RDP 2002-06: Output Gaps in Real Time: are they Reliable Enough to use for Monetary Policy? Appendix C: Detailed Phillips Curve Specifications

Specifications for the Preferred Phillips Curve Method

The five broad specification types for the preferred Phillips curve method are summarised in Table 1 in Section 2. Table C1 provides a complete listing of the specifications used for each of our 121 data vintages.

The specifications over the period 1971 :Q4 to 1973:Q3 are the only ones which do not contain any oil-price-inflation terms. This presumably reflects the relative lack of movement in oil prices before the OPEC I oil shock (see Figure 1). Our inability to identify a separate role for oil prices in these early data vintages may also reflect the fact that our measure of import prices before 1985 does not exclude oil (see Appendix D). Hence, oil-price effects may be captured in these early-vintage equations indirectly through lagged import-price-inflation terms, rather than showing up directly.

The first appearance of oil-price-inflation terms in our Phillips curve specifications occurs in 1973:Q4. Frequent re-specifications are required over the following two years of data vintages because of the extreme volatility in oil prices over this period. These re-specifications principally involve changes in the lags of oil-price-inflation terms with the inclusion, for 1974:Q4 and 1975:Q4 vintages, of that lag which includes the near quadrupling of Australian-dollar oil prices in 1974:Q1. Interestingly, despite the continued volatility of oil prices over the remainder of the 1970s, the 1975:Q4 specification continues to perform well until the mid 1980s – and when the Phillips curve specification is next changed in 1986:Q3, the required modifications are relatively minor.

Finally, the introduction of chain-linking in the National Accounts in 1998:Q3, and the associated switch from System of National Accounting (SNA) 1968 to SNA 1993 as the basis on which the accounts are prepared, results in significant revisions to the entire history of real GDP. As a consequence of these revisions, re-specification of our Phillips curve is required, with the coefficient on the output gap falling considerably for the 1998:Q3 specification relative to that for 1998:Q2, although part of this fall is retraced in subsequent vintages.

The optimal Phillips curve specification for the final data vintage (2001:Q4) is explicitly checked, since we regard the results from this vintage as giving us our best available estimate of the output gap over history. As it turned out, no further change in specification is required for this final vintage. Unlike in 1973 and 1974, the considerable rise in the oil price in the second half of 2,000 seems not to have substantially affected the performance of the 1998:Q3-optimised specification, with the effects of this rise captured by the terms already included in this equation.

Specifications of Constant Potential Output Growth Phillips Curves

Here, potential output is assumed to follow a simple linear trend, . We then conduct specification searches for optimal Phillips curves for each data vintage, following the same approach as for the preferred Phillips curves. The estimation results for the final-vintage Phillips curve are shown in Table C2. While the equation appears quite impressive in terms of goodness of fit, the derived output-gap estimates appear very poor, as the results in Tables 3 and 4 make clear. The optimal Phillips curve specifications for all vintages are set out in Table C3.