The Effects of Federal Funds Target Rate Changes on S&P100 Stock Returns, Volatilities, and Correlations

Transcription

1 The Effects of Federal Funds Target Rate Changes on S&P100 Stock Returns, Volatilities, and Correlations Helena Chulia-Soler Department of Economics and Business Universitat Oberta de Catalunya Martin Martens Department of Finance Erasmus University Rotterdam Dick van Dijk Econometric Institute Erasmus University Rotterdam September 18, 2007 Abstract We study the impact of FOMC announcements of Federal funds target rate decisions on individual stock prices at the intraday level. We find that the returns, volatilities and correlations of the S&P100 index constituents only respond to the surprise component in the announcement, as measured by the change in the Federal funds futures rate. For example, an unexpected 25 basis points increase of the target rate leads on average to a 113 basis points negative market return within five minutes after the announcement. It also increases market volatility during the 60-minute window around the announcement with 147 basis points. Positive surprises, meaning bad news for stocks, provoke a stronger reaction than negative surprises. Market participants also respond differently to good and bad news. In case of bad news for stocks the fact that there is a surprise matters most, whereas in case of good news the magnitude of the surprise is more important. Across sectors, Financials and IT show the strongest response to target rate surprises. Keywords: monetary policy announcements, high-frequency data JEL Classification: E44, E52, G14 We thank participants at the Stanford Institute for Theoretical Economics (SITE) Summer 2007 Workshop on the Economic Analysis of High-Frequency Data and the Impact of Economic News, in particular Refet Gürkaynak, Michael Melvin and Jonathan Wright, for useful comments and suggestions. Any remaining errors are our own. Econometric Institute, Erasmus University Rotterdam, P.O. Box 1738, NL-3000 DR Rotterdam, The Netherlands, djvandijk@few.eur.nl (corresponding author)

2 1 Introduction In this paper we examine the effects of Federal Open Market Committee (FOMC) decisions concerning the Federal funds target rate on intraday returns, volatilities and correlations of the constituents of the S&P100 index. This analysis provides insights into the efficiency of the stock market around monetary policy news announcements and into possible asymmetry in the response of market participants to bad news and good news. The effects of macroeconomic and monetary policy news announcements on exchange rates, interest rates, and stock prices have been well documented. For stocks, most previous research examines the effects on the market as a whole, see Bomfim (2003), Ehrmann and Fratzscher (2004), Guo (2004), Bernanke and Kuttner (2005), Boyd et al. (2005), Gürkaynak et al. (2005), Wongswan (2005), Hausman and Wongswan (2006), Zebedee et al. (2006), Andersson (2007), and Andersen et al. (in press), among others. Bernanke and Kuttner (2005), Guo (2004) and Ehrmann and Fratzscher (2004) do consider heterogeneity in the response to FOMC announcements across sectors, across size portfolios and across individual stocks, respectively, but not at the intraday level as we do in this study. As noted by Bomfim (2003), the use of daily data may diminish the precision with which the impact of news is estimated as other economic data could be released on the same day. In an extreme scenario, endogeneity may arise if the stock market and monetary policy both respond simultaneously to other new information, see Rigobon and Sack (2003) and Bernanke and Kuttner (2005) for discussion. The use of high-frequency intraday data makes it possible to obtain more accurate measures of the response of stock prices to news announcements, as suggested by Fair (2002). Wongswan (2006) illustrates this point in the context of transmission of information contained in macroeconomic announcements across international equity markets. 1 Furthermore, the prior literature focuses on returns and volatilities, but does not consider correlations between assets. A notable exception is Christiansen and Ranaldo (2007) who examine the behavior of the correlation between the stock and the bond market around news 1 The effects of macroeconomic announcements on exchange rates and interest rates are more commonly analyzed with high-frequency intra-day data, see Andersen et al. (2003) and Faust et al. (2007), among others, for recent examples. 1

3 announcements. We focus on Federal funds target rate announcements for two reasons. First, many other macroeconomic news announcements take place prior to or at the opening of the stock exchange trading day, which bars the use of high-frequency intraday data. Second, there is a natural link between the discount rate and stock prices. The dividend discount model states that the value of a stock is equal to the future cash flows expected to be generated by the firm discounted at an appropriate risk-adjusted rate. In its most standard setting the value of a stock is equal to the dividend per share divided by the difference between the discount rate and the dividend growth rate. 2 Hence both an expected increase (decrease) in the future discount rate as well as an unexpected increase (decrease) in the current discount rate should lower (increase) stock prices. A further advantage is that surprises in rate changes can be derived from the Federal funds futures rate, as first demonstrated by Kuttner (2001) and subsequently adopted by Bomfim (2003), Rigobon and Sack (2004), Fleming and Piazzesi (2005), Gürkaynak et al. (2005), Wongswan (2005) and Zebedee et al. (2006), among others. In our empirical analysis we focus on two issues. First, we test whether intraday returns, volatilities and correlations respond to the fact that there is a monetary policy announcement, or whether they respond to the surprise component therein. In addition, we explore whether the mere occurrence of a surprise triggers a stock market reaction or whether the magnitude of the surprise plays a role. Second, we test for the presence of asymmetries in the effects of news announcements, by examining whether their effects are different for positive and negative surprises, for target rate increases and decreases, and during economic contractions and expansions. Consistent with previous studies, our results overwhelmingly confirm that surprises in target rate decisions matter, and that the actual rate changes do not. For example, an unexpected 25 basis points increase of the target rate leads to a 113 basis points negative market return in the five minutes immediately following the announcement 3 and to a 147 basis points increase in volatility during a 60-minute 2 See Bernanke and Kuttner (2005) for an extensive analysis of the question why stock prices respond to news announcements and to target rate surprises in particular, and also Campbell and Vuolteenaho (2004) on decomposing stock betas into news about future cash flows and future discount rates. 3 For a different sample period ( ) and using a different methodology Zebedee et al. 2

4 window around the announcement. The results for sector-level returns and volatilities show considerable differences in the effects of target rate surprises across sectors. As expected, Financials and IT show the strongest response, while Utilities and Energy show the weakest response. We also find that the response of stock returns, volatilities and correlations is rather different after positive surprises than after negative surprises. In the latter case (positive news for stocks), the increase in stock returns, volatilities and correlations is more affected by the magnitude of the surprise, whereas after positive surprises (negative news for stocks) the decrease in stock returns and increase in volatilities and correlations is mainly due to the mere occurrence of a surprise. Hence in case of good news market participants respond in a more rational way than in the case of bad news. As a result, the effects of positive news for stocks are smaller than the effects of negative news for small magnitudes of the target rate surprise (less than eight basis points) and vice versa for large surprises. For example, an unexpected decrease of the target rate with 5 (10) basis points provokes a positive market return of 16 (46) basis points, compared to a 25 (39) basis points negative return following an unexpected increase of the target rate of the same magnitude. The asymmetric response in asset prices to bad and good news has been widely analyzed in the literature. Bernanke and Kuttner (2005) study the effect of FOMC decisions on daily stock returns but find no evidence for asymmetry. We attribute the fact that we do find a different response to positive and negative news to our use of intraday data. Andersen et al. (2003) find that bad news has greater impact than good news on exchange rates using high-frequency data. Engle and Ng (1993) study the asymmetric response of volatility to lagged (residual) stock returns finding that volatility increases more after negative returns. Bomfim (2003) finds that positive surprises in Federal funds target rate changes tend to have a larger effect on daily S&P500 index volatility than negative surprises. Using intraday data we also find positive target rate surprises to result in higher stock volatility than negative ones for realistic magnitudes of the surprise. Christiansen and Ranaldo (2007) find no evidence for an asymmetric response in the correlation between the bond and stock market. We do find that the correlation between stocks increases more after negative (2006) report a negative market return of 43 basis points after a 25 basis points positive surprise. 3

5 news for stocks. This is intraday evidence of a well-documented asymmetric response of correlations to positive and negative news. Cappiello et al. (2006), for example, find that conditional correlation among regional groups increases dramatically when bad news hits financial markets. Ang and Chen (2002) find that correlations between US stocks and the aggregate US market are much higher for downside moves, especially for extreme downside moves, than for upside moves. The remainder of the paper is organized as follows. Section 2 describes our data. Section 3 lays out the methodology we use to analyze the effects of FOMC announcements on returns, volatilities and correlations of the S&P100 constituent stocks and discusses the empirical results. Section 4 concludes. 2 Data 2.1 FOMC announcements and Federal funds target rate surprises We examine the effects of Federal funds target rate changes decided upon at scheduled FOMC meetings between May 1997 and November The FOMC meets eight times a year (approximately every six weeks), such that our sample period contains 77 target rate decisions. In addition, four unscheduled meetings were held during our sample period, on October 15, 1998, January 3, 2001, April 18, 2001, and September 17, At each of those meetings the FOMC decided to lower the target rate, by 25 basis points on October 15, 1998 and by 50 basis points in each of the three meetings in We omit these unscheduled meetings from our analysis, as they have rather different effects than regular, scheduled meetings, see Bernanke and Kuttner (2005), Fleming and Piazzesi (2005), and the discussion below. Given the forward-looking nature of financial markets, it is likely that anticipated target rate changes have already been incorporated in stock prices before the actual change is announced. At least we should allow for the possibility that investors respond differently to expected and unexpected FOMC decisions. We use the methodology proposed by Kuttner (2001) to obtain a measure of the surprise in the Federal funds target rate change from the change in the current-month Federal funds futures rate on the day of the FOMC announcement. As the futures contract s payoff depends on the monthly average Federal funds rate, the change in the futures 4

6 rate is scaled up by a factor reflecting the number of remaining days in the month, which are affected by the change. In sum, we compute the unexpected target rate change ( the surprise ) S as where f 0 d S = D ( ) f 0 D d d fd 1 0, (1) is the current-month futures rate at the end of the announcement day d, and D is the number of days in the month. In order to avoid disturbances from end-of-month effects in the effective funds rate, we use the (unscaled) change in the next month s futures rate for announcements within the last seven calendar days of the month, following Bernanke and Kuttner (2005) and Fleming and Piazzesi (2005). Throughout the empirical analysis below the surprise S is measured in basis points. Appendix A shows the dates of the FOMC meetings and the changes in the Federal funds target rate together with the corresponding surprises. The most remarkable sequence of target rate changes is obviously the large decrease from 6.50% to 1.75% between January 3, 2001 and December 11, 2001, reflecting the attempts of the Federal Reserve to stimulate the US economy, followed by the two-year period of tightening monetary policy between June 30, 2004 and June 29, 2006, during which the target rate was increased from 1.25% to 5.25%. The mean (median) surprise for the 77 scheduled FOMC meetings is equal to 0.23 (0.00) basis points with a standard deviation of 4.35 basis points, suggesting that market expectations of these monetary policy actions are essentially unbiased. The distribution of surprises is somewhat asymmetric with skewness equal to The asymmetry is also reflected by the fact that the average of the 28 negative surprises at 4.06 basis points is larger in absolute value than the average of the 26 positive surprises at 3.14 basis points. A similar difference occurs for target rate increases and decreases, with the average surprise being equal to 0.32 and 0.80 basis points, respectively. More remarkable is the difference in variability of the surprises for the different types of FOMC decisions. The standard deviation of surprises in case of a target rate decrease is equal to 9.51 basis points, compared to only 2.73 and 2.56 basis points in case the FOMC decides to lower the target rate or to leave it unchanged, respectively. Admittedly these numbers are based on small samples (during our sample period, the target rate was increased (decreased) 23 (12) times), but they nevertheless suggest that uncertainty is considerably larger when monetary policy actions 5

7 are tightening than when they are loosening. The relevance and effectiveness of central bank communication has been a topic of intense recent debate, see Kohn and Sack (2004), Rigobon and Sack (2004), Ehrmann and Fratzscher (2005, 2007), Gürkaynak, Sack and Swanson (2005), and Poole (2005a,b), among others. Starting with the decision to limit target rate changes to multiples of 25 basis points in 1989, the FOMC has taken several measures to improve the transparency of its monetary policy actions, see Poole (2005b) for an overview. These include the release of a press statement at the conclusion of its meetings since February These statements included an indication of the policy bias until December 1999, which has been replaced by a balance of risks assessment since January 2000; both have generally been interpreted as being suggestive of future policy actions. In the statement related to its August 2003 meeting, the FOMC introduced more explicit forward-looking language into its press statement, indicating the probable direction of the target rate over the next meetings, as discussed in Bernanke, Reinhart and Sack (2004), Poole (2005b), and Rudebusch and Williams (in press), among others. 4 Especially the latter action appears to have been effective in improving the market s understanding of the FOMC policy actions, as the standard deviation of the target rate surprises has declined from 5.34 basis points before the August 2003 meeting to just 1.22 basis points thereafter. We refer to Poole and Rasche (2003) and Swanson (2006) for a more detailed analysis of the effects of the increased transparency of the Federal Reserve on financial market expectations. 5 A final remark about the Federal funds target rate changes concerns the four unscheduled FOMC meetings during our sample period. Appendix A shows that the target rate decisions made at these unscheduled meetings took the market almost completely by surprise, with the unexpected target rate change being almost equal 4 As discussed in Rudebusch and Williams (in press), the statements following the six meetings held between May 1999 and December 1999 also contained an explicit announcement of the expected future direction of FOMC policy. These were abandoned from January 2000 onwards, as market participants did not seem to respond well to these statements. 5 Using a Markov-Switching modeling framework, Davig and Gerlach (2006) document the presence of two regimes in the response of the stock market to monetary policy announcements. The regimes are characterized by different levels of volatility in the market response, which according to Davig and Gerlach (2006) may be related to different interpretation by the market of information in the FOMC announcements across the sample period. 6

8 to the actual change. Given that their surprise components are so different from scheduled announcements, we omit the unscheduled meetings from the analysis. For all the regressions in the empirical analysis discussed in the next section, we compute influence statistics for each observation in the remaining sample to determine whether some might have an unduly large effect on the regressions results. Reassuringly, it does not appear to be the case that any of the announcements following scheduled meetings is overly influential. 2.2 Intra-day stock returns The stock price data set consists of open, high, low, and close transaction prices at the one-minute sampling frequency for the June 2004 S&P100 index constituents for the period from April 16, 1997 until November 3, For some stocks the price series start at a later date, while for others they stop earlier. The average number of stocks for which prices are available at the time of the 77 scheduled FOMC meetings is equal to 94. From the closing prices we obtain intraday returns, volatilities and correlations. First, we compute five-minute returns for a window surrounding the FOMC announcements, from 10 minutes before until 80 minutes after the announcement. In the return regressions discussed in Section 3 we use the stock returns during the fiveminute interval immediately following the announcement as the dependent variable, following the evidence reported in the literature that the effects on price levels occur instantaneously and are short-lived, see Ederington and Lee (1995) and Andersen et al. (in press), among others. The use of accurate announcement times is therefore essential for this analysis. Although the FOMC announcements are scheduled to be released at 2:15PM, in practice the exact timing has varied by up to several minutes. We therefore make use of the announcement times reported in Fleming and Piazzesi (2005), extended until November 2006 to cover our complete sample period. For volatilities, it has been found that announcement effects are more sluggish and last longer, see Ederington and Lee (1993) and Fleming and Remolona (1997), among others. For that reason we focus on the 60-minute window between 10 minutes before the announcement until 50 minutes thereafter, following Fleming and Piazzesi 7

9 (2005). 6 In particular, we compute the realized volatility (RV) for the announcement on day t as RV t = 10 Rt,k 2, (2) k= 2 where R t,k is the five-minute (market, sector, or individual stock) return on day t in interval k and the interval k = 0 corresponds to the first five minutes after the announcement. Similarly, the realized correlation (RC) between stocks (or sectors) i and j for the announcement at interval t is computed as RC ij,t = 10 k= 2 R i,t+kr j,t+k RV i,t RV j,t, (3) where RV i,t and RV j,t are the realized volatilities for stocks i and j computed according to (2). Finally, there is ample empirical evidence that the level of volatilities and correlations changes over time. As this may obscure the results on the effects of Federal funds target rate changes, we also compute volatilities and correlations for the same window on the five trading days before the announcement, and consider the change in volatilities and correlations on the day of the FOMC announcement relative to their original level. 3 Methodology and empirical results This section consists of three parts, analyzing the effects of FOMC target rate decisions on stock returns, on volatilities, and on correlations, respectively. In each case both the employed methodology and the results are presented. The regressions below are implemented at the individual stock level, the sector level, and the market level. The sector and market returns are equally weighted returns of the individual stocks. To save space, we present detailed results obtained for the market and sector returns, which enable us to assess the aggregate effects of target rate changes on the market as a whole and any differences therein across sectors. Results for the individ- 6 We examine the robustness of our results with respect to the length of this window, by repeating the complete analysis with volatilities and correlations over 30- and 90-minute windows (both starting 10 minutes before the announcement) as well. In general we find that results and conclusions are robust to the choice of window. Detailed results are available upon request. 8

10 ual stocks are occasionally presented to illustrate the pervasiveness of the response of stock prices to monetary policy actions The effect of FOMC meetings on stock returns Table 1 shows means and standard deviations of the five-minute returns of individual stocks aggregated to the market and sector levels following the Federal funds target rate announcements. The first column shows the averages taken over all 77 scheduled FOMC meetings. The overall average returns are negative and significantly different from zero at the 1% level for the market and for all sectors. In order to gain more insight in the cause for these negative returns, we split the sample into several subgroups defined by the nature of the target rate change and the surprise component. - insert Table 1 about here - The second and third columns of Table 1 show the average return for those meetings at which the FOMC decided to increase (23 times) or decrease (12 times) the target rate. In most cases the average return following a target rate decrease is positive as expected, but they are significant only for the market and the sectors Energy and Financials (at the 5% level). The average returns following a target rate increase generally are significant, but they have the wrong positive sign for both the market as a whole and for most sectors. After all, an interest rate increase implies a higher discount rate and hence should result in lower stock prices. For Health Care and Industrials sector returns we do observe a significantly negative return. Note that the fact that average returns generally are positive following both rate increases and decreases together with the negative average returns across all announcements implies that the average return following FOMC decisions to leave the target rate unchanged is negative and large. The final two columns show the results for positive and negative surprises separately. Now we do observe the expected signs for the average returns. A positive surprise means that the target rate was increased more or decreased less than the 7 Detailed results for the individual stocks can be found in the supplemental Appendix to this paper, available at 9

11 market anticipated. Hence the market underestimated the future discount factor and we see a significantly negative reaction in the stock returns following the announcement of the FOMC decision. Similarly, a negative surprise implies that the target rate ended up lower than expected. This is good news for stocks, which duly show a significant positive return. Also note that the response to negative news for stocks is larger than to positive news. After positive target rate surprises the average five-minute return is 20.1 basis points compared to 10.6 basis points after negative interest rate surprises. Moreover, this difference is statistically significant, with the t-statistic for testing the equality of the absolute values of the average response being equal to 9.3. We observe significantly negative and positive returns following positive and negative surprises for all sectors as well, although there is considerable heterogeneity in magnitude. Average returns are largest for Financials, IT and Consumer Discretionary, while they are smallest for Utility stocks. In sum, the average returns indicate that the stock market is efficient to the extent that expected target rate changes are already incorporated into stock prices prior to the FOMC meetings. Only surprises lead to a significant reaction of stock prices. This is investigated more rigorously using regression analysis below. We examine the speed at which the market responds to the FOMC announcements by computing the average returns in five-minute intervals around the time of the news release, as shown in Figure 1. The positive return for the first five minutes after the announcement for negative surprises, and the negative return during the same interval for positive surprises can also be found in the first row ( All stocks ) and the final two columns of Table 1. As expected the response to interest rate changes occurs mostly in the first five-minute interval after the announcement. Interestingly, in the case of a positive surprise, that is, negative news for stocks, the next five-minute interval also shows a negative average return. We also observe that the average returns between five and 20 minutes after the announcement are negative for both negative and positive target rate surprises. This suggests some sort of overreaction of stock prices to positive news and underreaction to negative news. - insert Figure 1 about here - The summary statistics in Table 1 suggest that only surprises in the FOMC target rate decisions affect stock prices. We continue with running three regressions, 10

12 intended to confirm this formally and to examine whether the magnitude of the surprise is important. The first regression relates the stock return to the actual target rate change, R t = α + β F F t + ε t, (4) where R t is the return in the five-minute interval following the news release, and F F t is the actual change in the Federal funds target rate, both measured in basis points. The second regression uses the surprise, R t = α + βs t + ε t, (5) where S t is the surprise (in basis points) measured by means of the change in the Federal funds futures rate, as defined in (1). Finally we consider the regression that includes both the surprise and the expected change, following Bernanke and Kuttner (2005), R t = α + βs t + γ( F F t S t ) + ε t. (6) As the variance of stock returns changes over time, we use heteroskedasticity-consistent standard errors to assess the significance of the coefficients in these regressions. Results for market and sector returns are presented in Table 2. - insert Table 2 about here - The results overwhelmingly confirm that surprises matter while the actual target rate changes do not. Consider the regression of the five-minute post-announcement market return on the surprise and the expected target rate change as given in (6). The coefficient estimates imply a five-minute return of 45 basis points following a surprise of 10 basis points in the target rate change, corresponding quite closely with estimates reported by Bernanke and Kuttner (2005) based on daily returns. The coefficient of the expected rate change is positive at 0.12 but not significant. This is consistent with market efficiency in that stock prices fully reflect expected interest rate changes, and react to new information only. The regression R 2 of (5) and (6) equal 0.27 showing that while the target rate surprises undoubtedly are important for explaining the stock returns following the FOMC announcements, they still leave a large part of the variance of returns unaccounted for. The estimation results for sector returns reveal considerable differences in the effects of target rate 11

13 surprises across sectors, cf. Ehrmann and Fratzscher (2004). The IT and Financials sectors show the strongest response, with the estimate of the surprise coefficient in (6) being almost one and a half times larger than the market average. For the Financials sector the strong reaction to unexpected interest rate changes is not surprising. The large effects on the IT sector can perhaps be explained by the fact that during our sample period IT firms are largely built on future earnings instead of current earnings. Hence the discount factor would have a larger effect on the value of IT stocks. The R 2 is also higher for the IT sector at 0.32, showing that the target rate surprises are relatively more important for this sector. The Utilities sector responds the least to the FOMC decisions, with the surprise coefficient being equal to 2.56, while surprises also explain only 17% of the variation in the postannouncement returns. The target surprises also have relatively little explanatory power for the Energy and Telecommunications returns. Our results for the sector returns are in partial agreement with Ehrmann and Fratzscher (2004). We confirm their finding that non-cyclical sectors such as Utilities, Consumer Staples, Health Care, and Energy are less responsive to target rate surprises than the aggregate market. On the other hand, we do not find that cyclical and capital-intensive sectors such as Telecommunications, Consumer Discretionary, and Industrials are more responsive than the market average. Conversely, while we find that Financials is the sector most sensitive to target rate surprises, Ehrmann and Fratzscher (2004) find its reaction to be very close to the market average. The pervasiveness of the response of stock prices to target rate surprises can be seen from the regression results for the 100 individual stocks. Based on (5), the estimate of the surprise coefficient is significantly negative at the 1%, 5% and 10% levels for 71, 19, and 7 stocks, respectively. The majority of the individual betas, which are shown in Figure 2, are between 3 and 5, but some are even below 8. We refer to Ehrmann and Fratzscher (2004) for a detailed analysis of the possible determinants of the heterogeneity in response to target rate surprises across firms. In addition to the industry effects discussed above, Ehrmann and Fratzscher (2004) document that small firms and financially-constrained firms respond significantly more to monetary policy surprises than large and less constrained ones. 8 8 Using decile portfolios, Guo (2004) finds that the size effect in the responsiveness to monetary policy only occurs during economic recessions but not during booms. 12

14 - insert Figure 2 about here - Given the above results, we focus on the effects of surprises in the target rate decisions in more detail. In particular, we examine the presence of different types of asymmetries in their effects on stock prices. Several such asymmetries have been suggested in the literature, including the possibility that the response of stock prices depends on the surprise being positive or negative or on the direction of the actual target rate change (Lobo, 2000; Bernanke and Kuttner, 2005), or on the phase of the business cycle (Guo, 2004; Andersen et al., in press). 9 The average returns in Table 1 suggest that these asymmetric effects may also be relevant for the S&P100 constituents. We extend the regression in (5) to assess the empirical evidence for these asymmetries more formally. For example, we test whether the post-announcement return is different following positive and negative surprises, and whether only the occurrence of a surprise matters or also its magnitude by means of the regression R t = α 0 + (α 1 + β 1 S t )D(S t < 0) + (α 2 + β 2 S t )D(S t > 0) + ε t, (7) where D(A) is a dummy variable taking the value 1 if the event A is true and 0 otherwise. The results for this regression are presented in Table 3 for the market and sector returns. - insert Table 3 about here - We observe a marked difference between positive and negative surprises. The estimate of β 1 is significant at the 1% level for both the market return and all sector returns, while α 1 is not significant except for the IT and Telecommunications sectors. This implies that for negative surprises, meaning good news for stocks, the magnitude of the surprise is much more important than the fact that the actual target rate decision is different from market expectations. In contrast, in the case of positive target rate surprises, or bad news for stocks, it is not possible to determine whether the fact that there is a surprise or its magnitude is most important, as α 2 and β 2 are not significant individually, with a few exceptions. Note that this should 9 Fleming and Piazzesi (2005) document an asymmetric response of Treasury note yields depending on the slope of the yield curve. We find no such asymmetry for the S&P100 stocks and hence do not consider this further. 13

15 not be interpreted as saying that stocks do not respond to positive surprises at all. In fact an F-test shows that jointly α 2 and β 2 are significant for the market and all sectors except Energy and Financials. In addition, it is useful to note that the coefficient estimates imply that the response to negative news is in fact larger than the effects of positive news for realistic magnitudes of the target rate surprise of less than eight basis points, which occurs for more than 90 percent of the observations in our sample. For example, an unexpected increase of the target rate with 3.6 basis points (which is the mean absolute surprise across the non-zero surprises) provokes a negative market return of 21.2 basis points, compared to a positive return of only 8.0 basis points following an unexpected decrease of the target rate of the same magnitude. The asymmetric effect of positive and negative surprises is confirmed by the regressions for the individual stocks, see Figure 3. The estimate of β 1 in (7) is significant at the 1%, 5% and 10% levels for 74, 11, and 6 stocks, respectively, and negative in all 100 cases. The corresponding numbers for β 2 are 0, 7, and 16. The difference between β 1 and β 2 is negative for all stocks, with the average being equal to 3.2 with a t-value of 9.9. A final remark about the results in Table 3 concerns the estimate of α 0, representing the average return in case the surprise target rate change is zero. Its insignificance confirms the notion that stock prices are not affected when the actual target rate change corresponds with the market s prior expectations. - insert Figure 3 about here - The above results differ markedly from the findings of Bernanke and Kuttner (2005), who document weak support at best for asymmetric effects of positive and negative surprises based on daily stock returns. This demonstrates the advantage of using intraday data. Repeating the regression analysis using daily returns, we find that also at the daily level stock prices respond to surprises but not to expected target rate changes. The regression R 2 s, however, are substantially lower, that is, noise contaminates the analysis when using daily data. For example, using daily market returns as dependent variable, the R 2 of (6) is 0.10 only. We also find some asymmetry, with α 2 being significant (such that in case of negative news for stocks only the fact that there is a surprise matters), but in general the much lower significance levels when using daily data make it difficult to draw the same clearcut 14

16 conclusions as we can do here based on intraday data. 10 Our findings for the other two types of asymmetries are much less supportive. 11 First, we do not find any significant asymmetry in the effects of surprises depending on whether the target rate is increased or decreased or left unchanged. Bernanke and Kuttner (2005) reach a similar conclusion with respect to the effects for rate increases and decreases, but do find that the market responds very little, if at all, to policy inactions. Second, we find no evidence for different effects of target rate surprises in business cycle recessions and expansions. Using the official NBER turning points according to which the recession period is April 2001 through November 2001, we find no support for asymmetric business cycle effects. Possibly this is due to the small number of observations during the recession period. However, when we follow Andersen et al. (in press) and define the start (end) of a recession when there are three consecutive monthly declines (increases) in nonfarm payroll employment leading to a recession from March 2001 through December 2002, we do not find more evidence for business cycle asymmetry. This is in contrast to Andersen et al. (in press), who find that equity markets only react according to expectations to target rate surprises during recessions, while during expansions the response is not significant. 12 Possibly this is due to our use of a longer sample period and, in particular, the inclusion of the recent period during which the FOMC has attempted to improve its transparency. 3.2 The effect of FOMC meetings on stock volatilities We now turn our attention to the impact of FOMC target rate decisions on stock return volatility. As discussed in Section 2, we measure volatility by means of the realized volatility (RV) using the popular five-minute frequency for the 60-minute window starting 10 minutes before the target rate decision is made public. In order to control for variation in the level of volatility over time for reasons other than the FOMC announcements, we consider the difference between RV on the announcement 10 Detailed results for daily returns are given in Tables 1 and 2 of the supplemental Appendix. 11 Detailed results are not shown here to conserve space but can be found in the supplemental Appendix. 12 Andersen et al. (in press) document even more pronounced business cycle effects in asset prices response to news concerning real activity, with bad news unexpectedly having a positive impact during expansions and the expected negative impact during recessions; see also Boyd et al. (2005). 15

17 day and the day before insert Table 4 about here - Table 4 clearly demonstrates that FOMC news leads to a substantial volatility increase for the S&P100 constituents. For the equally-weighted market return, the average change in realized volatility taken over all 77 scheduled FOMC meetings equals 32.4 basis points, with a standard deviation of As expected, among the different sectors Financials experiences the largest volatility change at 46.6 basis points, closely followed by IT and Telecommunications. Again it seems that the impact of FOMC announcements is smallest for the Utilities and Energy sectors. The second and third columns show that the average volatility change is considerably larger following target rate decreases than increases. For the market as a whole, on days with target rate decreases the change in volatility equals 62.8 basis points on average, compared to 30.9 on days with target rate increases. This difference may be related to the asymmetry mentioned in Section 2 that the average surprise after target rate decreases equals 0.80 basis points compared to 0.32 basis points after target rate increases. In addition, all 12 target rate decreases happened prior to August 2003 when FOMC introduced more explicit forward-looking language into its press statement resulting in a decline of the standard deviation from 5.34 basis points before August 2003 to 1.22 basis points after August Related to this, we also note that the average level of realized volatility after the 17 target rate increases that occurred after August 2003 is substantially smaller (38.4 basis points) than following the six target rate increases that occurred before (78.4). From the final two columns of Table 4 we observe that for FOMC decisions which surprised the market the change in volatility is larger than when considering all announcements. This holds for the market return as well as for all sectors. There does not appear to be a large difference between negative and positive surprises when looking at these average volatility changes. For example, the average change 13 Detailed results for the level of volatility are available in the supplemental Appendix. It is also useful to remark that Bomfim (2003) documents the presence of a so-called calm-before-thestorm effect, that is, volatility tends to be depressed on pre-announcement days. This does not affect the results of our analysis, as we obtained qualitatively and quantitatively similar results when using the difference between RV on announcement days and the average RV on the previous five trading days. 16

18 in market volatility considering all 77 FOMC meetings is equal to 32.4 basis points, whereas it is higher at 41.7 after the 28 positive surprises and 40.6 after the 26 negative surprises. From these numbers it also follows that the average change in volatility in case the FOMC announcement does not contain a surprise is equal to 12 basis points only. To examine the persistence of the increase in volatility after announcements with surprises we compute the squared market returns in five-minute intervals around the time of the release of the FOMC target rate decisions. From Figure 4, we observe that the increase in volatility after surprises only lasts for 5 to 15 minutes. The fact that the average squared return immediately following the news is largest after negative surprises does not contradict Figure 1. The smaller average positive return is based on more extreme positive and negative returns for single meetings than the higher negative return after positive surprises. We therefore discuss the asymmetry based on our regression analysis presented in Table 6 below. - insert Figure 4 about here - We examine the effects of target rate changes and surprises on stock volatility by means of regressions similar to the ones used for stock returns in the previous subsection, except that for obvious reasons we now use absolute target rate changes and surprises. Specifically, we regress the daily change in realized volatility on the absolute actual target rate change, on the absolute surprise, and on the absolute surprise and the absolute expected target rate change, that is RV t = α + β F F t + ε t, (8) RV t = α + β S t + ε t, (9) RV t = α + β S t + γ F F t S t + ε t, (10) where RV t denotes the difference between realized volatility during the 60-minute window starting 10 minutes before the target rate decision is made public and during the same window on the day before. The results in Table 5 show that the stock volatilities respond to the (absolute) surprise and not to the expected target rate change. We do find significantly positive estimates of the actual target rate change coefficient in (8) for the market and nine 17

19 sector returns, but splitting this into the expected target rate change and the surprise as in (10) shows that this is due entirely to the latter component. The estimates of the surprise coefficients are positive and highly significant, in both (9) and (10). The estimate of β in (9) for the market RV implies that an unexpected target rate change of 3.6 basis points (the mean absolute surprise across the non-zero surprises) leads to an increase of volatility by 21.1 basis points during the 60-minute window around the announcement. Financials, as expected, show the strongest response to surprises with the estimated coefficient for S t being equal to 8.2 in (10), considerably larger than the market average. The Utilities sector forms the other extreme, with a surprise coefficient in (10) of 1.9 only. These conclusions are confirmed by the regressions for the realized volatilities for the individual stocks. The estimate of β in (9) is positive for all stocks and significantly different from zero at the 1% significance level for 82 stocks, with another 9 and 5 coefficients being significant at the 5% and 10% significance levels, respectively. The majority of the individual betas are between 3 and 10, see Figure 5, showing that there is considerable heterogeneity in the effects of target rate surprises on individual stock volatility. - insert Table 5 and Figure 5 about here - Given these results, we continue with analysing the effects of surprises on volatility in more detail, again focusing on possible asymmetries. In particular, the volatility literature strongly suggests that the response of volatility to negative news is stronger than the response to positive news. This can conveniently be examined by the regression RV t = α 0 + (α 1 + β 1 S t )D(S t < 0) + (α 2 + β 2 S t )D(S t > 0) + ε t, (11) where D(A) equals 1 if the event A is true, and 0 otherwise. The results are presented in Table 6, both for market-wide and sector volatility. Again we observe marked differences between positive and negative surprises. In the case of negative surprises, representing good news for stocks, the magnitude of the surprise is far more important than the occurrence of a positive surprise, as β 1 is significant for both the market return and all sector returns at the 1% level, while α 1 is not significant at all. In contrast, in the case of positive surprises, thus bad news for stocks, both the fact that there is a negative surprise and the magnitude of the surprise 18

20 are relevant, as both α 2 and β 2 are significant for the market and for most sectors. The magnitude of the estimates of α 2 also suggest that negative news leads to a sizable jump in volatility, irrespective of the magnitude of the surprise. The results for the realized volatilities of the individual stocks confirm that negative news has a larger effect on volatility than positive news. This asymmetry shows up in both the effects of the mere occurrence of good or bad news and in different responses to the magnitude of the surprise. Across all 100 stocks, the average response to positive surprises (β 2 ) is equal to 3.7 compared to an average response coefficient to negative surprises (β 1 ) of 5.7, see also panel (b) of Figure 6. The average difference of 2.0 is significantly different from zero at the 1% level. Hence, at first sight it may seem that individual stocks in fact respond more strongly to positive news than to negative news. This is not the case, however, as it is more than compensated for by differences in the coefficients α 1 and α 2, shown in panel (a) of Figure 6. The average estimates of these coefficients equal 7.3 and 20.9, respectively, showing that the occurrence of negative news leads to a substantially larger increase of volatility. The average difference between α 2 and α 1 of 13.6 is highly significant. Overall, the average coefficient estimates in (11) imply that the response of stock volatility is stronger to negative news for surprises less than seven basis points and stronger to positive news for larger surprises. - insert Table 6 and Figure 6 about here - For the other possible asymmetries that we examine, 14 we do observe some differences in the response of volatility to FOMC surprises depending on the type of target rate change, but according to Wald tests these generally are not significant. This is in contrast to Lobo (2000), who documents moderate evidence for an increase in aggregate stock market volatility following target rate cuts but not following rate hikes. Allowing for different news effects over the business cycle, we find more pronounced asymmetries. In particular, the average volatility change following an FOMC announcement is larger during recessions, while the surprise coefficient is larger during expansions. The differences are significant for market volatility and for quite a few sectors. Hence, this suggests that FOMC announcements lead to a larger jump in 14 See the supplemental Appendix for detailed results. 19

21 volatility, independent of the magnitude of the target rate surprise during recessions, while the magnitude of the surprise is more important during expansions. 3.3 The effect of FOMC meetings on stock correlations We next consider the impact of FOMC meetings on intraday realized correlations during the 60-minute window around target rate announcements. The below-diagonal part of panel A of Table 7 shows the average realized correlations between the sector returns on announcement days, while the above-diagonal elements are the average changes relative to the day before. On announcement days, the inter-sector correlations range between 0.53 for Energy and Utilities and 0.84 for Financials and IT, with the average correlation being equal to All 45 inter-sector correlations are higher on announcement days, with an average increase relative to the day before of This is considerably lower than the average change of 0.27 when the target rate is increased and 0.31 when it is lowered, see panel B of Table 7. Hence, on days that the target rate is not moved the change in correlation during the hour around the news announcement is substantially smaller. Similarly, on days with a nonzero target rate surprise the correlation change is also higher than average, at 0.27 and 0.23 after positive and negative surprises, respectively, see panel C of Table 7. Note that positive target rate surprises (negative news for stocks) have a larger impact on correlations than negative surprises. This is intraday evidence of a general belief (see Longin and Solnik, 2001, among many others) that at times of negative news correlations increase. A final observation that is made from Table 7 is that the Energy and Utilities sectors show the least co-movement with other sectors as they have lower average correlations. At the same time, the increase in correlations on FOMC announcement days is largest for pairs involving one of these two sectors. - insert Table 7 about here - We also compute correlation changes between individual stocks, and aggregate these to within sectors and between sectors. Generally the realized correlations between individual stocks are somewhat lower than the ones shown in Table 7. The reason is that sector returns are more diversified and hence have less specific risk. A higher specific risk will lower the correlation Detailed results are not shown here, but are available upon request. 20