The Effect of Macroeconomic News on Stock Returns: New Evidence from Newspaper Coverage

Transcription

1 Draft: November 2008 The Effect of Macroeconomic News on Stock Returns: New Evidence from Newspaper Coverage Gene Birz Department of Economics State University of New York at Binghamton Binghamton, NY John R. Lott, Jr. University of Maryland Foundation University of Maryland College Station, MD Abstract In this paper we choose a different approach of measuring real sector macroeconomic news to better estimate its effect on stock returns. Previous literature has produced weak evidence to support the hypothesis that real economic news affects stock returns. This is, in part, attributed to the difficulty of measuring how investors interpret macroeconomic news, as captured by the statistical releases on economic variables. Since newspaper headlines provide an interpretation of the statistical releases, we choose Lott and Hassett (2006) headlines classification as our measure of news. Our findings indicate that news on and unemployment does affect stock returns.

2 I. Introduction The last three decades of finance research have produced a tremendous number of papers examining the effect of news announcements on financial markets. While many of these papers study the effect of firm-specific news, others explore a theoretically appealing idea that financial markets react to macroeconomic news. Asset pricing theory argues that any variable that affects the level of consumption or the investment opportunity set should be priced in the equilibrium, thus making macro announcements excellent candidates to correlate with stock returns [Merton (1973), Breeden (1979)]. However, stock market effects of macroeconomic factors are hard to find empirically. Previous studies have found the effects of news about money growth and interest rate variables on stock prices. 1 But evidence for stock price effects of news about statistics on real variables such as, unemployment, durable goods orders, and others, is mixed. One difficulty in finding these effects comes from the fact that it is hard to measure the component of the statistical release relevant for stock prices. According to efficient market hypotheses, stock prices already incorporate all existing and expected public information and should only respond to new information. Thus, to capture new information in the economic releases, previous papers calculated economic surprises of the releases, measured by the difference between the release and financial market participants previous expectations of the release, as revealed by surveys. This difference would then represent unanticipated, new information about economic conditions and, if different from zero, should lead to a change in stock prices. For example, growth 1 Previous research found a statistically significant relationship between stock returns and interest rate variables such as short-term interest rates [Chen (1991)], term structure measured by the spread between long- and short-term interest rates [Chan, Karceski and Lakonishok (1998), Chen, Roll and Ross (1986), Campbell (1987)], and yield spreads between high and low grade bonds [Chen, Roll and Ross (1986), Chan, Chen and Hsieh (1985)]. In addition, studies found strong evidence of money growth affecting stock returns [Cornell (1983), Pearce and Roley (1983, 1985)]. 2

3 rate higher than expected would lead investors to expect good economic conditions in the future, which would raise the demand for securities and raise security prices. Another difficulty is that the interpretation of the release is another determinant of investors expectations and, therefore, a determinant of stock prices as well. In other words, investors form expectations based on their interpretation of the statistical releases of macroeconomic variables, which then affect their demand for securities and leads to changes in stock prices. For example, growth rate higher than expected can be interpreted as good news for stock prices in recessions but bad news in expansions [McQueen and Roley (1993)]. This finding can be explained by the fact that high growth rate in recessions can be interpreted by investors as a sign of economic improvement and lead to higher stock prices. Alternatively, the same high growth rate in expansions can lead to investors expectations of a contractionary monetary policy resulting in higher interest rates and, therefore, leading to lower stock prices. In this paper, we choose a different approach to find the relationship between macroeconomic factors and stock prices. We use newspaper headlines to measure macroeconomic news because headlines represent an interpretation of the statistical releases and can, therefore, be a good indicator of investors expectations about the economy. Newspaper articles often provide a detailed discussion of the economic reports by analyzing all the potential implications. However, a final conclusion about whether the release overall represents positive or negative economic news is captured by the headlines. This type of interpretation is then a good indicator of investors expectations about future economic conditions. For example, in response to release, on January 31, 2004, a headline in the New York Times was, Economy Remained Strong in 4th Quarter, US Reports. This headline interprets the release as positive news and informs the readers about good economic conditions. On February 16, 1996, Palm Beach 3

4 Post s headline in response to the release of durable goods orders read, Reports Indicate Economy Could Be Heading for Stall. This particular headline concludes that the durable goods report signals bad economic conditions. The headlines data were obtained from Lott and Hassett (2006), who examine whether media coverage of economic announcements depends on the party affiliation of the president. The dataset was created for nonfinance-related study, which contributes to the unbiasness of our study. Lott and Hassett (2006) collect all headlines related to macroeconomic releases and employ an unbiased research team to classify each headline as positive, negative, mixed, or neutral. Since a newspaper headline is the interpretation of the data releases, then Lott and Hassett (2006) classification should be able to capture whether the statistical release represents good news or bad news, which is what matters for investors expectations about future economic conditions. We examine four macroeconomic variables, specifically Growth, Rate, Retail Sales, and Durable Goods, and find that news on and does affect stock returns. The remainder of the paper is organized as follows. Section II will provide a more detailed discussion of theory and previous work. Section III will describe the data. We will discuss our empirical results in Section IV. Finally, Section V will conclude. II. Theory and Existing Research The channel by which macro news affects stock returns can be shown through the fundamental definition of the price of a security: P t = C t + E τ k τ = t + τ 4

5 where Pt is the price of a security at time t, Ct+τ is the expected future cash flows paid at time t+τ, and kt+τ is the future discounting factor that is a function of the risk-free interest rate and the risk premium. Macroeconomic news affects stock prices through investors expectations about future economic activity because economic conditions can affect cash flows and the discounting factor. According to the efficient market hypothesis, only new and unanticipated information of the economic release can affect stock prices. To proxy for this new information, most previous studies calculated economic surprises measured by the difference between the actual statistical release of economic variables and their expected values, which are obtained from surveys. 2 This difference then represents the unexpected component of the economic announcement, which, if different from zero, could affect investors expectations about the future state of the economy. Thus, researchers regressed stock returns on the economic surprises to see whether there is a correlation between stock returns and real economic activity. As mentioned in the previous section, existing research has numerously shown the effect of monetary policy variables on stock returns. However, stock market effects of real economic variables are difficult to find empirically. Pearce and Roley (1985) find a strong stock price effect of money stock surprises and a weaker effect of the discount rates. However, the correlation of stock returns and real activity, measured by the industrial production index and unemployment rate surprises, is not statistically significant. Alternatively, Chen, Roll, and Ross (1986) do not only find stock market effects of monetary policy variables, but they also find the effect of the industrial production index, thus finding the effect of real economic variables on stock returns. 2 The data on expectations of economic conditions are derived from weekly surveys of market participants collected by MMS Survey. 5

6 Flannery and Protopapadakis (2002) examine 17 macro announcements altogether. Their findings show that although conditional volatility of returns is lower on GNP announcement days, real GNP surprises are not significantly related to stock returns. The analysis of retail sales and employment announcements also failed to produce a statistically significant correlation with stock returns. Other studies, however, have shown that unemployment surprises change S&P 500 returns (Boyd et al. (2005), and McQueen and Roley (1993)). These papers recognize the fact that it is not only the surprise of the release that affects investors expectations and stock prices, but also the interpretation of the release. In particular, the authors argue that investors interpretation of economic growth depends on the state of the economy. For example, the news about high economic growth can have a positive impact on stock prices in a recession since this would indicate the end of the recession and would lead investors to expect better economic conditions. Alternatively, the news about high economic growth in expansions can result in lower stock prices if investors expect the central bank to increase interest rates. The most closely related article to what we propose is by Mitchell and Mulherin (1994), who investigate the relationship between stock returns and the importance of news announcements measured by the number of headlines transmitted by Dow Jones & Company wire services. They examine a variety of national news stories, including macroeconomic news about employment, retail sales, and durable goods, and find that none of the economic news is correlated with stock returns. As the authors themselves point out, the weakness of their approach lies in the fact that a mere count of newspaper headlines does not convey the importance of different news stories, nor does it differentiate among the nature of those stories. The advantage of our approach is that Lott and Hassett (2006) data classifies all headlines as positive, negative, neutral, 6

7 or mixed. As a result, this type of classification allows us to identify the sign of the relationship between news and returns. III. Data A. Headlines Data and News Confidence Index Data on macro announcements and the analysis of newspaper headlines are from Lott and Hassett (2006). They studied news coverage of four macroeconomic data series:, unemployment, retail sales, and durable goods. Newspaper and wire articles are only analyzed for the day of and the day after the release of the economic data. Out of 389 newspapers collected from LexisNexis files, only 31 exist prior to We therefore we begin our analysis from January Top ten newspapers, which are the Wall Street Journal, the New York Times, USA Today, the Los Angeles Times, Washington Post, New York Daily News, New York Post, Newsday, Chicago Tribune, and the Houston Chronicles, are also analyzed separately. Out of these ten newspapers, only four were included prior to In order to be included in this sample, headlines have to acknowledge the release of the economic data series on either the same day or the day after. The headlines were then classified as positive, negative, neutral, or mixed. 3 Tables 1 and 2 report the average number of news stories published on the release day by all newspapers in the sample and top ten newspapers separately for the period between January 1991 and April Newspaper stories are circulated the day after the release of macro series. and unemployment stories account for the majority of news articles. 3 For a detailed explanation of headlines classification, see Lott and Hassett (2006). Here are the examples of their classification: Economy Remained Strong in 4th Quarter, US Reports, NY Times, 01/31/2004 positive March Jobs Growth Worries Homebuilding Industry, Dallas Morning News, 04/03/2004 mixed Reports Indicate Economy Could be Heading for Stall, Palm Beach Post, 02/16/96 negative 7

8 On average, there are stories on a release day that cover news about. Out of these 28.87, are positive, 7.32 are negative, 2.15 are neutral, and 7.75 are mixed. In addition, out of the stories about from all newspapers, 7.26 are from top ten newspapers only. They include 3.04 positive stories, 1.85 negative, 0.23 neutral, and 2.15 mixed stories. The average number of stories about unemployment is close to the amount of stories about. However, there is less coverage on retail sales and durable goods releases. On average, there are 13.7 stories about retail sales and 13.4 stories about durable goods on a release day. Tables 3 reports the average amount of Associated Press (AP) stories issued per day for the period between January 1991 and April Unlike newspaper articles, AP stories are transmitted on the day of macro series releases. To determine the net effect of economic news, we then calculate a News Confidence Index. Since Lott and Hassett (2006) classification of newspaper headlines is similar to the Conference Board s Consumer Confidence Survey responses (positive, negative, etc.) we follow their methodology to calculate our News Confidence Index. First, we derive the amount of positive and negative stories in percent terms. Then we subtract the percent of negative stories from the percent of positive stories and add 100 to get rid of all negative values. Table 4 summarizes the daily news index for each macro statistic from all newspapers, top ten newspapers, and the Associated Press. A potential weakness of our study is the fact that we compare stock returns on the day of macro releases to the news coverage of economic events from the following day. We are forced to employ this method since, with the exception of wire services such as the Associated Press, virtually all news stories are in fact transmitted on the day after the release of economic variables and the usage of the news stories is crucial for our analysis. As a result, we have to control for reverse causation in our analysis. 8

9 B. Macro Announcements In order to show that it is not just the unanticipated component of the release that affects stock returns, but also the way in which the release is interpreted by the investors, we have to control for the economic surprises of the statistical releases. The surprises are traditionally measured by calculating the difference between the release and the surveybased expectations of the release. As a result, we need to obtain the real-time values of macroeconomic releases for each of the four macro variables that we use. The reports on all four variables are released at 8:30 AM. Durable goods orders data is from the Advance Report on Durable Goods Manufacturers Shipments, Inventories, and Order, which is released monthly by the Department of Commerce, U.S. Census Bureau. Retail Sales data is from Advance Monthly Sales for Retail Trade and Food Services, which is also released monthly by the Department of Commerce, U.S. Census Bureau. We collect the Advance Estimate of from monthly Economic Indicators issued by the Council of Economic Advisers and available at the Federal Reserve Bank of Chicago. rate data is from the Bureau of Labor Statistics. We use Advance estimates of growth rate, durable goods orders, and retail sales because Lott and Hassett (2006) only collected headlines following the Advance estimates of these variables. C. Traditional Macro Surprises and Expectations We calculate traditional macro surprises by calculating the difference between the government s actual releases and their expected values. The expectations on,, Durable Goods, and Retail Sales are collected by MMS Survey and were purchased from Haver Analytics. Usually approximately one week before the release of the economic indicators, MMS conducts weekly telephone surveys of market participants 9

10 which include employees of commercial banks, consulting firms, and academic institutions. D. Asset Prices The time series of closing daily returns on the S&P 500 were obtained from Wharton Research Data Services (WRDS) from January 1, 1991 to April 30, We then picked out returns for the release days of the economic data series. III. Results We first examine how the traditional macro surprises on, unemployment, retail sales, and durable goods predict stock returns for the period between 1991 and The results are reported in Table 5. S&P 500 returns on the day of the release of macro series are regressed on macro surprises. None of the coefficients are statistically significant, once again showing that traditional macro surprises do not seem to predict stock returns. Some have noted that the impact of traditional macro surprises depends on when they occur during the business cycle. Two notable examples of this are Flannery and Protopapadakis (2002) and McQueen and Roley (1993), where they define economic regimes according to the growth rate of the industrial production index. 4 Again, we follow these authors in separating out three different periods of growth rate in this index (Low, Medium, and High), with which we interact the macro surprise data released by 4 Step 1: Run the following regression: ln (Industrial Production) = a + b*time Trend + e Step 2: Derive residuals from this regression. Low economic regimes correspond to the months with the lowest 25% of all the residuals. High regime corresponds to the highest 25% of all the residuals. And Medium regime is the remaining 50%. Step 3: Stock Returns = a +L*Xb1 + M*Xb2 + H*Xb3 + d +e where H = 1 if High economic state and 0 otherwise, L = 1 if Low economic state and 0 otherwise, M = 1 if Medium economic state and 0 otherwise, X is a vector of macro surprises, d = day of the week dummy. 10

11 the government. Table 6 reports these results. Again, there is little evidence, at least over the period of time that we are examining, that macro surprises impact S&P 500 returns on the day of macroeconomic releases. None of the results is statistically significant at the 10-percent level for a two-tailed t-test. Table 7 examines the effect of news coverage of macroeconomic releases from all newspapers in our sample on S&P 500 returns on the day of macro releases. In order to prove our hypothesis that it is not just the unanticipated component of the release that affects stock returns, but also the way in which the release is interpreted by the investors, we regress stock returns on our newspaper variable, controlling for the traditional economic surprises. A one-percent increase in the positive news coverage of increases S&P 500 returns by % (2.2 basis points). Alternatively, a one-percent increase in the negative news coverage of decreases S&P 500 returns by 0.013% (1.3 basis points). Jointly, a one-point increase in the news confidence index on derived from all newspapers increases S&P 500 returns by 0.01% (1 basis point). Since the daily standard deviation of the news index reported in Table 4 is 60.8 points, on average, S&P 500 returns increase by approximately 60.8 basis points. The average response of S&P 500 returns to the news index on unemployment is approximately 37.6 basis points. News coverage about durable goods and retail sales, although having a correct sign, do not seem to affect stock returns in a statistically significant way. One explanation of this result is that news about durable goods and retail sales are not viewed by the market as important indicators of future economic conditions. However, statistical insignificance in this case can also be the result of fewer news stories about these variables. 11

12 Table 8, which breaks down the macro surprises by where they occur during the business cycle, provides very similar results. Again, news coverage of and unemployment impacts S&P 500 returns. Since the results are so similar, throughout the remainder of the paper, we will only report the results that include information on the macro surprises and not those that account for the business cycle. Table 9 shows a similar test with the news coverage derived from the top ten newspapers only. The stock market s response to news about is slightly greater when measured by the news stories from top ten newspapers. This result can be explained by the fact that the top ten newspapers are better at capturing the true meaning of the release. In addition, the coefficients on news stories about durable goods are statistically significant as well. IV. The Issues As mentioned before, our concern is the possibility of a causation problem. The headline writers and journalists covering the story the day after the release of macro series are already aware of stock market events the day before and might be influenced by it when assigning a headline or writing the article. For instance, if stock indicators behave poorly on the day of the release that shows constant growth, a journalist s coverage of the release on the following day might have a negative connotation. This, of course, would result in what we would classify as a negative headline. As a result, our main finding that stock returns on the day of the release of macro announcements are correlated with their newspaper coverage the day after the release may not mean that this news coverage is a priced factor. To determine whether such a causation problem really exists, we redefined our news variable as the number of positive news stories covered by the Associated Press 12

13 Wire stories only. The advantage of using AP Wire news is that these particular news articles are written and transmitted on the same day as the releases of the actual macroeconomic announcements. Since the announcements of our four macro variables are released in the morning, 5 the AP wire stories that cover the announcements are published early, with some being published even before the market opens. Therefore, a journalist covering the story is less likely to be affected by the stock market news since the closing day s returns are not yet known. For instance, on April 30, 1998, the day of the first quarter Advance Estimate release, the first AP story, under the headline U.S. Economic Growth Accelerates, was published at 8:40 AM, only ten minutes after the release of the announcement and 50 minutes before the market opened. Of course, it is arguable that the journalist writing the story may be influenced by the stock market s activity before its closing, thus resulting in a biased coverage of macro announcements. Unfortunately, given our dataset on news coverage, we cannot control for this type of bias. Nevertheless, if we find that stock returns on the day of the macroeconomic announcements are significantly related to positive AP stories, then the results in the previous section most likely support our original hypothesis. Table 10 reports regression analysis of S&P 500 returns and the news coverage derived from the AP stories. Both growth and unemployment news significantly affect S&P 500 returns on the day of the release of macro series. V. Conclusion In this paper, we examine whether real sector economic news affects stock returns. While anecdotal evidence suggests that there is such a correlation, the empirical evidence 5,, Retail Sales, and Durable Goods announcements are released at 8:30 AM. 13

14 in support of this hypothesis was rather weak. Previous literature has indicated that investors interpretation of the statistical releases is a determinant of stock prices since it affects their expectations about future economic conditions. As a result, we choose a different approach and use newspaper coverage as our proxy for investors interpretation of macroeconomic news. Overall, we examine news about growth, unemployment, retail sales, and durable goods and find that the news about growth and unemployment significantly affects stock returns. Although having a correct sign, the correlations between stock returns and news about durable goods and retail sales are statistically insignificant. This can be a result of these variables not being indicative of future economic conditions or lower statistical power due to a smaller number of stories about retail sales and durable goods reports. Unfortunately, there is a possibility of a causation problem in our analysis since the newspaper articles might be written after the closing of the stock market on the day of the economic releases. To control for this, we employ AP stories as another measure of economic news. The advantage of using AP stories is that they are transmitted, and thus written, much earlier in the day, before the closing of the stock market. It is still possible that the coverage is based on earlier activity of the stock market, but given the nature of the dataset we cannot control for it. We believe that our findings provide evidence for stock market effects of real economic news. However, due to the fact that the newspaper articles are transmitted the day after economic releases, it is impossible to develop trading strategies based on the information contained in the news articles. Therefore, future research should identify better proxies of real economic news that can be employed to generate profits from trading. 14

15 References Boyd, John H., Jian Hu, and Ravi Jagannathan, 2005, The Stock Market s Reaction to News: Why Bad News Is Usually Good for Stocks, Journal of Finance, 60, Breeden, D., 1979, An Intertemporal Asset Pricing Model with Stochastic Consumption and Investment Opportunities, Journal of Financial Economics, 7, Chan, K.C., N. F. Chen, and D.A. Hsieh, 1985, An Exploratory Investigation of the Firm Size Effect, Journal of Financial Economics, 14, Chan, K, C., J. Karceski, and J. Lakonishok, 1998, The Risk and Return from Factors, Journal of Financial and Quantitative Analysis, 33, Chen, Nai-fu., 1991, Financial Investment Opportunities and the Macroeconomy, Journal of Finance, 46, Chen, Nai-fu, Richard Roll, and S. Ross, 1986, Economic Forces and the Stock Market, Journal of Business, 46, Cornell, B., 1983, The Money Supply Announcements Puzzle, American Economic Review, 83, Cutler, David M., J.M. Poterba, and L.H. Summers, 1989, What Moves Stock Prices? Journal of Portfolio Management, 15, Dyck, Alexander and Luigi Zingales, 2003, The Media and Asset Prices, Working Paper, NBER. Flannery, M. J. and A. A. Protopapadakis, 2002, Macroeconomic Factors Do Influence Aggregate Stock Returns, Review of Financial Studies, 15, Ghent, A., 2008, Why Do Markets React Badly to Good News? Evidence from Fed Funds Futures, Working Paper. Hautsch, Nikolaus, Hess, Dieter, and Müller, Christoph, 2008, Price Adjustment to News with Uncertain Precision, Available at SSRN: Jain, P., 1988, Response of Hourly Stock Prices and Trading Volume to Economic News, Journal of Business, 61, Lott, John R., Jr and Kevin A. Hassett, 2006, Is Newspaper Coverage of Economic Events Politically Biased?: , Working paper, AEI. McQueen, Grant and Vance V. Roley, 1993, Stock Prices, News and Business Conditions, Review of Financial Studies, 92,

16 Merton, R., 1973, An Intertemporal Capital Asset Pricing Model, Econometrica, 41, Mitchell, M. L. and J. H. Mulherin, 1994, The Impact of Public Information on the Stock Market, Journal of Finance, 49, Taylor, J.B., 1993, Discretion versus Policy Rules in Practice, Carnegie-Rochester Conference Series on Public Policy 39, December, Tetlock, Paul C., 2006, Giving Content to Investor Sentiment: The Role of the Media in the Stock Market, Journal of Finance, forthcoming. Wolfers J. and E. Zitzewitz, 2004, Prediction Markets, Journal Economic Perspectives, 18,

17 Table 1: Daily Summary Statistics for News Stories from All Newspapers ( ) Average Std. Dev. Min Max Total Positive Negative Neutral Mixed Retail Sales Total Positive Negative Neutral Mixed Total Positive Negative Neutral Mixed Durable Goods Total Positive Negative Neutral Mixed

18 Table 2: Daily Summary Statistics for News Stories from Top 10 Newspapers ( ) Average Std. Dev. Min Max Total Positive Negative Neutral Mixed Retail Sales Total Positive Negative Neutral Mixed Total Positive Negative Neutral Mixed Durable Goods Total Positive Negative Neutral Mixed

19 Table 3: Daily Summary Statistics for Associated Press (AP) Stories ( ) Mean Std. Dev. Min Max Total AP AP positive AP negative AP neutral AP mixed Retail Sales Total AP AP positive AP negative AP neutral AP mixed Total AP AP positive AP negative AP neutral AP mixed Durable Goods Total AP AP positive AP negative AP neutral AP mixed

20 Table 4: Daily Summary Statistics for News Confidence Index from All Newspapers, Top 10 Newspapers, and AP Stories ( ) Variable Mean Std. Dev. Min Max News Index News Index AP Index Retail Sales News Index News Index AP Index News Index News Index AP Index Durable Goods News Index News Index AP Index

21 Table 5: The Effect of Traditional Macro Surprises on S&P 500 Returns The table shows the results of regressions in which the dependent variable is daily percentage changes in the closing values of the S&P 500 Index. Macroeconomic surprises were derived by calculating the difference between actual the government s releases and their expected values. All regressions include days-of-the-week dummy variables that are not reported here. Coef.= OLS estimates, N = number of observations, F = F-statistic, p-values are in the parentheses. Italics denote significance at the 10% level; bold denotes significance at the 5% level; bold and italics denote significance at the 1% level. Coef. N R² F Rate Surprise (0.506) Retail Sales Surprise (0.652) Rate Surprise (0.574) Durable Goods Surprise (0.853) 21

22 Table 6: The Effect of Traditional Macro Surprises on S&P 500 Returns Accounting for the Business Cycle Effects The table shows the results of regressions in which the dependent variable (daily percentage changes in the closing values of the S&P 500 Index) is regressed on traditional macro surprises after breaking down the impact by segment of the business cycle. Low, Medium, and High represent different states of the economy as defined by McQueen and Roley (1993). All regressions include days-of-the-week dummy variables that are not reported here. Coef.= OLS estimates, N = number of observations, F = F-statistic, p- values are in the parentheses. Italics denote significance at the 10% level; bold denotes significance at the 5% level; bold and italics denote significance at the 1% level. Low Medium High N R² F Rate Surprise (0.324) (0.452) (0.372) Retail Sales Surprise (0.181) (0.742) (0.535) Rate Surprise (0.73) (0.714) (0.472) Durable Goods Surprise (0.239) (0.657) (0.161) 22

23 Table 7: Explaining the Effect of News Stories from All Newspapers on S&P 500 Returns, Controlling for Traditional Macro Surprises The table shows the results of 12 different regressions (3 for each of the 4 different macro variables) in which the dependent variable is S&P 500 returns. % Positive and % Negative represent the percentage of positive and negative news stories collected from all newspapers in the sample. News Index represents the News Confidence Index that was calculated to account for the effect of both positive and negative news stories. All regressions include days-of-the-week dummy variables that are not reported here. Coef.= OLS estimates, N = number of observations, F = F-statistic, p-values are in the parentheses. Italics denote significance at the 10% level; bold denotes significance at the 5% level; bold and italics denote significance at the 1% level. Rate Durable Goods Retail Sales Unemploy ment Rate Durable Goods Retail Sales Rate Retail Sales Durable Goods (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) %Positive (0.001) (0.635) (0.000) (0.435) %Negative (0.081) (0.469) (0.007) (0.377) News Index (0.008) (0.910) (0.000) (0.389) Rate Surprise (0.054) (0.439) (0.118) Retail Surprise (0.428) (0.787) (0.586) Rate Surprise (0.045) (0.202) (0.072) Durable Goods Surprise (0.389) (0.342) (0.352) N R² F

24 Table 8: Explaining the Effect of News Stories from All Newspapers on S&P 500 Returns, Controlling for Traditional Macro Surprises in Different States of the Business Cycle The table shows the results of 12 different regressions (3 for each of the 4 different macro variables) with the following specification: R = a + Newspaper Coverage + Low*Xb1 + Medium*Xb2 + High*Xb3 Where R is daily percentage changes in the closing values of the S&P 500 Index. X is a vector of real economic surprises on, Retail Sales,, and Durable Goods. Newspaper Coverage consists of either % Positive, % Negative, or News Index depending on the specification. % Positive and % Negative represent the percentage of positive and negative news stories collected from all newspapers in the sample. News Index represents the News Confidence Index that was calculated to account for the effect of both positive and negative news stories. Low, Medium, and High represent different states of the economy as defined by McQueen and Roley (1993). All regressions include days-of-the-week dummy variables that are not reported here. Coef.= OLS estimates, N = number of observations, F = F-statistic, p-values are in the parentheses. Italics denote significance at the 10% level; bold denotes significance at the 5% level; bold and italics denote significance at the 1% level. Retail Sales Rate Durable Goods Retail Sales Rate Durable Goods Retail Sales Rate (1) (2) (3) (4) (5) (6) (7) (8) 9) (10) (11) (12) %Positive (0.00) (0.39) (0.00) (0.53) %Negative (0.08) (0.63) (0.01) (0.44) News Index (0.01) (0.83) (0.00) (0.46) Low (0.94) (0.12) (0.81) (0.41) (0.62) (0.24) (0.78) (0.45) (0.91) (0.17) (0.72) (0.44) Medium (0.02) (0.89) (0.02) (0.50) (0.12) (0.63) (0.11) (0.43) (0.04) (0.78) (0.04) (0.45) High (0.28) (0.21) (0.34) (0.13) (0.92) (0.42) (0.94) (0.11) (0.44) (0.31) (0.56) (0.12) N R² F Durable Goods 24

25 Table 9: Explaining the Effect of News Stories from the Top 10 Newspapers on S&P 500 Returns, Controlling for Traditional Macro Surprises The table shows the results of 12 different regressions (3 for each of the 4 different macro variables) in which the dependent variable is S&P 500 returns. % Positive10 and % Negative10 represent the percentage of positive and negative news stories collected from the top 10 newspapers in the sample. News Index10 represents the News Confidence Index that was calculated from the top 10 newspapers to account for the effect of both positive and negative news stories. All regressions include days-of-the-week dummy variables that are not reported here. Coef.= OLS estimates, N = number of observations, F = F-statistic, p-values are in the parentheses. Italics denote significance at the 10% level; bold denotes significance at the 5% level; bold and italics denote significance at the 1% level. Retail Sales Rate Durable Goods Retail Sales Unemploy ment Rate Durable Goods Retail Sales Rate Durable Goods (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) %Positive (0.000) (0.294) (0.000) (0.051) %Negative (0.032) (0.363) (0.003) (0.018) News Index (0.000) (0.880) (0.000) (0.020) Rate Surprise (0.007) (0.263) (0.012) Retail Surprise (0.241) (0.654) (0.399) Rate Surprise (0.196) (0.264) (0.161) Durable Goods Surprise (0.128) (0.053) (0.059) N R² F

26 Table 10: Explaining the Effect of News Coverage from AP Wire Stories on S&P 500 Returns, Controlling for Traditional Macro Surprises Each panel of this table shows the results of 12 different regressions (3 for each of the 4 different macro variables) in which the dependent variable is the S&P 500 returns. % PositiveAP and % NegativeAP represent the percentage of positive and negative news stories collected from AP Wire Stories. News IndexAP represents the News Confidence Index that was calculated from AP Wire Stories to account for the effect of both positive and negative news stories. All regressions include days-of-the-week dummy variables that are not reported here. Coef.= OLS estimates, N = number of observations, F = F-statistic, p-values are in the parentheses. Italics denote significance at the 10% level; bold denotes significance at the 5% level; bold and italics denote significance at the 1% level. Retail Sales Rate Durable Goods Retail Sales Rate Durable Goods Retail Sales Rate (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) %PositiveAP (0.284) (0.688) (0.005) (0.779) %NegativeAP (0.084) (0.839) (0.251) (0.840) News IndexAP (0.094) (0.912) (0.023) (0.793) Rate Surprise (0.993) (0.631) (0.641) Retail Surprise (0.681) (0.527) (0.616) Rate Surprise (0.140) (0.632) (0.214) Durable Goods Sales Surprise (0.502) (0.528) (0.544) N R² F Durable Goods 26