 6.7) [1965 and 1966], The Australian Consumer Price Index: Concepts, Sources and Methods (ABS Cat No 6461.0), and additional data from the ABS.

Before 1966:Q3, the weighted-median series is back-cast using All Groups CPI from the December quarter 2001 Consumer Price Index (ABS Cat No 6401.0).

Import Prices

From 1985:Q3 onwards, import prices are the implicit price deflator for merchandise imports, excluding fuels and lubricants, civil aircraft and RBA imports of gold, from National Income, Expenditure and Product (ABS Cat No 5206.0) [2001:Q4].

Before 1985:Q3, the series is back-cast using the imports implicit price deflator for goods from National Income, Expenditure and Product (ABS Cat No 5206.0).

The series we use is adjusted for tariffs (but not for the Balassa-Samuelson effect) using the approach described in Appendix C of Beechey et al (2000). For 1969:Q3–2001:Q4 the tariff rate is customs duty receipts divided by the value of merchandise imports (excluding fuels and lubricants, civil aircraft and RBA imports of gold), seasonally adjusted. Tariff revenue is from the Australian Customs Service. For 1959:Q3–1969:Q2 customs duties receipts are sourced from Overseas Trade (CBS) [1959/60–1968/69]. These annual figures are linearly interpolated to obtain a quarterly series.

Oil Prices

The US$ oil price is the average-quarter value of the price per barrel of West Texas Intermediate crude. For 1982:Q1 onward, this is sourced from the nearest contract price on Bloomberg, CL1 CMDTY. For the period before this it is back-cast using the average quarterly spot price for crude from the International Monetary Fund (IMF) International Financial Statistics (IFS) database (Datastream code WDI76AAZA).

The US$ oil price is converted to A$ using a quarter-average AUD/USD bilateral exchange rate. For 1970:Q1 onwards this is obtained from the IMF IFS database (Datastream code AUI..RF.). For the period before 1970:Q1, quarterly data are generated by linear interpolation of annual figures for the AUD/USD bilateral rate from Foster (1997), Table 1.19a, available at <http://www.rba.gov.au/Statistics/op8_index.html>.

Inflation Expectations from the Bond Market

The bond market inflation expectations series is derived by splicing together two different series. For the period from 1993:Q1 onwards we use the difference in the yield between a 10-year government bond and an indexed bond of comparable maturity. For the period prior to 1993, for the bulk of which indexed bonds were not issued by the Commonwealth, we use a variant of the approach used by Debelle and Laxton (1997) to generate inflation expectations estimates.

Following Debelle and Laxton (1997), we construct an Australian equilibrium real 10-year government bond rate series, , based on the ratio of the stock of OECD net public debt to GDP, debt_t :

and use Debelle and Laxton's value of 0.07 for β, which was based on the work of Tanzi and Fanizza (1995), so that a 1 percentage point increase in debt_t increases by 7 basis points.

The constant C in equation (D1) is a sum of the world real interest rate when OECD net public debt is zero and an Australia-specific risk premium, which we assume is constant. We choose a reference quarter, 1959:Q3, and assume that inflation expectations in that quarter were equal to average year-ended inflation for the preceding two years, since it was a period of quiescent inflation. Then setting the nominal Australian bond rate, i_t, equal to the sum of the Australian equilibrium real government bond rate, , and inflation expectations, , in that quarter, gives a value for C of −0.186.

These choices for β and C yield a series for inflation expectations for the period 1959:Q3 to 2001:Q4, based on the assumption:

This series is then spliced together with our inflation expectations series from indexed bond data, with the latter replacing the former in 1993:Q1, when the two measures differ by only 15 basis points.

As regards data sources, we use the end-quarter Australian 10-year government bond yield from Bulletin (RBA) Table F.2. Australian Treasury capital-indexed bond yields are from Bloomberg (screen: ILB).

An annual series for the OECD net public debt to GDP ratio is sourced from OECD Online Information Services (OLISnet). This series starts in 1970, and over the 1970s and 1980s moves quite closely with US General Government Net Financial Liabilities as a ratio of Nominal GDP from OECD Economic Outlook Database, Annex Table 34 (Datastream code USOCFNF%). In light of this co-movement, for the period 1960–1970 we back-cast our OECD debt to GDP series based on changes in this US-debt-to-GDP series. For the few years in our database before 1960, we assume that the OECD net public debt to GDP ratio is constant. To obtain a quarterly series, the annual series thus constructed is linearly interpolated.

Real-time Issues

There are a few real-time issues relating to these data. First, we use the final (2001:Q4) vintage of consumer price data throughout. While Australian CPI data are not subject to revision, periodic re-basing of the index may, for early periods, have resulted in very minor differences between inflation rates reported in real-time and those reported in 2001:Q4, resulting from the ABS's practise of rounding the index to one decimal place. These differences, however, are negligible.

Secondly, the 2001:Q4 vintage of import prices is also used throughout. Inspection of several vintages of the import price data suggests that, while revisions do occur, they tend to be small, especially relative to real GDP revisions, and are therefore unlikely to materially affect our results.

Finally, the construction of the bond market inflation expectations series requires an estimate of β, the assumed sensitivity of the equilibrium real 10-year bond rate to increases in OECD public debt. We use the value β = 0.07, based on Tanzi and Fanizza (1995), which was clearly unavailable before that time. Likewise, our use of a single, current series for the ratio of OECD public debt to GDP in our calculation of pre-1993 bond market inflation expectations is also, strictly speaking, subject to a real-time problem, as it neglects revisions over time to estimates of OECD GDP.

We have, however, examined the sensitivity of our results to the chosen value of β. Assuming a value twice as large, β = 0.14 (which also requires an adjustment to the constant, C, assuming that inflation expectations in the reference quarter, 1959:Q3, are equal to average year-ended inflation for the preceding two years, as before), leads to real-time output-gap estimates with a root-mean-square difference of 0.5 ppt from real-time estimates assuming our standard value, β = 0.07, over the period, 1971:Q4 to 2001:Q4. The corresponding RMSD between the final output-gap estimates over the same period is 0.7 ppt. We conclude that even large changes in the value of β lead to only small changes in our estimated output gaps.