Exceeding Expectations: Economic Forecasts and Underreaction to Macroeconomic Announcements

Exceeding Expectations: Economic Forecasts and Underreaction to Macroeconomic Announcements Gene Birz Sandip Dutta* This paper was previously circulated under the title Exceeding Expectations: Economic Forecasts, Anchroing Bias and Stock Returns. We thank Thomas C. Howard, Dennis Lasser, and Avanidhar Subrahmanyam, as well as participants at the 2014 meetings of Academy of Behavioral Finance and Economics for helpful comments. *Department of Economics and Finance; Southern Connecticut State University; Birz: birzg1@southernct.edu ; Dutta: duttas2@southernct.edu 1

Exceeding Expectations: Economic Forecasts and Underreaction to Macroeconomic Announcements Abstract Existing research relies on the assumption of rational behavior and argues that only unexpected macroeconomic news may affect stock prices. Unlike these papers, we find significant stock price effects of expected components of macroeconomic announcements, which are revealed by survey forecasts of market participants. Since economic forecasts are systematically biased towards the values of previous announcements, we hypothesize that the historical information embedded in these forecasts may impact stock prices. Our results confirm this hypothesis and suggest that some investors trade on historical macroeconomic announcements. 2

1. Introduction In the past several decades, a large body of research has examined the effect of macroeconomic announcements on the stock market. While these papers differ from each other in terms of the macroeconomic factors and time periods they examine, most of them use a similar statistical model. Mainly, stock prices are regressed on the unexpected component of macroeconomic announcements, i.e., the difference between the released statistic and its expected value, which is revealed earlier by survey forecasts of financial market participants. Pearce and Roley (1985) were one of the first to implement this approach. They argued that while, theoretically, both the expected and the unexpected components of economic announcements could affect stock prices, the empirical evidence of their study supported the Efficient Market Hypothesis (EMH) and showed that only unexpected economic news affected the S&P 500 returns during the 1977-1982 period. Since the Pearce and Roley study, research papers have only looked at the relationship between stock prices and the unexpected component of macroeconomic announcements, thereby relying on the assumption of efficient markets. However, recent research in behavioral finance has shown that the behavior of stock prices is not always consistent with EMH. 1 In particular, a large subset of this research examines stock price effects of public announcements and provides empirical evidence that stock prices may not quickly adjust to all available information [Bernard and Thomas (1989, 1990), Birz (2014), Gilbert et. al (2012), Huberman and Regev (2001), Tetlock (2011)]. In light of this evidence, we propose to examine whether stock prices are impacted by expected macroeconomic announcements revealed by survey forecasts of market participants. To 1 Hirshleifer (2001) and Barberis and Thaler (2003) provide a detailed literature review of research on market efficiency and behavioral finance. 3

test our hypothesis, we implement the unrestricted model proposed by Pearce and Roley (1985, p. 62). We find a strong, statistically significant relationship between the S&P 500 returns and the survey forecasts of macroeconomic statistics collected by Money Market Services (MMS), a popular source of economic forecasts in existing literature. We believe that there are two possible explanations for our finding. On the one hand, due to the fact that these forecasts are made public before the official releases of the economic statistics, these forecasts may constitute the first public signal about changes in economic conditions. As such, the forecasts may contain new information, and therefore, their impact on stock returns may be consistent with rational behavior. On the other hand, our finding may mean that investors are prone to systematic cognitive biases causing them to trade on expected or historical components of economic statistics. Previous research finds that the survey forecasts of economic statistics contain a systematic anchoring bias, i.e., these forecasts are very similar to the macroeconomic statistics released in the past (for example, see Campbell and Sharpe (2009) and Hess and Orbe (2013)). 2 These papers argue that the surveyed market participants put too much weight on past economic statistics and too little weight on new information when making forecasts of future economic statistics. Therefore, we hypothesize that some market participants trading on macroeconomic information may have a similar bias wherein they underweight new announcements and tend to rely more on recent historical information. Consequently, our findings of stock price effects of economic forecasts may mean that some investors trade on past economic information embedded in these forecasts and that the stock market underreacts to macroeconomic announcements. In light of these hypotheses, our second objective is to determine whether stock price 2 Tversky and Kahneman (1974) were one of the first studies in experimental psychology that describe various cognitive biases including anchoring. 4

effects of survey forecasts are driven by a behavioral bias or whether market participants trade on new information included in these forecasts. We do this in three ways. First, if forecasts contain new information, we should see the largest stock price responses as soon as these forecasts become public. By the time of the official macroeconomic announcements, which are released five days after the release of the forecasts, the relevant new information embedded in the forecasts should be already reflected in stock prices. In contrast, we find that stock price effects of economic forecasts are statistically equal to zero during the first three days since they become public followed by statistically significant effects the day before and the day of the macroeconomic announcements. This finding suggests that it is unlikely that investors specifically trade on the forecasts of economic statistics and that stock price effects of these forecasts are related to the arrival of new information. In our second test, we start with confirming previous research and find that in our sample, forecasts of future macro statistics are anchored to macroeconomic statistics released in the recent past. In other words, we show that these forecasts, to a large extent, contain old information released in the past. We then econometrically split the value of each forecast into the component formed as a result of anchoring on past statistics and the component formed as a result of new information. Subsequently, we compare both components to stock returns in the framework of the Pearce and Roley (1985) unrestricted model. We find a statistically and economically significant relationship between the S&P 500 returns and the historical component of the forecasts, which suggests that some investors may overweight and trade on past announcements embedded in these forecasts. Finally, if stock price effects of economic forecasts imply that some investors trade on past economic announcements, stock prices should also correlate with values of these past 5

announcements. In our third test, we find statistically and economically significant correlations between the S&P 500 returns and the average values of the previous three announcements. A large body of research on firm-specific events often cites either over- or underreaction as the reason for stock price effects of stale information. The empirical difference between these two hypotheses is the return reversal, which only happens in the case of overreaction [Brav and Heaton (2002, p. 585)]. We do not find the reversal of the stock price effects of past economic announcements, which suggests that the stock market underreacts to these announcements. Throughout the paper, we focus on monthly unemployment rate and Nonfarm Payroll Employment releases. We do this for two reasons. First, employment announcements are the most influential announcements for asset prices among all macroeconomic releases [Andersen and Bollerslev (1998), Andersen et al. (2007), Carnes et al. (1991), Lahaye et al. (2011)]. The second reason is the schedule of employment announcements, which allows us to determine the mechanism through which forecasts affect stock returns. The Bureau of Labor Statistics (BLS) releases both the unemployment rate and Nonfarm Payroll Employment statistics on the first Friday of each month. Economic forecasts of these statistics are made public one week before the official announcements of these data. As mentioned above, the five-day difference allows us to determine whether investors respond to new information provided by economic forecasts or whether some investors trade on historical information embedded in the forecasts. In summary, our paper provides two significant contributions to the existing literature. First, we contribute to research on asset price effects of macroeconomic news, which currently only documents the stock price effects of unexpected economic announcements [Andersen et al. (2007), Boyd et. al. (2005), Chen, Roll, and Ross (1986), Jain (1988), McQueen and Roley (1993), Pearce and Roley (1985)]. To our knowledge, we are the first to find that the expected 6

component of economic announcements could also affect stock prices. Our second contribution is to the literature on investor biases and asset prices. Specifically, our results are consistent with underreaction a tendency for stock prices to only gradually respond to the arrival of new information. Many researchers argue that it is hard to reconcile underreaction with rational behavior [Daniel et. al. (1998) and Barberis et al. (1998)]. For example, relying on research in experimental psychology, Barberis et al. (1998) develop a theoretical model to show that underreaction is caused by conservatism a tendency for individuals not to update their beliefs as much as a rational Bayesian agent would in the event of new information [Edwards (1968)]. In other words, conservatism is a tendency for individuals to undervalue new information and overvalue past information. The existing empirical evidence of underreaction includes post-earnings announcement drift [Bernard and Thomas (1989, 1990)], stock price responses to various past firm-specific events [Ikenberry et al. (1995), Loughran and Ritter (1995), Michaely et al. (1995)], as well as momentum and autocorrelation in stock returns [Jegadeesh and Titman (1993)]. However, most of this research focuses on underreaction to firm-specific events. We contribute to this literature by documenting the stock market s underreaction to macroeconomic announcements. 3 Finally, this paper is also related to Campbell and Sharpe (2009) and Hess and Orbe (2013). Campbell and Sharpe (2009) document the anchoring bias of economic forecasts. However, their focus is to examine whether this bias affects the bond market. They find that the anchoring bias does not impact bond prices. The authors also employ a different statistical model as they assume that bond traders are initially rational and only trade on the unanticipated 3 This paper is also related to Birz (2014) and Gilbert et al. (2012) who study whether stale economic information may affect stock prices. Birz (2014) finds that stock prices are impacted by stale economic news reported in newspapers. Gilbert et al. (2012) find asset price effects of Leading Economic Index (LEI) announcements, which may contain stale information. However, while Birz (2014) and Gilbert et al. (2012) also focus on economic news, unlike this study, they find evidence of overreaction. 7

component of the announced statistics. Hess and Orbe (2011) also find anchoring in economic forecasts. However, they do not focus on whether anchoring affects financial markets. The focus of their paper is to determine whether economic forecasts can be improved by adjusting the bias. The remainder of the paper is organized as follows. Section 2 describes our data and methodology. We report our empirical results in Section 3. Section 4 provides further discussion of our findings. Finally, Section 5 concludes. 2. Data and Methodology 2.1 Data We focus on the unemployment rate (UN) and Nonfarm Payroll Employment (NFE) releases, which existing research consistently finds to be important for the stock market. The official reports on these variables are released at 8:30 AM on the first Friday of each month. The expectations on macroeconomic releases are collected by MMS, the most popular source of economic forecasts in existing literature. Every Friday, MMS surveys market participants on the variables scheduled to be released during the following week. The surveys are conducted by phone in the morning and the results of the surveys become available in the late afternoon. Table 1 provides summary statistics on the economic releases, their MMS forecasts, and macro surprises. Table 2 reports correlations among surprises and forecasts, which are among our explanatory variables. The time series of closing prices on the S&P 500 were obtained from CRSP for the period between January 1, 1992 to December 31, 2008. In our 17-year sample, four UN and NFE release days coincided with Good Friday, during which the stock market was closed. 8

Consequently, the sample consists of 200 observations. 4 2.2 Model Existing studies rely on the assumption of efficient markets and only study whether unexpected economic announcements affect stock prices. We examine whether stock prices are affected by all macroeconomic information, including expectations of economic announcements revealed by previously released survey forecasts. To test our hypothesis, we use the following specification: R t = b 0 + b 1 F 1t + b 2 S 2t + e t (1) where Rt denotes the difference between the closing price on the S&P 500 at time t, the release day of the macroeconomic statistics, and the closing price at time t-1. F1t is the expected value of each macroeconomic statistic, revealed by the median survey response of financial market participants. 5 Since the examined statistics on the unemployment rate and nonfarm employment are measured in different units, we normalize each expectation by dividing it by its sample standard deviation. S2t, often referred as surprise, is the difference between the released statistic and its expected value. We calculate surprises as in Andersen, Bollerslev, Diebold, and Vega (2003, 2007). First, we calculate the difference between the values of released statistics and their expected values. Second, we divide this difference by its sample standard deviation. Therefore, the surprise for announcement k is 4 The following days of unemployment and NFE releases coincided with Good Friday: April 1, 1994; April 5, 1996; April 2, 1999; April 6, 2007. 5 Ft = E (At ); Ft is a survey forecast of the economic announcement, At. These forecasts are made public 5 days before the official releases of the unemployment rate and the nonfarm employment statistics. 9

S k,t = (A k,t F k,t ) σ k (2) where Ak,t is the value of the released statistic k, Fk,t is the expected value of the statistic k provided by MMS forecasts, and σk denotes the sample standard deviation of Ak,t - Fk,t. Moreover, because of a small sample size, we compute bootstrapped standard errors for all regressions in the paper in order to provide more accurate statistical inference (see Efron (1979)). Some researchers examine intraday stock prices to show that the impact of economic surprises depends on when they occur during the trading day [Andersen, Bollerslev, Diebold, and Vega (2007)]. We do not follow this methodology for two reasons. First, both of our employment variables are released at the same time 8:30 am. This means that there should not be any difference in the impact of these variables on stock returns due to the time of their release. More importantly, the research goal of this paper is not to identify or measure the impact of surprises, but to find out if investors trade on recent past macro announcements that constitute the survey forecasts of economic statistics. Consequently, we are only interested in examining the relationship between forecasts and closing stock returns. Moreover, some researchers argue that stock price effects of macroeconomic news depend on the state of the economy [McQueen and Roley (1993)]. To show this, they regress stock returns on the interactions between economic surprises and different stages of the business cycle. Since the goal of this study is not to examine the impact of surprises on stock returns, but to find out if past economic announcements affect stock prices, we do not follow this methodology. 3. Results 3.1 Stock Returns and Survey Expectations of Economic Data We begin our analysis by implementing the unrestricted Pearce and Roley (1985) model 10

the model without the assumption of EMH to examine whether stock prices are affected by the expected economic statistics forecasted by financial market participants. The results are shown in Table 3. In regressions (1) and (4), the S&P 500 returns on the day of the release of macro statistics are regressed on economic surprises and the survey expectations of economic statistics. As Table 2 shows, surprises and forecasts are not highly correlated, therefore, multicollinearity is not an issue in this analysis. The coefficients on the expectations of both variables, i.e. the coefficients of interest in this study, have expected signs and are both statistically and economically significant. For example, a one-standard deviation increase in the expected unemployment rate decreases S&P 500 returns by 16.7 basis points (bps). The coefficient is statistically significant at the 5 % level. Similarly, a one-standard deviation increase in the expected nonfarm payroll employment increases S&P 500 returns by 17.3 bps and the estimate is also statistically significant at the 5 % level. These results are also robust to standard time series controls. In regressions (2) and (5), we include two lags of S&P 500 returns to control for residual autocorrelation, a January dummy to control for the January effect, and a Friday dummy to control for the weekend effect. We also ensure that our main results are not driven by past return-volume interactions as in Campbell, Grossman and Wang (1993). Therefore, we include two lags of S&P 500 returns multiplied by the detrended natural logarithm of S&P 500 volume for the same time period as part of the control variables. The results with respect to stock price effects of expected economic factors are even stronger after we include these control variables. For example, a one-standard deviation increase in the expected unemployment rate decreases S&P 500 returns by 20.2 bps and the coefficient is statistically significant at the 1 % level. 11

In regressions (3) and (6), we check that our results are not impacted by a small number of outliers. We follow previous research and use Huber (1981) M-estimator that produces robust estimates in the presence of outliers [for example, see Garcia (2012)]. As in previous regressions, we find a statistically and economically significant relationship between the expected economic statistics and the S&P 500 returns on the days of macroeconomic announcements. All regressions in Table 3 show that the effect of surprises on the daily S&P 500 returns are statistically insignificant. While the goal of this study is not to examine the impact of surprises, these results are not inconsistent with existing research, which finds that the effect of economic surprises is short-lived. In particular, many studies that examine daily stock returns find statistically insignificant effects of surprises. However, those that look at intraday returns find that stock prices adjust to surprises within an hour. Therefore, our findings in Table 3 are consistent with findings in existing literature with respect to the daily stock price effects of economic surprises. 6 The fact that we find that expected economic statistics affect daily stock prices, perhaps, suggests the existence of different investor clienteles. The findings of existing papers suggest that there are some investors, probably more sophisticated ones, that trade on surprises right after the release of macroeconomic statistics. Our findings, while do no refute the findings of these papers, suggest that there are other investors that trade on expected economic information embedded in the forecasts. 3.2 New Information or Anchoring Bias We continue our analysis by investigating the source of the results reported in Table 3. In 6 For example, Jain (1988), Flannery and Protopapadakis (2002), and Birz and Lott (2011) find that real economic news announcements do not affect daily stock returns. However, Lahaye et al. (2011) and Andersen et al. (2007), among many others, find that real economic news announcements affect intraday stock returns. 12

our first test, we hypothesize that if investors specifically trade on forecasts because they contain new information, we should observe larger stock price effects as soon as these forecasts become public. Subsequently, these effects should diminish and become statistically equal to zero in several days since the new information should be already reflected in stock prices. In Table 4, we examine the impact of survey forecasts on the daily S&P 500 returns starting from day t-5 the day when these forecasts become public. In contrast, we find that stock price effects of economic forecasts are statistically equal to zero on the day the forecasts become public, as well as on the following three days. The effects become statistically significant on the fourth day since the release of the forecasts, which is one day prior to the day of the official BLS announcements of UN and NFE statistics. Thus, the findings of Tables 3 and 4 show that the market only responds to information included in the economic forecasts days after they become public, which suggests it is unlikely that the forecasts contain new information. We further investigate the type of information driving the stock price effects of economic forecasts shown in Table 3. We do it in two steps. In the first step, we follow previous research and examine whether UN and NFE forecasts in our sample contain the anchoring bias, i.e., whether they are highly correlated with the values of past economic statistics. Specifically, we test whether the forecasts are biased towards the previous month s value of the economic releases, as well as, towards the average value of the three previous releases. We use the following specification: F t = b 0 + b 1 A h + e t (3) where Ft denotes the MMS forecast of each economic statistic and A-h stands for the average of 13

h lags of previously released statistics. 7 Columns 2-5 of Table 5 show the results of two regressions (one for each of the 2 employment announcements) comparing forecasts to the value of the previous month s release. Columns 6-9 show the results of regressions comparing forecasts to the average value of the three previous releases. In both cases (one-month model and three-months model), the anchoring bias embedded in NFE and UN forecasts is very strong, which is indicated by high economic and statistical significance. For example, columns three and seven show that almost 100% of the forecast on the unemployment rate is based on the values of the previously released rates. The results for Nonfarm Payroll Employment are also economically meaningful regardless of the used model. Therefore, these results confirm previous findings and indicate that market participants overvalue past information when making forecasts of economic statistics. Consequently, we hypothesize that the market participants trading on macroeconomic announcements may also overvalue the importance of past UN and NFE announcements. Thus, in the second step, we examine whether it is the historical component of the forecasts that impacts stock prices. We econometrically decompose the value of the forecast into the component formed as a result of anchoring on past announcements and the component formed as a result of other information. We then compare both components to stock returns using the following specification: R t = b 0 + b 1 F 1t ANC + b 2 F 2t NANC + b 3 S 3t + e t (4) 7 Campbell and Sharpe (2009) find the anchoring bias in MMS surveys using the following specification: S t = γ(f t A h )+ e t where S denotes surprise. This is because the authors assume that bond traders are initially rational since they only trade on unanticipated component of the announced statistics, i.e., surprises. Since surprises are a function of forecasts, the authors estimate the part of the surprises that are attributed to the anchoring bias of economic forecasts. 14

where F 1t ANC represents the anchored or stale component of each forecast, predicted by the model in equation (3), F 2t NANC represents new information or a non-anchored component of the forecast, measured as F t F 1t ANC. 8 S 3t is the surprise component of the announcement calculated as the difference between the announced value of the release and both components of the forecast. Regressions (1) and (4) of Table 6 report OLS estimates for the model without time series controls. In regressions (2) and (5), we include two lags of S&P 500 returns, two lags of returnvolume interactions, a January dummy, and a Friday dummy. Finally, in regressions (3) and (6), we use Huber (1981) M-estimator with time series controls. The results in all regressions are robust and show that only the anchored or historical component of the forecasts impacts stock returns. Moreover, the results of Table 4 and Table 6 suggest that investors do not specifically trade on survey forecasts of economic statistics, but on the values of past announcements that, to a large extent, comprise these forecasts. To further confirm this hypothesis, we employ the average value of the three past announcements instead of the forecasts and rerun the model from Table 3. The results, which show a statistically and economically significant relationship between the S&P 500 returns and past announcements, are reported in Table 7. For example, a one-standard deviation increase in the average unemployment rate during the previous three months, decreases Rt by 14.9 bps and the coefficient is statistically significant at the 5 % level. These estimates are also robust with various time series controls (models 1, 2, 5, 6). In fact, the ANC coefficients on A-3 are similar to the coefficients on Ft in Table 3 and the coefficients on F 1t in Table 6. Therefore, these results support our hypothesis and show that stock price effects of 8 We estimate F1t ANC using the average of the three previous releases, however the results remain similar if we use the previous month s release. 15

economic forecasts imply that some investors trade on past macroeconomic announcements. We also examine the effect of these past announcements on future daily and weekly stock returns starting with time t+1. We find no evidence of statistically significant return reversals. 9 Thus, these results reject overreaction hypothesis and suggests that the stock market underreacts to past economic announcements. 4. Discussion The results of Table 6 and Table 7 suggest a behavioral explanation for stock price effects of expected economic announcements. According to EMH, if investors were completely rational, they would only trade on new economic information. Alternatively, our findings suggest that some investors specifically choose to trade on economic announcements released during the previous three months. It is difficult to reconcile this finding with rational behavior since stock price effects of past economic announcements take place after the release of new economic data. Consequently, these results suggest that some investors may overvalue the importance of historical information for forecasting future economic conditions and stock returns. Findings in experimental psychology may explain this type of behavior. Edwards (1968) finds that individuals do not update their beliefs enough or as much as Bayesian would after the arrival of new information. In his experiment, individuals tend to underweight new data and tend to rely more on past data. The author calls this tendency a conservatism bias. In behavioral finance literature, Barberis et al. (1998, p. 315) argue that conservatism could explain why individuals may disregard the full information content of new public announcements and rely more on past data. Interestingly, Hirshleifer (2001, p. 1536) believes that conservatism is a form 9 Regressions showing the effect of past announcements on future returns are not reported to save space but may be available from the authors on request. 16

of anchoring bias identified by Tversky and Kahneman (1974). Consequently, our results suggest that the market participants who trade on macroeconomic news may exhibit a similar behavioral bias as the forecasters of economic statistics. In the remaining part of our study, we would like to discuss the significance of the recent financial crisis for our analysis. As stated before, our sample does not cover the post-2008 time period. This is because starting from December 2008, financial markets have been dominated by news about various asset purchase programs, which were established by the Federal Reserve and became known as quantitative easing (QE). In fact, some recent research argues that the established relationship between stock prices and real sector economic news may not hold in the presence of QE. For example, Löffler and Posch (2013) show that while positive (negative) real estate news increased (decreased) stock prices prior to 2008, this relationship is reversed during the post-2008 time period. The authors argue that bad real estate news during the post-2008 time period led to the stock market s expectation of a government bailout and QE, which meant good news for the market and led to higher stock returns. Although we are limited in the number of available monthly observations for the post- 2008 period, we also examine whether the relationship between stock prices and past employment announcements is the same (not reported here). Interestingly, we also find that stock price effects of our announcements are reversed after 2008, however the estimates are statistically insignificant. 5. Conclusion In this paper, we find that stock prices are impacted by the expected components of macroeconomic announcements, which are revealed by the survey forecasts of market 17

participants. Remarkably, this relationship is both economically and statistically significant for both employment factors. Moreover, we find that while traders do not specifically trade on economic forecasts, they trade on past economic announcements, which to a large extent comprise these forecasts. Upon further investigation, we confirm that stock prices underreact to employment announcements released during the previous three months even after the new data become public. Thus, our results are consistent with several studies in experimental psychology [Edwards (1968), Tversky and Kahneman (1974)] that suggest that some investors tend to overweight past economic announcements despite the availability of new information. In conclusion, our paper contributes to the behavioral literature on underreaction, however, unlike the existing studies that focus on firm-specific events, we show that investors may also underreact to macroeconomic announcements. 18

References Andersen, T.G., Bollerslev, T. (1998) Deutsche mark dollar volatility: intraday activity patterns, macroeconomic announcements, and longer run dependencies, Journal of Finance 53, 219 265. Andersen, T.G., Bollerslev, T., Diebold, F.X., Vega, C., 2003. Micro effects of macro announcements: Real-time price discovery in foreign exchange. American Economic Review 93, 38-62. Andersen, T.G., Bollerslev, T., Diebold, F.X., Vega, C., 2007. Real-time price discovery in global stock, bond and foreign exchange markets. Journal of International Economics 73, 251-277. Barberis, N., A. Shleifer, and R. Vishny, 1998, A model of investor sentiment, Journal of Financial Economics 49, 307 343. Barberis, N. and R. H. Thaler. 2003. A survey of behavioral finance. In Handbook of the Economics of Finance (ed. G. Constantinides, M. Harris, and R. Stulz). Amsterdam: North- Holland. Bernard, V. L. and J. K. Thomas, 1989, Post-earnings-announcement drift: Delayed price response or risk premium? Journal of Accounting Research, Supplement 27, 1 48. Bernard, V. L. and J. K. Thomas, 1990, Evidence that stock prices do not fully reflect the implications of current earnings for future earnings, Journal of Accounting and Economics 13, 305 340. Birz, Gene, 2014, Stale Economic News, Media, and the Stock Market, SSRN Working Papers. Birz, Gene and John R. Lott, Jr. 2011, The Effect of Macroeconomic News on Stock Returns: New Evidence from Newspaper Coverage, Journal of Banking & Finance 35, 2791 2800. Boyd, J. H., Hu, J., Jagannathan, R., 2005. The Stock market s reaction to unemployment news: Why bad news is usually good for stocks. Journal of Finance 60, 649-670. Brav, A., and Heaton, J. B., 2002, Competing theories of financial anomalies. Review of Financial Studies, 15(2), 575-606. Campbell, Sean D., and Steven A. Sharpe, 2009, Anchoring bias in consensus forecasts and its effect on market prices, Journal of Financial and Quantitative Analysis 44.02, 369-390. Carnes, W. Stansbury & Slifer, Stephen D. (1991) The Atlas of Economic Indicators. Harper Collins Publishers, Inc. Chen, N., Roll, R., Ross, S. (1986) Economic forces and the stock market, Journal of Business 46, 529-554. 19

Daniel, K. D., D. Hirshleifer, and A. Subrahmanyam, 1998, Investor psychology and security market under- and over-reactions, Journal of Finance 53, 1839 1886. DellaVigna, S. and J. Pollet, 2009, Investor inattention and Friday earnings announcements, Journal of Finance 64, 709-749. Edwards, W. (1968), Conservatism in human information processing, in: B. Kleinmutz, ed., Formal Representation of Human Judgment (Wiley, New York) pp. 17 52. Efron, B. (1979). Bootstrap methods: another look at the jackknife. The Annals of Statistics, 1-26. Garcia, Diego (2012) Sentiment During Recessions, The Journal of Finance, forthcoming. Gilbert, T., Kogan, S., Lochstoer, L., & Ozyildirim, A., 2012, Investor inattention and the market impact of summary statistics, Management Science 58(2), 336-350. Hess, D., & Orbe, S. 2013, Irrationality or efficiency of macroeconomic survey forecasts? Implications from the anchoring bias test. Review of Finance17(6), 2097-2131. Hirshleifer, D., 2001, Investor psychology and asset pricing, Journal of Finance 64, 1533 1597. Hirshleifer, D., S. S. Lim, and S. H. Teoh, 2009, Driven to distraction: Extraneous events and underreaction to earnings news, Journal of Finance 64, 2289-2325. Huber, Peter J., 1981, Robust Statistics (Wiley, New York). Huberman, Gur, and Tomer Regev (2001) Contagious speculation and a cure for cancer: A Nonevent that Made Stock Prices Soar, Journal of Finance 42, 483-510. Ikenberry, D., Lakonishok, J., Vermaelen T., 1995. Market underreaction to open market share repurchases. Journal of Financial Economics 39, 181-208. Jain, P., 1988. Response of hourly stock prices and trading volume to economic news. Journal of Business 61, 219-231. Jegadeesh, N., Titman S., 1993. Returns to buying winners and selling losers: implications for stock market e-ciency. Journal of Finance 48, 65-91. Lahaye, J., Laurent, S., Neely, C.J., 2011. Jumps, cojumps and macro announcements. Journal of Applied Econometrics 26, 893-921. Loughran, T., Ritter, J., 1995. The new issues puzzle. Journal of Finance 50, 23-51. Löffler, Gunter, and Peter N. Posch. Wall Street s bailout bet: Market reactions to house price releases in the presence of bailout expectations. Journal of Banking & Finance 37.12 (2013): 20

5147-5158. McQueen, G., Roley, V.V. (1993) Stock prices, news and business condition, Review of Financial Studies 92, 307-328. Michaely, R., Thaler, R., Womack K., 1995. Price reactions to dividend initiations and omissions: overreaction or drift? Journal of Finance 50, 573-608. Pearce, D. K., and V. V. Roley, 1985, "Stock Prices and Economic News," Journal of Business, 58, 49-67. Tetlock, Paul C. (2011) All the News That s Fit to Reprint: Do Investors React to Stale Information? Review of Financial Studies, forthcoming Tversky, A., and D. Kahneman (1974), Judgment under Uncertainty: Heuristics and Biases, Science, 185, 1124-1131. 21

Table 1: Summary Statistics: Forecasts, Releases, and Surprises (1992-2008) Panel A: MMS Forecasts Mean Std. Dev. Min Max Obs. Nonfarm Payroll Employment ( in thousands) 127.14 114.39-324 375 200 Unemployment rate (% ) 5.4 0.93 3.9 7.7 200 Panel B: Macroeconomic Releases Nonfarm Payroll Employment ( in thousands) 132.75 237.24-732 1226 200 Unemployment rate (% ) 5.36 0.93 3.9 7.8 200 Panel C: Macroeconomic Surprises Nonfarm Payroll Employment ( in thousands) 5.61 183.77-684 1056 200 Unemployment rate (% ) -0.03 0.15-0.4 0.4 200 22

Table 2: Correlations Among Forecasts and Surprises Panel A: Nonfarm Payroll Employment Forecasts Surprises Forecasts 1.00 Surprises 0.22 1.00 Panel B: Unemployment Rate Forecasts Surprises Forecasts 1.00 Surprises -0.10 1.00 23

Table 3: S&P 500 Effects of Macroeconomic Forecasts and Surprises The table shows the results of 6 (3 for each variable) regressions explaining the effect of macroeconomic factors on stock returns. Rt is the difference between the closing price on the S&P 500 at time t, the day of UN and NFE announcements, and the closing price at time t-1. F 1t denotes forecasts of UN and NFE announcements measured by MMS surveys. S 2t denotes surprises on UN and NFE calculated as in Andersen, Bollerslev, Diebold, and Vega (2003, 2007). Controls include two lags of S&P 500 returns, two lags of return-volume interactions, Friday dummy, and a January dummy. N = number of observations, p- values are reported in the parentheses, and standard errors are in brackets. * denotes significance at the 10% level; ** denotes significance at the 5% level; *** denotes significance at the 1% level. Unemployment Rate Nonfarm Payroll Employment OLS OLS M-estimator OLS OLS M-estimator Rt (1) (2) (3) (4) (5) (6) Ft - 0.167** - 0.202*** - 0.239*** 0.173** 0.194** 0.215** [0.071] [0.074] [0.070] [0.089] [0.091] [0.091] (0.02) (0.01) (0.00) (0.05) (0.03) (0.02) St -0.053-0.108-0.076-0.082-0.086-0.031 [0.096] [0.096] [0.091] [0.073] [0.083] [0.067] (0.58) (0.26) (0.40) (0.26) (0.30) (0.64) Controls No Yes Yes No Yes Yes Constant 1.04** 1.47** 1.68*** - 0.107 0.67-0.700 [0.440] [0.605] [0.596] [0.135] [0.402] [0.398] (0.02) (0.02) (0.01) (0.43) (0.87) (0.86) N 200 200 200 200 200 200 R² 0.02 0.09 0.10 0.01 0.09 0.09 24

Table 4: S&P 500 Effects of Macroeconomic Forecasts Around the Forecasts Release Times The table shows the results of 10 (5 for each variable) regressions explaining the effect of macroeconomic factors on stock return. Rt-5 is the difference between the closing price on the S&P 500 at time t-5, the day UN and NFE forecasts become public, and the closing price at time t-6. Rt-4, Rt-3, Rt-2 denote S&P 500 returns on the first day, second day, and third day since the release of the forecasts. Rt-1 denotes S&P 500 returns on the fourth day since the release of the F 1t forecasts, i.e., the day before BLS official announcements. denotes standardized forecasts of UN and NFE announcements measured by MMS surveys. N = number of observations, p-values are reported in the parentheses, and standard errors are in brackets. * denotes significance at the 10% level; ** denotes significance at the 5% level; *** denotes significance at the 1% level. Unemployment Rate Nonfarm Payroll Employment R(t-5) (1) R(t-4) (2) R(t-3) (3) R(t-2) (4) R(t-1) (5) R(t-5) (1) R(t-4) (2) R(t-3) (3) R(t-2) (4) R(t-1) (5) Ft 0.102-0.072 0.021-0.136-0.194*** -0.091 0.346-0.091-0.068 0.186* [0.066] [0.098] [0.093] [0.074] [0.073] [0.078] [0.236] [0.115] [0.110] [0.104] (0.12) (0.46) (0.82) (0.85) (0.01) (0.25) (0.14) (0.43) (0.54) (0.07) Constant -0.585 0.420 0.003 0.219** 1.04** 0.109-0.381 0.220 0.216-0.293* [0.422] [0.541] [0.564] [0.433] [0.440] [0.121] [0.321] [0.178] [0.167] [0.174] (0.17) (0.44) (0.99) (0.61) (0.02) (0.37) (0.24) (0.22) (0.20) (0.09) N 200 200 200 200 200 200 200 200 200 200 R² 0.01 0.003 0.003 0.002 0.02 0.01 0.06 0.01 0.01 0.03 25

Table 5: Anchoring Bias And Macroeconomic Forecasts The table shows the results of 4 regressions (2 for each of the 2 economic factors) relating macroeconomic forecasts to the lags of macroeconomic releases: F t = b 0 + b 1 A h + e t where Ft denotes the MMS forecast of each economic statistic. A-h stands for average of h lags of previously released statistics. Columns 2-5 show the results of regressions comparing forecasts to previous month s release, i.e. h = 1. Columns 6-9 show the results of regressions comparing forecasts to the average of the three previous releases, i.e. h = 3. N = number of observations, p-values are reported in the parentheses, and standard errors are in brackets. * denotes significance at the 10% level; ** denotes significance at the 5% level; *** denotes significance at the 1% level. One - Month Anchoring (h = 1) Three - Month Anchoring (h = 3) Ft, b0 b1 N R² b0 b1 N R² UN 0.045 0.998*** 200 0.99 0.056 0.995*** 200 0.98 [0.03] [0.01] [0.05] [0.01] (0.13) (0.00) (0.24) (0.00) NFE 90.75*** 0.269*** 200 0.29 62.92*** 0.467*** 200 0.48 [9.91] [0.05] [8.68] [0.048] (0.00) (0.00) (0.00) (0.00) 26

Table 6: The Effect of New and Old Information on S&P 500 Returns The table shows the results of 6 (3 for each variable) regressions explaining the effect of macroeconomic factors on stock return. Rt is the difference between the closing price on the S&P 500 at time t, the day of UN and NFE announcements, and the closing price at time t-1. F ANC 1t denotes the anchored or historical component of the forecast, which is predicted using the model in equation (3) and F NANC 2t denotes the unexpected component of the forecast, which is measured S 3t as F t F 1t ANC. denotes surprises on UN and NFE calculated as in Andersen, Bollerslev, Diebold, and Vega (2003, 2007). Controls include two lags of S&P 500 returns, two lags of return-volume interactions, Friday dummy, and a January dummy. N = number of observations, p-values are reported in the parentheses, and standard errors are in brackets. * denotes significance at the 10% level; ** denotes significance at the 5% level; *** denotes significance at the 1% level. Rt F 1t ANC Unemployment Rate OLS (1) OLS (2) M-estimator (3) OLS (4) Nonfarm Payroll Employment OLS M-estimator (5) (6) - 0.150** - 0.186*** - 0.219*** 0.207* 0.235** 0.261** [0.074] [0.074] [0.073] [0.125] [0.122] [0.115] (0.04) (0.01) (0.00) (0.10) (0.05) (0.02) F 2t NANC - 0.139-0.147-0.173 0.102 0.113 0.120 [0.114] [0.098] [0.107] [0.128] [0.099] [0.089] (0.22) (0.15) (0.11) (0.43) (0.26) (0.18) S3t - 0.053-0.110-0.074-0.082-0.086-0.031 [0.093] [0.093] [0.086] [0.084] [0.076] [0.062] (0.57) (0.24) (0.39) (0.33) (0.26) (0.62) Controls No Yes Yes No Yes Yes Constant 0.944** 1.40** 1.60*** - 0.145-0.092 0.094 [0.473] [0.603] [0.578] [0.173] [0.426] [0.442] (0.05) (0.02) (0.01) (0.40) (0.83) (0.83) N 200 200 200 200 200 200 R² 0.03 0.10 0.11 0.02 0.09 0.09 27

Table 7: S&P 500 Effects of Past Announcements The table shows the results of 6 (3 for each variable) regressions explaining the effect of past macroeconomic announcements on stock returns. Rt is the difference between the closing price on the S&P 500 at time t, the day of UN and NFE announcements, and the closing price at time t-1. A-3 denotes the standardized average value of the three previous announcements. St denotes surprises on UN and NFE calculated as in Andersen, Bollerslev, Diebold, and Vega (2003, 2007). Controls include two lags of S&P 500 returns, two lags of return-volume interactions, Friday dummy, and a January dummy. N = number of observations, p-values are reported in the parentheses, and standard errors are in brackets. * denotes significance at the 10% level; ** denotes significance at the 5% level; *** denotes significance at the 1% level. Unemployment Rate Nonfarm Payroll Employment OLS OLS M-estimator OLS OLS M-estimator Rt (1) (2) (3) (4) (5) (6) A-3-0.149** - 0.185*** - 0.218*** 0.141** 0.155** 0.174** [0.073] [0.075] [0.072] [0.072] [0.073] [0.087] (0.04) (0.01) (0.00) (0.05) (0.03) (0.05) St -0.051-0.105-0.071-0.064-0.067-0.021 [0.096] [0.096] [0.091] [0.081] [0.073] [0.069] (0.59) (0.27) (0.44) (0.43) (0.35) (0.76) Controls No Yes Yes No Yes Yes Constant 0.935** 1.35** 1.54*** -0.030 0.078 0.098 [0.457] [0.610] [0.609] [0.111] [0.382] [0.394] (0.04) (0.03) (0.01) (0.79) (0.84) (0.80) N 200 200 200 200 200 200 R² 0.02 0.09 0.09 0.01 0.08 0.08 28