RDP 2004-02: The Impact of Rating Changes in Australian Financial Markets 5. Results

Given that ratings relate specifically to bonds rather than equities, we present the results for bond yields first. In each case, we present results for average abnormal returns on the announcement day and for the two days prior to and following announcements. In addition, we present cumulative average spread changes or abnormal returns for three windows: pre-announcement (day −20 to −1); announcement (day 0 and 1, allowing for some lagged response of prices); and post-announcement (day 2 to 20). Most of the discussion of our results focuses on the market movements over these three broader windows.

5.1 Bonds

While the small sample of events must remain a caveat, the overall picture from the event study for bond yields is that there is essentially no movement in yield spreads prior to ratings announcements, and then a small but statistically significant change immediately following the announcements (Figure 2 and Table 1). Looking first at the estimation window (days −120 to −21), the data for cumulative average spread changes in the estimation window suggest that bonds subject to upgrades and downgrades both exhibit some very modest tendency to narrow in the period leading up to the announcement. At first glance, this common narrowing of spreads – although it is not statistically significant – appears puzzling, but it appears to be due to the natural tendency for credit spreads to narrow gradually as maturities shorten. In particular, the analysis in Appendix B suggests that for bonds of around 2 to 3 years outstanding maturity, the passage of 100 days should normally be associated with a narrowing in spreads of around 4 basis points (bps), which is approximately what is observed in Figure 2.

Looking next to the event window, the cumulative spread change in the pre-announcement (days −20 to −1) and post-announcement (days 2 to 20) windows are statistically insignificant. There is, however, a statistically significant movement in spreads on the day of both downgrades and upgrades. The cumulative average changes on the day of the announcement and the day following the announcement (days 0 and 1) are a fall in spreads of 3.4 bps for upgrades and a rise of 10.3 bps for downgrades. In addition, around 80 per cent of the events show changes in spreads of the expected sign during this two-day window, with the observed proportions for downgrades being statistically significantly different from 0.5.

Given the average levels of spreads prior to announcements (66 bps for upgrades and 90 bps for downgrades), these changes around the immediate announcement period correspond to percentage changes in spreads of around 5 per cent for upgrades and 11 per cent for downgrades. It is noteworthy that the impact for downgrades is both larger and more protracted, with a widening of about 4 bps on the day of the announcement and a further 6 bps the following day. The size and pattern of this impact partly reflects the effect of one particularly influential observation, the downgrade of Southcorp from BBB to BB+, a non-investment grade. The measured spread of this bond widens by 90 bps on the day after the downgrade announcement and contributes about 4 bps to the average spread change on day 1.^[16] While this delayed adjustment is puzzling, the average spread change on day 1 remains significant even when this bond is excluded. It is unclear whether this delayed adjustment indicates that markets do not absorb new information immediately, that some rating changes were made late in the trading day, or simply that it is difficult to obtain accurate daily data on bond spreads.

Comparing our results with earlier studies, it is perhaps noteworthy that we find little evidence of bond spreads moving in advance of the changes by rating agencies, as would be expected if the decisions of agencies are based at least partly on information that is already in the public domain and reflected in pricing. Data problems due to the illiquidity of the market would be one possibility, but this seems unlikely given that we do see a significant change in bond spreads in the immediate announcement window. Indeed, based on the analysis in Appendix B, it would seem that the estimate of the average change in spreads on days 0 and 1 represents around half of the total adjustment in spreads that would be expected. In particular, the median rating change in our sample is between A and A−, and the analysis in Appendix B would suggest that this difference in ratings is typically associated with an ‘equilibrium’ spread differential of around 9 bps. Given that the median two-day change in spreads in our analysis is around 4.5 bps, it seems that half of the total adjustment we would expect to see in yields is occurring in the immediate period around ratings announcements.

The analysis in Appendix B also gives an indication of the magnitude of change in bond prices that might be expected to result from a one-notch rating change. For the median bond in our sample, a 9 bps change in yields (assuming, for illustrative purposes, a coupon and yield of 7 per cent and maturity of 2½ years) will result in a price change of only about 0.21 per cent. Given the relatively small size of this implied change and the various data problems in the bond market, it is perhaps not surprising that other event studies have frequently failed to find significant impacts on bond prices or returns.

5.2 Equities

The behaviour of equity returns around rating announcements is quite different to the behaviour observed in bond markets. The greatest contrast is in terms of the significant market movements that take place over a long period prior to rating announcements. This is illustrated most starkly in Figure 3, which shows that companies subject to downgrades on average have underperformed the broader market by around 12 per cent in the estimation window (days −120 to −21).^[17] By contrast, there is little evidence of outperformance by companies that are subsequently upgraded by rating agencies.

For downgrades, we find continuing evidence of underperformance by companies in the pre-announcement period of the event window (days −20 to −1). The cumulative average abnormal return for companies that are subsequently downgraded is around −6 per cent; and 64 per cent of the events are associated with negative returns, which is statistically significantly different from 50 per cent (Figure 4 and Table 2). By contrast, there is only very weak evidence of positive abnormal returns in this period for companies that are subsequently upgraded. In particular, the cumulative average abnormal return of 1 per cent in this period is not significant, nor is the proportion of events with positive returns.

Around the announcement, we find that downgrades are associated with negative returns both on the day of the announcement and the day after the announcement, with cumulative average abnormal returns over this period (days 0 to 1) of −1.3 per cent. However, this underperformance is small compared with the underperformance that is seen over the previous 20 days, and indeed the 100 days prior to the event window. Thus, it appears that the information that prompts rating agencies to lower ratings (or place ratings on negative watch) is largely already reflected in equity prices, and that the decisions of the agencies convey little additional information to the market. In addition, there is no evidence of any protracted market reaction to rating changes, since the cumulative average abnormal return for days 2 to 20 is insignificant.

Turning to upgrades, we find some evidence that companies that are upgraded by rating agencies outperform the general market on the day of the announcement. The average abnormal return of 1.2 per cent on the event day is statistically significant, and the proportion of events showing positive abnormal returns is 74 per cent which is also statistically significant. There is no evidence for ongoing adjustment in market prices in either the average abnormal returns for day 1 or for days 2 to 20. Overall, the cumulative abnormal return in the entire event window (days −20 to +20) is around only 2 per cent for upgrades, versus around −8 per cent for downgrades.

Our results for the immediate market impact of upgrades and downgrades are slightly at odds with earlier research. In particular, earlier studies have mostly suggested that the impact of downgrades tends to be larger than the impact of upgrades. Notwithstanding firms' obligations to provide market-relevant information in a timely manner, one explanation for this earlier finding would be that firms have an incentive to release positive information as soon as possible, and to hide negative information, which would make downgrades more newsworthy. Our point estimates for the two-day announcement window (days 0 and 1) suggest little difference between the magnitude of the effects for upgrades and downgrades (1.1 per cent and −1.3 per cent, respectively).^[18] However, the two-day effect for upgrades is not strictly statistically significant, so perhaps the contrast with earlier work is not that large. It is also noteworthy that the results for bonds are consistent with the notion that downgrades have larger immediate impacts than upgrades.

While our results provide evidence that both bond and equity prices respond to ratings announcements, the statistical significance of the results is stronger for bond yields than for equity prices. This might reflect the fact that ratings refer specifically to probability of default on a company's bonds, and are not necessarily broader indicators of a company's health. Indeed, as was discussed in Section 2, the response of equity prices to changes in bond ratings is theoretically ambiguous. In theory, some rating changes may reflect an agency's view that a firm's management is acting in a way that benefits one class of claimants at the expense of the other. Hence, an upgrade (downgrade) to the firm's bond rating might be bad (good) news for equityholders. The results presented so far provide no evidence for the presence of such cases: on average, upgrades (downgrades) are associated with positive (negative) price effects for equities.

However, in principle it is possible that there might at least be some proportion of events where the rating announcement prompts some perverse response in equity prices. If so, the distribution of abnormal returns for upgrades or downgrades on the announcement day would consist of a mixture of two different distributions, one with a positive mean and the other with a negative mean. We can test for this by examining the cross-sectional standard deviation of announcement-day abnormal returns to see if this is substantially larger than the equivalent standard deviation in the estimation window. When we do so, we find that the standard deviation of abnormal returns on upgrade days is 1.8 per cent, which is actually less than the estimation-window average of 2.8 per cent. The standard deviation for downgrades of 2.9 per cent is higher than the estimation-window average of 2.6 per cent, but it is far from being statistically significantly different. Hence, these results provide no support for the existence of a subset of events for which equity prices respond perversely to the announcement of bond rating changes.

5.3 Firm Size and Type of Rating Change

In addition to differences in the response to upgrades and downgrades, it is possible that the effects of rating changes may vary in other ways across types of rating changes or firms. To explore this possibility, we split the sample in three ways to test hypotheses relating to different response patterns. Since the sample of bond events and equity upgrades is quite small, we limit this analysis to equity downgrades.

The first possibility is that the impact of a ratings announcement may depend on whether or not the rating change is ‘anticipated’. More specifically, does the fact that a firm has earlier been placed on a rating watch reduce the market impact of the announcement of a subsequent downgrade, relative to the market impact of a watch announcement or a rating change that was not preceded by a watch? For this test, our data suggest that we can reject the hypothesis that the market impact of the two types of events is similar. In particular, the data suggest that only unanticipated downgrades have an impact in the immediate event window (Table 3). For the two-day announcement window (days 0 and 1), the impact of unanticipated events is −1.8 per cent versus an insignificant −0.7 per cent for anticipated events.

The second hypothesis is that the reaction of a security's price to a rating change might depend on the size of the particular firm. A difference might arise if large firms are subject to more market scrutiny, so that the opinions of the rating agencies would be less influential and have a smaller impact on prices. To test this, we divide the 62 firms into 2 equal-sized groups according to their relative market capitalisations. We find some evidence for differential market impact based on firm size. Announcement window returns are significantly negative for smaller firms but insignificant for larger firms (−2.0 per cent versus −0.6 per cent). In addition, there is also a large difference in abnormal returns in the 20-day period leading up to the rating announcement, when smaller firms have much larger cumulative abnormal returns (−8.8 per cent versus −1.4 per cent). It is unclear what the explanation for the latter difference might be.

The third hypothesis is that rating changes might matter more when they carry an issuer from investment to speculative (or junk) grade. This is because some portfolio managers face restrictions on the type of credits that they can hold, and because movements below the investment grade threshold could have important psychological effects. However, the non-linear relationship between credit ratings and default probabilities might also explain why rating changes toward the lower end of the ratings spectrum, such as a downgrade into speculative grade, would be associated with larger price movements. The small number of rating changes (only 7) is a constraint in drawing strong conclusions in this regard, but there is some evidence that rating changes from investment to sub-investment grades do have larger market impacts than other rating changes (−3.2 per cent versus −1.2 per cent). This would be consistent with the notion that markets do pay more attention to rating changes that lower credit ratings to sub-investment grade levels. There is also evidence that rating changes to sub-investment grades are preceded by substantially larger abnormal returns in the 20-day period prior to announcements (−15.1 per cent versus −4.9 per cent). This may reflect the particular 7 cases in our sample, or it may reflect a broader wariness on the part of agencies to downgrade companies through this important barrier, so that such downgrades only occur after a very substantial amount of adverse news about the firms in question.^[19]

Footnotes

The median two-day spread change of 8.4 bps is only modestly smaller than the average change of 10.3 bps, so our results do not appear to be excessively affected by any particular outlier. [16]

It is, however, noteworthy that the downgrades in our sample have historically not occurred in times of falling equity markets. The data suggest that downgrades occur independently of movements in the broader markets, so that rating agencies have responded to idiosyncratic weakness of issuers, rather than weakness in conditions for the broader corporate sector. [17]

However, an explanation for the lack of evidence in our results for larger impacts from downgrades may be that, in our sample, downgrades are more likely to be anticipated than are upgrades: 41 per cent of downgrades are preceded by a watch, compared with only 20 per cent of upgrades. As we show in Section 5.3, the response of equity prices to rating changes is apparently smaller when the changes are (at least partly) anticipated. [18]

Johnson (2003) suggests that agencies may be aware of the importance of the investment grade barrier, and are mindful of the significance of downgrades through this barrier, so that the BBB−/Baa3 rating may in effect be ‘wider’ than other surrounding ratings. [19]