RDP 9808: What Moves Yields in Australia? 5. Announcement Effects on Yields in Australia

This section first discusses any general intra-day pattern of volatility in yields, then examines whether specific large movements tend to be associated with the announcements nominated, and finally, examines the extent to which there are systematic influences on yields from these announcements.

5.1 Patterns of Intra-day Volatility

Figures 3 and 4 show a strong intra-day pattern of volatility in yields on days when announcements were made compared with days on which there was no announcement.^[14] Volatility is defined as the standard deviation of the change in the yield in five-minute intervals over the trading day for the period from January 1994 to September 1997. The impact of news on bill and bond yields is clearly visible in the spikes in volatility at times when announcements were made, i.e. at 9.30 am for monetary policy announcements, and at 11.30 am for most other news.^[15] On days when no announcements were made, volatility was much flatter.

Other patterns evident from Figures 3 and 4 are:

• monetary policy announcements, as would be expected, had a notably larger effect on volatility in bill yields than on bond yields, whereas releases of economic news had roughly the same effect on volatility of both yields; and
• economic news releases had a larger effect on bond yields, relative to monetary policy announcements (and post-Board dealing intentions), whereas the reverse was true for bill yields.

Volatility in yields was also higher at the open and close of trading than at other times. This feature is slightly more evident on days when announcements were not made, possibly because, on days when announcements were made, investors were focusing on forthcoming news, while on other days the domestic market looks for direction from trading in overseas markets. There is also more pronounced volatility on both sets of days as the SFE opens for afternoon trading.

5.2 The Largest Movements

The 25 largest movements in five-minute intervals in each episode are listed in Appendix C.^[16] There is no science in the choice of the number 25. However, the twenty-fifth largest movement takes the large movements near a level that might be regarded as a ‘normal’ response of yields to any announcement, while also giving a reasonable representation across the range of news items. The largest move in bill yields in a five-minute interval over the whole period was 40 basis points, while the smallest ‘large’ move was 4 basis points; the corresponding figures for bond yields were 26 basis points and 4 basis points.

A high proportion of the 25 largest moves in bill and bond yields has been associated with the announcements nominated in Table 1, as illustrated in Table 4 below. They are also widely spread over the range of news. The announcements which have most frequently affected bill yields by a large amount are (in descending order): the labour force, monetary policy announcements, the CPI and AWOTE. The corresponding announcements for bond yields are: the labour force, the CPI and monetary policy announcements. While the CPI and AWOTE announcements have affected bill and bond yields less frequently than the monthly labour force news, the proportion of the quarterly CPI and AWOTE announcements contained in the list is higher than for other items of statistical news, including labour force news.

Most of the largest moves associated with these announcements came in the interval immediately after they were made. There are, however, a number of exceptions: for bill yields, a large move in the 5–10 minute interval after the news of the tightening in policy in December 1994; a lagged reaction of five minutes to the CPI (released simultaneously with AWOTE) in April 1995 and again in January 1997; and large moves within quarter of an hour of labour force news in May 1995, and in September 1997.^[17] A large move in both bill and bond yields in July 1997 immediately preceding building approvals and retail sales news; and a large move in bond yields between 11.40 am–11.45 am on the morning of the Governor's first semi-annual appearance before the Parliamentary Committee in May 1997, which coincided with the release of labour force news at 11.30 that morning.

5.2.1 Comparison between bills and bonds

While similar items of news seem to affect bill and bond yields, there appears to be some difference of emphasis:

• ‘monetary policy’ news – monetary policy announcements, regular Reserve Bank economic and market commentary and announcement of post-Board dealing intentions – has been more often associated with a large move in bill yields than in bond yields;
• information about the CPI, AWOTE and economic activity has been equally likely to be associated with large moves in bill yields and bond yields; and
• fiscal policy announcements were more likely to produce a large move in bond yields than in bill yields.

These features would be consistent with markets viewing short-term interest rates as being tied down by the setting of monetary policy whilst revisions to expectations about the inflation or economic outlook have implications for both short- and long-term interest rates. All fiscal policy announcements in the period affected bond yields (whereas only half affected bill yields), suggesting that the bond market reacted to information about the supply of bonds implied by the Budget balance.

5.2.2 Comparison of episodes

Almost all monetary policy announcements had an effect on both measures of yield.^[18] Beyond that, the CPI and labour force releases were the two announcements which consistently and frequently affected both measures of yield. Both series have direct relevance for monetary policy; and, as noted, the Labour Force Survey is the most timely reading on the economy of major significance.

These announcements aside, there are notable changes of pattern between episodes. News on AWOTE became progressively more prominent, affecting bill yields and bond yields with increasing frequency as time passed. This perhaps reflects greater experience of investors and funds managers with the inflation-targeting framework, and the vital role of wages as a forward-looking indicator of inflation. The incidence with which Reserve Bank economic commentary and post-Board dealing intentions affected bill yields also increased. Monthly news on retail sales was associated with more large movements in bond yields in the third episode than earlier, but announcement effects from GDP became less prominent. Import news was important in the second episode, soon after this release was introduced, but not at other times. This may have been because imports provided a more timely reading on domestic spending than retail sales, at a time when policy-makers were looking for signs of slowing in the economy.

Balance of payments announcements became much less important over the period studied. This item was associated with a number of large movements in yields in the first episode but with none in the last. This may reflect several factors such as: improvement in the balance of payments situation in Australia over the third period, which took this indicator out of the headlines, i.e. there was a negative ‘flavour-of-the-month effect’; or, better understanding of the inflation targeting framework meant that this factor was simply no longer seen as directly relevant for monetary policy; or, traders came to appreciate that monthly balance of payments statistics were soft ground on which to rest policy or investment decisions. It is also possible that, more recently, news on the balance of payments had its main market impact on the exchange rate, not interest rates.

5.2.3 15-minute intervals

We also examined the 25 largest movements in 15-minute intervals in each episode. The large moves, and the frequency distribution of announcements, are listed in Tables D1 to D7 in Appendix D. In the first and second episodes, the number of large moves associated with news declines noticeably from the five-minute intervals – particularly for bill yields. The majority of these large moves not associated with announcements occurred within a half-hour of the morning open of trading, or in the quarter-hour reopening after lunch.

Broadly speaking, the items of news associated with large moves in yields are similar to those for the five-minute intervals. There are, however, two important exceptions: monetary policy announcements and labour force news produced fewer large movements in bill yields over the 15-minute intervals than over five-minute intervals. This may suggest that any large ‘knee jerk’ reactions to changes in monetary policy or labour force news tended to be quickly, at least partially, reversed.

5.3 Information Content of Announcements

The next step is to assess which announcements systematically affect yields. There are two aspects of interest. Firstly, we attempt to measure whether the market systematically differentiates between the information content of different announcements. A lay-person might refer to this as trying to establish the ‘importance’ of different items of news. Secondly, any market response should also reflect the size and direction of any surprise in the news. Putting these two points together simply says that it might be expected that a large surprise in the CPI would move bond yields by more than a comparable surprise in, say, building approvals since, for a variety of reasons, the information contained in the CPI is of inherently more value to bond and bill traders (or policy-makers) than that contained in building approvals.

We measure the information content of announcements by conducting a pooled regression, regressing the absolute movement in the relevant yield on a constant and dummy variables defined for each news item listed in Table 1 (Equation 2):

|Z_nt| is defined as the absolute value of the change in the yield (we loosely refer to this as yield volatility), on day n in interval t; where t = 9.30 am–9.35 am (for monetary policy announcements and post-Board dealing intentions), 11.30 am–11.35 am (for economic statistics) and 3.00 pm–3.05 pm (for the Reserve Bank's Bulletin). K is the number of different announcements included in the regressions. D_knt is the set of dummy variables defined for each announcement listed in Table 1.^[19] D_knt = 1 if announcement k is made on day n during interval t, D_knt = 0 otherwise.

We include an additional set of dummy variables to control for general yield volatility during the three time intervals. D₁ = 1, for interval 9.30 am–9.35 am, 0 otherwise; D₂ = 1, for interval 11.30 am–11.35 am, 0 otherwise.

The coefficient b_k provides an estimate of the impact of announcement k on yield volatility. Hence, we can ‘rank’ announcements in order of their average impact on yield volatility.

One factor that determines the degree of market reaction to an announcement is how hard it is to forecast the relevant variable. Indicators with large forecasting errors, i.e. those which tend to be associated with large surprises, might tend also to have large announcement effects, reflected in a large estimated coefficient in Equation (2). (The surprise element in news is discussed more fully in Section 5.4 below.) But an indicator might still be important for the operation of monetary policy (or for investment decisions) even if it has a very small coefficient in this regression: for example, if inflation could be forecast perfectly, there might be no effect on yields from CPI releases, yet inflation would, of course, still be important in considering monetary policy. (Presumably, in such a case, changes in yields would occur at the time market forecasts were changed.)

While acknowledging this point, it is also true that if, say, the coefficient for a particular variable were larger than that for another variable, while the forecasting error of the former were the same as (or smaller than) the latter, this would be prima facie evidence that the former variable contained more information for the market than the latter. We return to this point below.

The impact of announcements on yield volatility is summarised in Table 5. The estimated coefficients in the table show the average difference in yield volatility during the five-minute interval after an announcement, as compared with the same period on days on which announcements are not made. For example, the announcement of the CPI is estimated to have been associated with a move of about 4 basis points, on average, in the 90-day bill yield in the following five minutes, compared with days on which no such announcement was made.

These results suggest strongly that investors and portfolio managers do differentiate between information contained in different items of news; they are also broadly consistent with the pattern in Table 4.

Taking the whole period, the largest effect on yield volatility has come from monetary policy announcements, followed by the CPI. There is then a tier of news composed of AWOTE, the labour force and GDP. There has also been a smaller systematic response from announcements about the balance of payments, imports, retail trade and building approvals, which had similar effects to the release of the Bank's commentary and announcement of its post-Board dealing intentions. It is of interest that, as summarised in Table E2 of Appendix E, forecasting errors for the major economic statistics – the CPI, AWOTE and employment – are smaller than for other statistics, reinforcing the point that the market perceives that there is more information contained in these statistics than in the others.

The rescheduling of the financial aggregates release – which is published by the Reserve Bank – from 9.30 am to 11.30 am seems to have been associated with the market showing more interest in this announcement, although evidence over a longer period would be needed to test whether this is a robust conclusion.^[20]

5.3.1 Comparison between bills and bonds

The effect of monetary policy announcements was larger on volatility in bill yields than on bond yield volatility, as was AWOTE in the second and third episodes. The CPI and labour force news had a larger effect on volatility in bond yields than in bill yields over the entire period.

Only announcements of monetary policy changes, the CPI and the labour force had statistically significant effects on volatility in bill and bond yields in each episode. While the Bank's economic and market commentary and announcements of its post-Board dealing intentions had significant effects on bill yields over the whole period, these effects were relatively small. The Bank's post-Board dealing intentions had no significant effect on bond yields.

5.3.2 Comparison of episodes

The announcement effect of monetary policy changes on both bill and bond yields was larger in the third episode than in the first, notwithstanding smaller moves in the cash (policy) rate. This mainly reflected the fact that, in the first episode, the market moved well ahead of the policy announcement, and had typically overshot the eventual policy change. A major feature of the second episode is that, in the absence of any monetary policy announcements, the CPI had by far the largest announcement effect on bill yields and bond yields.

Merchandise imports were again newsworthy only in the second episode – apparently as an indicator of domestic spending, at the expense of retail sales. The importance of balance of payments news was much diminished from the first through to the third episode.

Again consistent with Table 4, AWOTE was much more newsworthy in the second and third episodes than in the first. In the later episode, such announcements had a bigger effect on volatility in bill and bond yields than any other news item, save monetary policy announcements.

The progressively more pronounced announcement effect of AWOTE may reflect the evolution of wage-setting and the monetary policy framework in Australia as well as the unfolding economic cycle in the 1990s. For much of the decade subsequent to 1983, wage-setting in Australia was determined under the Accord – an agreement between the Government and the ACTU, the peak trade union council. Under these arrangements, the main news about wages was typically contained in announcements about renegotiation of the Accord, rather than in the time series on AWOTE.

Early stages of the process of deregulating wage-setting in Australia overlapped with a period of recession, so that wages growth fell to historically low levels, which tended to remove this news from the financial headlines. Thus, for the best part of a decade, the release of statistics on average weekly earnings did not attract much comment, certainly much less in the mid and late 1980s than, say, news about the balance of payments. It was not until 1995, in response to unsustainable economic growth in 1994, that the latent threat of wages growth again exercised commentators' attention. Within the forward-looking inflation-targeting framework, AWOTE has since remained an important factor in discussions about the outlook for inflation. As the newly introduced Wage Cost Index builds up a track-record, however, it is possible that it will replace AWOTE as the main focus of public discussion of trends in wages.

5.4 Surprises

As noted, the efficient market hypothesis implies that only unanticipated news should influence yields, since asset prices should already reflect prevailing market expectations about the economic outlook. In this section, we investigate the impact of the unanticipated news that an announcement contains, as distinct from the information content of the announcement itself, as in the previous section. The approach adopted in this section assumes that yields respond according to the size of surprises in the news and that this response is symmetrical. In other words, an unexpectedly large rise in AWOTE would be expected to increase market yields, and the size of the rise in yields would reflect the extent of the surprise.

Although some statistical releases contain a range of information, we limit our analysis to the statistic which we judge to be of most interest to the market. For the Labour Force Survey, this means the employment data, not the unemployment rate. Underlying inflation, rather than the ‘headline’ rate, is used for the CPI. Because of difficulties in quantifying expectations, we exclude from this analysis: Reserve Bank commentary, announcement of the Bank's post-Board dealing intentions, the Budget and financial aggregates.

To estimate the effect on bill and bond yields of unanticipated news, we again conduct a pooled regression, regressing the signed change in the yield on the signed surprise, measured as the published statistic minus the median market forecast (Equation 3):

ACT_knt is the published estimate for series k as announced on day n in interval t. FOR_knt is the corresponding forecast. S_knt = 0 on days when no announcement is made.

To compare more easily the ‘surprise coefficients’ between variables – ensuring that they are representative of typical surprises – the surprise variable, S_knt, is scaled by the absolute average surprise for the relevant series; a list of ‘average’ surprises is contained in Table E2 of Appendix E. The equation estimated then becomes Equation (4):

where:

(N_k is the number of releases of announcement k in the sample); and Z_nt is the signed change in the yield. The coefficient c_k measures the effect of an average surprise on bill and bond yields.

Estimates of the effect on yields of average surprises in the news are summarised in Table 6.

For the period as a whole, all coefficients are significant at the 5 per cent level or lower, except those for GDP surprises on bill yields, and monetary policy surprises on bond yields. The estimates can be interpreted along the following lines: a published outcome for, say, the underlying CPI which was above the median expectation by 0.2 of a percentage point – the average forecasting error – would be expected to increase both bill and bond yields by about 5 basis points within five minutes of the announcement.

The pattern of coefficients in Table 6 is familiar. The average monetary policy surprise had the biggest effect on bill yields followed by the average surprise for underlying inflation, employment and AWOTE. For bond yields, the largest coefficient was for the CPI; average surprises in news about employment had the next largest effect.

The declining newsworthiness of the balance of payments and GDP, and the rising prominence of AWOTE, is again evident. It is also notable, as discussed previously, that inflation surprises had a larger effect on yields in the second period when the market considered changes in monetary policy to be off the agenda. To an extent, the rising effect of AWOTE seems to have been at the expense of the CPI in the third period. This may reflect the market's recognition that the two pieces of information are intimately linked for policy decisions. Since, in three quarters out of four, AWOTE is released the day after the CPI, market participants may have come to the view that any reassessment of the policy or investment outlook should be made on the basis of the two pieces of information considered jointly. If this were the case, most of the market reaction might be expected to come after the AWOTE release.

The fact that a significant ‘surprise’ coefficient is not found for monetary policy announcements in the case of bond yields can largely be attributed to the experience in the first episode when the bond market rallied on the announcement of the first and second tightenings in 1994 but sold off on news of the third. This produced a negative (and statistically insignificant) coefficient on bond yields in this period.

This market reaction suggests that the tightenings of 1994 were, on the whole, viewed ‘credibly’ by the bond market, i.e. the policy tightenings were expected to work to contain inflationary pressures and, on two occasions, induced investors to buy bonds, in the process driving yields lower. The easings of 1996 and 1997 were also viewed as being consistent with falling inflation and, hence, with falling bond yields. Rallies in the bond market in response to both tightenings and easings of monetary policy do not comply with the assumption on which the analysis is based, i.e. the effect of a surprise must be symmetrical. This result in relation to monetary policy announcements is similar to that found for the United States by Fleming and Remolona (1997).

The results for bill yields might shed light on how markets perceive the Reserve Bank's reaction function. These results suggest that the market reacts to news on the view that the Bank gives a greater weight to the underlying CPI, followed by employment and AWOTE, than to other individual indicators of economic activity. The rise in the market's reaction to AWOTE news might suggest that – since current growth in labour costs is likely to affect inflation some quarters hence – market participants have increasingly accepted that the Bank sets interest rates in a forward-looking way. Of the indicators of economic activity, the market seems to view the Bank as giving the largest weight to labour market indicators. This might be because either labour market conditions are of intrinsic interest to policy-makers, or they have a crucial bearing on the climate for wage-setting and inflationary expectations, or both.

5.5 Does the Market Respond Only to Unexpected News?

The similarity between the estimated coefficients in Tables 5 and 6 is a striking feature of the results – i.e. the results suggest that, over the period considered, the ‘average’ information contained in an announcement was almost identical to the effect on yields of the ‘average’ surprise in these announcements. Such a result may be consistent with an interpretation that the market reacted only to the unanticipated information contained in these announcements.

To test this hypothesis (albeit somewhat crudely), we extend Equation (2) to include absolute surprises for each variable for the whole period, to attempt to differentiate between reactions coming from announcements per se and from surprises they might contain, i.e:

In a market which responded only to unanticipated news (and in which the response was proportional to the size of surprise), for each variable, we would expect a statistically significant estimate of coefficient c (on absolute surprises), and an insignificant estimate of coefficient b (on announcement dummy variables).

In interpreting the results, we ask: does the market respond only to unexpected news? The answers, as summarised in Table 7, are classified according to the following rules:

• ‘yes’, if the estimate of b is statistically insignificant and that for c is significant at the 5 per cent level;
• ‘mainly’, if both b and c are significant but c exceeds b; and
• inconclusive otherwise.

It is not possible to use this test to answer ‘no’ to the question posed. Since Equation (5) is specified in terms of absolute movements, it does not capture a key element of the market reaction to surprises – the direction of the move in yields; it also allows only for a linear relationship between the surprise and the market response.

On the basis of this test, there seems to be a tendency for fixed-interest markets in Australia to react only to unexpected information. Using the above criteria, reactions to nine of the 18 announcements contained in Table 7 are classified in this way. The results are, on the whole, stronger for the bond market than for bills. This might reflect the dominating influence of monetary policy announcements on bills, whereas the bond market adjusts more gradually to the content of underlying economic news as it comes to hand.

In the case of AWOTE, the results are inconclusive, i.e. the announcement itself appears to be the source of volatility, rather than the unexpected component of the announcement. There might be a number of strands to the explanation of this result. Firstly, it is influenced to an important extent by the earlier periods, when for reasons explained above, announcements of AWOTE were apparently regarded as containing little information and yields did not react systematically to surprises in the data. There is some evidence, however, from tests of the period since mid 1996, that the market has reacted mainly to surprises in AWOTE.

A second factor might be the interaction of two aspects of our approach: the frequency with which data on market expectations are published; and the fact that publication of AWOTE usually comes the day after the CPI. As noted, expectations for all of the series tested are published once a week, on the Friday before the relevant statistics are released.^[21] This approach assumes that expectations are not then revised. For most statistics, this seems to be a reasonable assumption. In the case of AWOTE, however, the release of the CPI (usually) on the day before AWOTE might induce some revision to forecasts of AWOTE, especially if the inflation news contains a surprise. Any such reassessment would not be captured in our approach, so the standard median market forecast might be out of date by the time AWOTE is announced.

On the balance of payments, the inconclusive results may reflect flavour-of-the-month considerations, perhaps partly due to growing acceptance that news about the balance of payments is not a major factor in monetary policy decisions.

Footnotes

Intra-day bill and bond yield volatility for each episode reviewed is illustrated in Figures B1 and B2 in Appendix B. [14]

There is also a rise in volatility, at around 3.00 pm, on days when the Bank's Bulletin is released. [15]

Tables C1, C2 and C3 for bill yields, and Tables C4, C5, and C6 for bond yields. [16]

In each of these instances, there was also a large move in the interval immediately after the announcement. This explains why there are four large movements in bill yields associated with policy announcements in the first episode, while there were only three policy tightenings. Most of these lagged large reactions were continuations of the initial movements, although some were reversals. [17]

The exceptions were the tightening on 14 December 1994 (for bond yields) and the easing on 6 November 1996 (for bill and bond yields). In both episodes, the policy adjustment was already fully priced into the relevant yields ahead of the announcement. [18]

A dummy variable is not defined for fiscal policy statements or Governors' speeches as their release times are variable. [19]

As well as changing the release time, the Reserve Bank also announced in advance the date of release. This foreknowledge may have been a factor in the increased significance of this news. The change in timing brought this release into line with the normal scheduling of information released by the Australian Bureau of Statistics. The change was made following a review prompted by the IMF's recommendations on international data standards. [20]

This approach is determined by data availability. [21]