RDP 9803: Forward-Looking Behaviour and Credibility: Some Evidence and Implications for Policy 2. Evidence on the Formation of Inflation Expectations

This section examines some aspects of inflation expectations. We start by analysing the inflation forecast errors of financial market economists and households. In Section 2.2, we examine whether there are interactions between actual and expected inflation, testing whether expectations influence actual inflation. Most economists draw a distinction between financial markets and other groups in the economy in terms of forward-looking behaviour. Accordingly, in Section 2.3, we look at how the foreign exchange market responds to expected future monetary policy. A summary of the results is presented in Section 2.4.

2.1 Inflation Expectations

If expectations are ‘rational’, in the sense that they use all currently available information, then they are unbiased predictors of the actual inflation rate. We test this for two very different measures of inflation expectations. The first is the median response to the Money Market Services (MMS) survey of financial market economists' expectations of the next release of quarterly headline inflation. The second is the median response to the Melbourne Institute survey of what households expect inflation to be over the coming year.[1] Given that financial markets are thought to be relatively efficient, their inflation expectations would be expected to exhibit a smaller bias.

The top panel of Figure 1 plots the inflation-expectation error of financial market economists. Financial market economists thought headline inflation would be lower than it was during the second half of the 1980s, and they then failed to predict the disinflation that occurred from 1990 to 1992, with inflation repeatedly coming in well below the market expectation. Errors have been smaller and more evenly balanced between under and over predictions in the past few years. While Figure 1 suggests that financial market economists generally underestimate inflation when it is rising and overestimate it when it is falling, the average forecast error over the past 13 years is indistinguishable from zero.

Figure 1: Actual Less Expected Inflation
Figure 1: Actual Less Expected Inflation

The bottom panel of Figure 1 plots the inflation-expectation error of households, as measured by the difference between underlying inflation (one-year ahead) and the Melbourne Institute's measure of household expected inflation for the coming year. In contrast to financial market economists, households have systematically expected inflation to be higher than it turned out to be, with this bias not falling substantially with the mean shift in inflation in the 1990s. Appendix A presents the results of a formal test of rational expectations, indicating that neither of these expectations series conform with the predictions of the rational expectations hypothesis, although the rejection is much stronger for households than for financial market economists.

2.2 Interactions Between Actual and Expected Price Inflation

We next examine the interaction of actual inflation and measured inflation expectations. The question we have in mind is whether movements in actual inflation occur before changes in expected inflation, or the other way around. Before we answer this question, there are a few preliminary data issues to resolve.

The first is to decide which inflation series is relevant. There are (at least) two candidates. On the one hand, monetary policy is made with reference to underlying CPI inflation, since this removes anomalous and volatile items, providing a more accurate representation of ‘true’ price pressures in the economy. On the other hand, the published ‘headline’ figure typically has more coverage in the non-specialist media, and so is more likely to influence popular opinion. Moreover, underlying inflation only gained prominence in the 1990s. The upshot is that it is unclear which series is relevant, and so both are examined. The second issue is the selection of the inflation expectations measure. A broad measure is preferred, and so the median response from the Melbourne Institute survey of households' inflation expectations is used.

Figure 2 plots headline and underlying CPI inflation with the quarterly reading of the Melbourne Institute survey measure of what households expect inflation to be one year ahead (hereafter called ‘inflation expectations’).[2] Eyeballing these data, it is difficult to draw a clear inference about whether actual or expected inflation ‘leads’ the other. Expected inflation hardly responded to the substantial changes in actual inflation through the late 1970s and 1980s, although this was not a bad call in the sense that two of the three reductions in inflation over this period proved to be transitory. (There is a ‘chicken and egg’ dimension to this, since the falls in actual inflation were bound to be transitory if expectations did not change.) The fall in inflation expectations over 1989 to 1991 to about 4½ per cent occurred on the back of a steady reduction in underlying inflation. Expected inflation ticked up slightly in late 1994, ahead of the rise in underlying inflation, but fell in 1996, to close to 3 per cent, following lower underlying and headline inflation. On this measure, expected inflation was around 3 per cent at the end of 1997, which was still well above actual inflation.[3]

Figure 2: Actual and Expected Inflation
Figure 2: Actual and Expected Inflation

Granger-causality tests provide a simple statistical tool to assess whether changes in actual inflation precede expected inflation or vice versa. Table 1 lists the Chi-square statistics for excluding lags of the relevant variables in a two-variable four-lag VAR (similar results apply to estimation in a two-lag VAR and in first differences of inflation). There are two points of interest. First, the results indicate that there is feedback between actual and expected inflation, such that it cannot be said that one systematically precedes the other. The statistical importance of expected inflation to actual inflation has increased in the 1990s, suggesting that expectations may have become more important to the inflationary process in recent years. Second, while actual inflation has consistently been a predictor of expected inflation, the headline rate seems to have been supplanted by the underlying rate of inflation as the driving force behind expected inflation.

Table 1: Granger-Causality Tests for Actual and Expected Inflation
  Dependent variable
  1977:Q3–1997:Q3 1980:Q1–1989:Q4 1990:Q1–1997:Q3
  Headline inflation Underlying inflation Expected inflation Headline inflation Underlying inflation Expected inflation Headline inflation Underlying inflation Expected inflation
Excluding lags
of headline
498** [0.00] n.a.
13.4** [0.01] 62.8** [0.00] n.a.
16.8** [0.00] 211** [0.00] n.a.
6.0 [0.20]
Excluding lags
of underlying
823** [0.00] 15.9** [0.00] n.a.
491** [0.00] 6.8 [0.15] n.a.
60.5** [0.00] 13.1** [0.00]
Excluding lags
of expected
147** [0.00] 17.4** [0.00] 176** [0.00] 78.1** [0.00] 20.1** [0.00] 277** [0.00] 144** [0.00] 30.6** [0.00] 63.3** [0.00]

Note: The table reports χ2(4) statistics; figures in brackets are marginal significance values; ** and * indicate 1 and 5 per cent significance; all statistics are Newey-West adjusted for serial correlation induced by overlapping observations.

2.3 Inflation Surprises and the Exchange Rate

Economists generally believe that financial markets are more explicitly forward looking and less likely to be systematically biased in forming their expectations than other groups. This is consistent with the analysis of inflation forecast errors in Section 2.1. We look here at how the foreign exchange market anticipates monetary policy will respond when actual inflation differs from the expected outcome.

In a floating exchange rate regime, when inflation is higher than expected, the nominal exchange rate depreciates in order to maintain the real exchange rate if no policy response is anticipated. If monetary policy is focused on inflation, however, policy would generally be tightened in response to the higher inflation forecast. In a rational market, foreign exchange market participants anticipate the higher real interest rates, and the exchange rate may appreciate in response to the positive inflation surprise. This is a testable proposition (Kim 1994; Amano et al. 1996). It provides an insight into the anti-inflation credibility of a central bank as assessed by the financial market, since it indicates whether or not the market believes that the monetary authority will accommodate inflation shocks.

Following Kim (1994), we estimate the equation:

where er is the US dollar per Australian dollar exchange rate (so a rise is an appreciation of the Australian dollar), πe is the expected inflation rate and surprise is actual inflation less the expected inflation for that quarter. If the foreign exchange market is efficient, so available information is already incorporated into the exchange rate, then α = β = 0 and γ ≠ 0. If the inflation surprise indicates higher future inflation, then γ < 0 if a policy response is not anticipated, but γ > 0 if a policy response is. Kim (1994) used quarterly data for Australia from 1985 to 1992, split the sample into two periods of about a dozen observations each, and found that the exchange rate tended to depreciate when published headline inflation was above expected inflation from 1985 to 1988 inclusive, but tended to appreciate after 1988.

The measure of expected inflation that we use is the MMS survey of financial market economists' expected headline annual CPI inflation the week before the ABS releases the CPI data. The data are available from March 1985 to June 1997. The inflation surprise is headline inflation less the MMS measure of expected inflation. The ABS released CPI data at 9.00 am until the December 1988 release, and at 11.30 am from the March 1989 release. The exchange rate change is the log difference between the opening rate (before 9.00 am rate) and the 11.00 am rate from March 1985 to December 1988, and the log difference between the 9.00 am rate and the midday rate from March 1989.

The top panel in Figure 3 plots the γ coefficient from Equation (1) estimated by rolling a regression forward from March 1985 to June 1997 with a fixed window of 20 observations (5 years). The bottom panel shows its marginal significance. The inference drawn from this exercise is that during most of this period, inflation surprises have tended to have no clear impact on the exchange rate, but this has changed – and significantly so – in recent years. The exchange rate now systematically moves in the same direction as the inflation surprise, so that, for example, the exchange rate depreciates if inflation comes in lower than expected.[4] This suggests that the market believes that the Reserve Bank of Australia will not accommodate inflation shocks.

Figure 3: Inflation Surprises and the Exchange Rate
20-quarter rolling-window estimation of γ in Equation (1)
Figure 3: Inflation Surprises and the Exchange Rate

2.4 An Assessment

The evidence on the information that people use to form their expectations is mixed. In the first place, survey measures of inflation expectations for households and financial market economists are not consistent with the predictions of rational expectations, which is a common result in the international literature (Maddala 1989; Roberts 1997). The degree of inconsistency, however, differs between these groups. The median financial market economist missed changes in the inflation regime, but basically has ‘got it right’ once the mean shift in inflation took place. The median household is yet to get this far, with the size of the bias only falling moderately despite the substantial fall in inflation in the 1990s relative to the 1980s.

Moreover, financial market participants also exhibit explicit forward-looking behaviour which takes account of a policy reaction when inflation outcomes differ to what was expected. The exchange rate now systematically moves in the same direction as the inflation surprise, suggesting that the market anticipates that real interest rates will be changed in response.

Inflation expectations also seem to matter to the inflation outturn, as shown by the interaction between actual and expected inflation over the 1990s. Actual inflation is explained by its own past values and by past inflation expectations, which indicates that inflation expectations contain separate information about future inflation. Inflation expectations, in turn, however, also depend on past inflation, suggesting that inflation expectations are based on a mix of different sets of information.


Households are asked ‘By this time next year, do you think the prices of the things you buy will go up or down? If up, by how much? If down, by how much?’ We compare household expected inflation with underlying inflation since the questions refer to broad inflationary pressures in the economy, of which underlying inflation is a better indicator. (The results are not substantively different when headline inflation is used in place of the underlying rate.) We compare financial market economists' inflation expectations with headline inflation since they are asked specifically about the headline rate. [1]

The CPI data are released about six weeks after the end of the quarter under consideration. The Melbourne Institute expectations data were collected on a quarterly basis until December 1986, after which they were collected on a monthly basis (with the survey conducted at the start of the month and released in the middle of the month). Quarterly expectations from March 1987 are calculated as the average of the monthly responses. [2]

It is worth observing that while this measure of inflation expectations is currently around 2 to 3 per cent, the Australian inflation target is expressed in terms of a medium-term mean, and does not specify a range or band within which inflation necessarily must lie in the short term for policy to be judged ‘successful’. This means that the sort of credibility test proposed by Amano et al. (1996) – whether inflation expectations over all periods lie within the band – is not necessarily an appropriate test of credibility. Given that the inflation target is a medium-term mean, short-term inflation expectations can lie away from the target rate at times even if the target is fully credible. The better test of credibility is that medium-term inflation expectations be consistent with the target (although short-term inflation expectations should also exhibit a tendency to revert to the target rate if the target is credible). [3]

Equation (1) also contains a test of whether the foreign exchange market is efficient, in the sense of whether it has used all available information (particularly about inflation) such that changes in the exchange rate are unpredictable. While the coefficients are not reported, the null hypothesis of financial market efficiency, α = β = 0, is not rejected for the 20-quarter rolling window regressions. [4]