Stock Prices, News, and BusinessConditionsGrant McQueenBrigham Young UniversityV. Vance RoleyUniversity of WashingtonPrevious research finds that fundamental mac- roeconomic news has little effect onstock prices. We show that after allowing for different stages of the business cycle, astronger relationship between stock prices and news is evident. In addition to stockprices, we examine the effect of real activity news on proxies for expected cash cowsand equity discount rates. We find that when the economy is strong the stock marketresponds negatively to news about higher real economic activity. This negativerelation is caused by the larger increase in discount rates relative to expectedcash flows.Apart from some types of monetary information, there is little empirical evidence tosupport the hypothesis that stock prices respond to macroeconomic news. Schwert (1981)finds that the daily response of stock prices to news about inflation from 1953 to 1978 isweak and slow. Pearce and Roley (1985) use survey data to measure expectations and findthat daily stock prices respond to monetary information between Sep- tember 1977 andOctober 1982, but news about the consumer price index, unemployment, and industrialWe are grateful to Lynda S. Livingston and Steven R. Thorley for research assistance, to Stephen J. Brown (the editor),Wayne E. Fetson, Douglas K. Pearce, J. Michael Pinegar. James M. Poterba. Simon M. Wheatley, an anon- ymous referee,and seminar participants at the National Bureau of Economic Research and the University of Washington for helpfulcomments, and to the Center for the Study of Financial Management, University of Washington, and Seafirst Bank, forresearch support. Address correspondence to V. Vance Roley, Department of Finance, DJ-10, University of Washington, Seattle,WA 98195.production have no significant effect on prices. Hardouvelis (1987)considers a somewhat broader set of variables through August 1984and concludes that stock prices respond primarily to monetary news.Finally, Cutler, Poterba, and Summers (1989) use vector autoregressions to measure news about macroeconomic time series from 1871to 1986. They conclude that less than one-third of the monthly returnvariance can be explained from these sources.’Each of these studies assumes that investors’ response to news isthe same over different stages of the business cycle. For instance,Cutler, Poterba, and Summers (1989) implicitly assume that a positivesurprise in industrial production at the end of the Great Depressionevokes the same response as a surprise in late 1969, after nearly adecade of expansion. A positive surprise in industrial productionduring the depression could indicate the end of the depression andhigher forecasts of firms’ cash flows. Such an announcement wouldlikely be “good news” for the stock market. In late 1969, with lowunemployment and factories running near full capacity, a positivesurprise in industrial production may result in fears of an overheatingeconomy, inflation, and possible efforts by policymakers to increasereal interest rates. Such an announcement could then be “bad news”for the stock market. If the same type of news is considered good insome states of the economy and bad in others, the response coefficienton the surprise in previous studies will be biased toward zero.2The popular press uses this good news/bad news story to interpretdaily stock price movements. For example, on February 4, 1983, after16 months of recession, the Labor Department reported that the unemployment rate fell to 10.4 percent. This represented a rate of 0.2 or0.3 percentage points below what was expected. This news was usedby the media to explain the 13.25-point jump in the Dow Jones Industrial Average, and prompted the Chairman of the Council of EconomicAdvisers, Martin Feldstein, to comment that “a recovery is eitherbeginning or already here” ( Wall Street Journal, February 7, 1983).In contrast, on November 4, 1988, after six years of expansion, theLabor Department reported that the unemployment rate fell to 5.3percent, matching a 14-year low. This represented a rate of 0.1 or 0.2percentage points below what was expected. The media’s interpreta1Chen, Roll, and Ross (1986) also investigate whether monthly stock returns covary with variousmacroeconomic variables. They again find that the explanatory power is low. The main focus oftheir study, however, is whether the covariance of economic variables with stock returns can explainex ante returns.2Several recent studies find significant effects from business conditions on stock returns. Ferson andMerrick (1987). for example, find shifts in consumption-based asset pricing parameters acrossstages of the business cycle measured by recession versus nonrecession. Fama and French(1989) and Fama (1990) consider term-premium and default-risk-premium variables asdeterminants of equity discount rates. They suggest that the term premium is related to NBERbusiness cycles, while the risk premium is related to business conditions over longer periods.684tion in this instance was “bond market investors reacted with gloom,sending interest rates higher on fears of tighter Fed policy. The stockmarket also fell” ( Wall Street Journal, November 7, 1988). The problem with this type of evidence, however, is that it is anecdotal andlargely after the fact.In this article we examine whether the response of stock prices tomacroeconomic news varies over different stages of the business cycle.By allowing the response to vary over different states of the economy,we can test the good news/bad news story and provide unbiasedestimates of the effects of fundamental information about the economy. We study daily percentage changes in closing values of theStandard & Poors 500 Index and several variables related to equitydiscount rates and cash flows. By considering these other variables,we can investigate the sources of any business-condition effect onthe response of stock prices.Following this introductory section, we present in Section 1 a simple theoretical framework to consider how news affects stock pricesand how this effect can vary over different stages of the business cycle.We describe the data in Section 2. In Section 3 we present the empirical results. We consider the robustness of the results in Section 4,and we summarize the main conclusions in Section 5.1. Theoretical FrameworkA common model that links stock prices to information posits thatstock prices equal the present discounted value of rationally forecasted future dividends. This model can be represented aswhere P, is the price of the stock at timedenotes the mathematical expectation conditional on information available at time tis the dividend paid at timeis the stochasticdiscount factor for cash flows that occur at timeEconomic announcements affect daily share price movements if thenew information revealed by announcements affects either expectations of future dividends or discount rates or both. The new information is represented by the difference in the announced value onday t + 1 and the expected value as of day t. Consequently, theunanticipated component of an announcement on day t + 1 is uncorrelated with information available on day t. The information set,includes past announcements of other economic variables, soannouncement surprises are uncorrelated under rational expectationsif they are made on different days. Combining daily stock-price changes685with announcement surprises on different days allows us to isolatethe effects of individual economic variables.1.1 Impact of real economic activity surprisesWe need not expect that real economic activity surprises will affectcash flows and discount rates in the same way across different statesof the economy. As a result, stock prices may well react differently tosurprises of this nature, depending on whether the economy is operating below capacity. When the economy is booming, for example, areal economic activity surprise could result in a larger increase indiscount rates than cash flows, causing stock prices to fall. In thiscase, high capacity utilization and employment may constrain furtherincreases in output and, consequently, cash flow in the absence ofnew investment in plant and equipment.The announcement effect we examine corresponds to the disclosure in month t of production growth that already occurred in montht – 1. Information about the previous month is relevant in that it maychange expectations about the future. That is, consistent with Fama(1990) and Schwert (1990), the information provided by an industrialproduction announcement causes stock prices to respond if this information causes revisions in expected future industrial production.1.2 Impact of other economic announcement surprisesWe also consider possible asymmetric effects of announcement surprises other than those related to real economic activity. This othereconomic information is, however, less closely related to the possiblebusiness-conditions effects discussed above. The announcements weconsider are for foreign trade, inflation, and money. We briefly discusseach in turn.First, foreign trade deficit announcements have at times receivedconsiderable attention in the popular press. For the 1979-1984 period,however, Hardouvelis (1987) does not find any significant effects onstock prices. We update his sample and test for varying effects overdifferent economic states.Second, following the empirical studies of Nelson (1976) and Famaand Schwert (1977), a number of studies estimate a significant negative relationship between inflation and stock returns. Among these,Feldstein (1980) argues that the tax treatment of depreciation andinventories results in lower real after-tax corporate profits and, hence,lower stock prices during times of inflation. Fama (1981), Geske andRoll (1983), and Kaul (1987) explain the negative relationship byappealing to real output effects. In terms of inflation announcementsurprises, the significance of the stock-price response is mixed [e.g.,Pearce and Roley (1985) and Hardouvelis (1987)]. We again extend686these announcement studies by lengthening the sample and by allowing business-condition-dependent responses.Third, Pearce and Roley (1983, 1985), Cornell (1983), and Hardouvelis (1987) find that stock prices respond significantly to moneyannouncement surprises. Varying responses over different monetarypolicy regimes are tested in these studies, but possible businessconditions effects are not considered.3 We estimate the stock-priceresponse not only to money announcements but also to Federal Reservediscount rate changes, over different economic states.2. DataOur sample period begins in September 1977 and ends in May 1988.The start of the sample period coincides with the initial availabilityof survey data from Money Market Services International (MMS). Wediscuss the robustness of the results using alternative sample periodsand expectation measures in Section 4.3.2.1 Asset prices and yieldsWe use daily percentage changes in the closing value of the S&P 500Index to estimate the response of stock prices to new macroeconomicinformation. For economic announcements occurring either beforeor while the stock market is open, we use the percentage change inthe index from the previous business day’s closing price to the closingprice on that day. For announcements made after the stock market isclosed, we use the percentage change in the index from that day’sclosing quote to the next business day’s closing quote. Throughoutthe sample, the stock market closed at 4:00 P.M. EST. (We use ESTfor all closing and announcement times.)To measure the response of equity discount rates to new information, we consider several proxies. These include daily changes inthe three-month Treasury-bill and lo-year Treasury-bond yields. Following Fama and French (1989) and Fama (1990), we also includevariables denoted as the term spread and the default spread as equitydiscount rate proxies. We represent the term spread by Moody’s Aaacorporate bond yield minus the three-month bill yield, and the default3Given the evidence that both short- and long-term interest rates respond differently to moneyannouncement surprises over different Federal Reserve policy regimes [e.g., Roley (1983, 1986).Cornell (1983), and Roley and Walsh (1985)], another potentially interesting hypothesis is thatstock prices respond differently to economic news over these regimes. For the October 1979 andOctober 1982 regimes, however, Pearce and Roley (1983, 1985) and Hardouvelis (1987) find nosignificant difference in the stock market’s response to money surprises. We nevertheless investigatethe effects of the monetary policy regimes in October 1979, October 1982, and February 1984. andthe hypothesis that the stock market’s response is the same across regimes for our set of economicannouncements can be rejected only at the 25 percent significance level. Consequently, we do notexamine the effects of monetary policy regimes further.687spread by Moody’s Baa corporate bond yield minus the Aaa yield.These yield data are from the Federal Reserve’s H.15 release, andthey correspond to yields based on bid prices prevailing at 3:30 P.M.42.2 Economic announcementsVirtually all of the economic announcements are well-publicizedevents with regular schedules. Data on industrial production (IP) areinitially released, seasonally adjusted monthly percentage changes inthe Federal Reserve Industrial Production Index, all items. BetweenJanuary 1979 and October 1985, the announcements were made at9:30 A.M.; since October 1985, at 9:15 A.M. Before 1979, the industrialproduction press releases give no specific announcement time, statingonly “for immediate release.” However, the announcements weremade before the market opened for our sample.Data on the unemployment rate (UNEM) and the percentage changein nonfarm payroll employment (NFP) are based on the initialannouncements by the Bureau of Labor Statistics, and both are seasonally adjusted. We convert the announced nonfarm payroll employment data into percentage changes from the previous month’sannounced level. During our sample period, both the unemploymentrate and payroll employment announcements were made at the sametime, typically the first Friday in the month. Each announcement may,however, contain unique information, since they are based on twodifferent surveys. The unemployment data are collected from a surveyof households, conducted and tabulated by the Bureau of the Censusfor the Bureau of Labor Statistics. The payroll employment data arecollected by state agencies from payroll records of employers and aretabulated by the Bureau of Labor Statistics. These employment datawere announced at 9:00 A.M. through March 1982 and at 8:30 A.M. fromApril 1982 to the present.The merchandise trade deficit (MTD) is announced by the ForeignTrade Division of the Department of Commerce, and it representsthe seasonally adjusted monthly trade deficit in billions of dollars(trade surpluses are negative). For most of the sample period, theseannouncements give information on the preceding month’s deficit.Starting in March 1987, the announcements were delayed severalweeks. So, an announcement in March, for example, would give information on January’s trade deficit. Between February 1979 and November 1983, the announcements were made at 2:30 P.M., and in December 1983 it was made at 9:30 A.M. Since January 1984, theannouncements have been made at 8:30 A.M.4We also use the 10-year Treasury-bond yield in the term and default spreads, replacing the Aaayield. The test results reported in the next section ate qualitatively the same using these alternativedefinitions.688The data on inflation are seasonally adjusted monthly percentagechanges in the Consumer Price Index (CPI) and Producer Price Index(PPI) as announced by the Bureau of Labor Statistics. Beginning inFebruary 1978, we use the CPI-U (all urban consumers), consistentwith the MMS expectations data. The PPI series corresponds to allfinished goods, again consistent with the MMS expectations data. ThePPI and CPI announcements were made on various days near themiddle of each month. The PPI announcement is, however, madeearlier in the month than the CPI announcement. With three exceptions, the inflation announcements were made before the stock marketopened, specifically at 9:00 A.M. before March 1982 and at 8:30 A.M.from April 1982 to the present.5The money stock data consist of seasonally adjusted weekly percentage changes in M1, as announced in the Federal Reserve’s H.6release. We convert the M1 data into percentage changes from theprevious week’s announced level. Before January 31, 1980, theannouncements were made on Thursdays at 4:10 P.M., and they corresponded to changes in “old M1.” Then, the announcements weremade at 4:10 P.M. on Fridays, and they corresponded first to Ml-B andthen to MI, where this latter M1 is equivalent to M1-B.6 Beginningon November 29, 1982, money announcements were made at 4:15P.M. Starting on February 16, 1984, money announcements wereswitched back to Thursdays, and since March 22, 1984, they havebeen made at 4:30 P.M. Changes in the Federal Reserve’s discountrate and surcharge were announced intermittently with no typicalannouncement day or time.2.3 Expected values of announcementsWe use the survey data compiled by MMS International to form measures of the market’s expectation of economic announcements. ForMl, the survey data start on September 27, 1977. The survey data forthe CPI, PPI, and the unemployment rate begin in November 1977.For industrial production, the data begin in December 1977. For themerchandise trade deficit and nonfarm payroll employment, the survey data begin in February 1980 and February 1985, respectively. Nosurvey data are available for discount rate and surcharge announcements. As a consequence, all such changes are treated as unantici-5The PPI announcements in October 1981 and August 1985 were made at 2:00 P.M., and theFebruary 1979 CPI announcement was made at 2:30 P.M.6Old M1 differs from the current definition mainly in that it excludes “other checkable deposits”at depository institutions. Following the introduction of nationwide NOW accounts in 1981, thiscategory became substantial,689pated.7 Finally, we convert the survey data for M1 and nonfarm payrollemployment into expected percentage changes from the previouslyannounced level.Although not reported here, we subject the survey data to unbiasedness and efficiency tests for the entire sample period and overvarious subsamples [e.g., Pearce and Roley (1985)]. The overall resultsof these tests are mixed. While the survey data are not always unbiasedand efficient, they generally have smaller root-mean-square errorsthan autoregressive models. To correct for any systematic biases, aswell as to update the survey data with new information, we formrevised expectations [e.g., Roley (1983, 1985) and Shiller, Campbell,and Schoenholtz (1983)]. Since the survey can be taken as long asfive business days before an announcement, we use the change inthe three-month Treasury-bill rate over the four business days beforean announcement as the new information proxy. We estimate regression equations for each calendar year to form revised expectations.82.4 Classification of economic statesTo test the hypothesis that the stock market’s response to news variesover business conditions, some classification of different levels ofeconomic activity is required. NBER business cycle turning pointsare one possibility, but they classify the direction of economic activity(i.e., expansion or recession) rather than the level. Unfortunately,widely accepted definitions analogous to NBER reference cycles arenot available for relative levels of economic activity.In this article, we define economic states using several alternativeeconomic variables. For most of the reported results, we use theseasonally adjusted monthly industrial production index, all items(1977 = 100), to define economic states. First, we estimate a trendin the log of industrial production by regressing the actual log ofindustrial production on a constant and a time trend from September1977. Then we add and subtract a constant from the trend, creating7Roley and Troll (1984) also make this assumption. Other researchers. however, attempt to forecastdiscount rate changes. See, for example, Smirlock and Yawitz (1985). Batten and Thornton (1984).and Hakkio and Pearce (1988). We do not use these approaches because they cannot isolate thespecific day in which the change is expected to occur. In contrast to these approaches, Cook andHahn (1988) simply classify changes into unexpected and expected categories based on FederalReserve statements.8When an economic announcementis made before the market opens, the revised expectation isthe within-sample fitted value of the equationwhereis the survey measure,is the 3-month Treasury-bill yield at the close of day t – 1.ei is a random error term. and a, b, and care coefficients. We perform the regressions over calendaryears instead of economic states to avoid possible biases in later tests that examine the effects ofbusiness conditions. We include the last few months of 1977 and the first live months of 1988 inthe 1978 and 1987 calendar years, respectively.690F igu re 1Natural log of industrial production, actual and bounds (trend ± .028)the upper and lower bounds illustrated in Figure 1. We choose theconstant 0.028 so that the log of industrial production is above theupper bound, denoted as “high” economic activity, 25 percent of thetime. The log of industrial production is below the lower bound,indicating “low” economic activity, about 25 percent of the time aswell. “Medium” economic activity is represented by the remainingobservations between the bounds. As we discuss in Section 4.1, theempirical results are not very sensitive to moderate changes in thebounds or different series used to classify the states.3. Empirical Results3.1 Response to economic announcementsWe first examine the impact of new economic information on stockprices, interest rates, and other discount rate proxies without conditioning on the state of the economy. The results for interest rates,the term spread, and the default spread are useful because they provide evidence that economic announcements contain relevant information for financial markets. Although there are over 3800 days inour sample period, we estimate how the markets respond to newsonly for the 932 days on which one or more announcements is made.Our initial estimation uses the following specification:691wherepercentage change in stock prices or change in interestrates (measured in basis points) from business dayt – 1 to business day t1 × 9 vector of unanticipated components of economicannouncements, calculated as1 × 9 vector of economic announcements1 × 9 vector of expected economic announcements1 × 4 vector of day-of-the-week dummy variables forMonday through Thursdayerror terma, b = scalar and 9 × 1 vector of coefficients, respectivelyFollowing Pagan (1984), ordinary least-squares (OLS) estimation ofEquation (2) results in consistent estimates of coefficients and standard errors in the absence of heteroskedasticity. In all tables, however,White’s (1980) procedure is used to calculate standard errors to takepossible heteroskedasticity into account [e.g., French, Schwert, andStambaugh (1987) and Schwert (1989))We report the results for Equation (2) in Table 1 for the September1977-May 1988 sample.9 The first row in the table shows, for example,that the S&P 500 Index falls by 0.1 percent in response to an unanticipated increase in industrial production of 1 percentage point. The loyear bond yield and the three-month Treasury-bill yield increase by5.5 and 9.5 basis points, respectively, in response to this sameannouncement. While interest rates exhibit statistically significantresponses to most of the new economic information, stock prices donot. The S&P 500 Index response coefficient is significant at the 5percent level only for unanticipated components of Ml announcements. These unconditional results are similar to those of other studies using much shorter sample periods [e.g., Pearce and Roley (1985)].109In addition to specification (2). we also obtain results for a specification including the expectedvalues of economic announcements The inclusion of these variables has no effect on theestimated response coefficients, 6, since the measures of unanticipated announced changes,are uncorrelated withby construction…

Stock Prices, News, and BusinessConditionsGrant McQueenBrigham Young UniversityV. Vance RoleyUniversity of WashingtonPrevious research finds that fundamental mac- roeconomic news has little effect onstock prices. We show that after allowing for different stages of the business cycle, astronger relationship between stock prices and news is evident. In addition to stockprices, we examine the effect of real activity news on proxies for expected cash cowsand equity discount rates. We find that when the economy is strong the stock marketresponds negatively to news about higher real economic activity. This negativerelation is caused by the larger increase in discount rates relative to expectedcash flows.Apart from some types of monetary information, there is little empirical evidence tosupport the hypothesis that stock prices respond to macroeconomic news. Schwert (1981)finds that the daily response of stock prices to news about inflation from 1953 to 1978 isweak and slow. Pearce and Roley (1985) use survey data to measure expectations and findthat daily stock prices respond to monetary information between Sep- tember 1977 andOctober 1982, but news about the consumer price index, unemployment, and industrialWe are grateful to Lynda S. Livingston and Steven R. Thorley for research assistance, to Stephen J. Brown (the editor),Wayne E. Fetson, Douglas K. Pearce, J. Michael Pinegar. James M. Poterba. Simon M. Wheatley, an anon- ymous referee,and seminar participants at the National Bureau of Economic Research and the University of Washington for helpfulcomments, and to the Center for the Study of Financial Management, University of Washington, and Seafirst Bank, forresearch support. Address correspondence to V. Vance Roley, Department of Finance, DJ-10, University of Washington, Seattle,WA 98195.production have no significant effect on prices. Hardouvelis (1987)considers a somewhat broader set of variables through August 1984and concludes that stock prices respond primarily to monetary news.Finally, Cutler, Poterba, and Summers (1989) use vector autoregressions to measure news about macroeconomic time series from 1871to 1986. They conclude that less than one-third of the monthly returnvariance can be explained from these sources.’Each of these studies assumes that investors’ response to news isthe same over different stages of the business cycle. For instance,Cutler, Poterba, and Summers (1989) implicitly assume that a positivesurprise in industrial production at the end of the Great Depressionevokes the same response as a surprise in late 1969, after nearly adecade of expansion. A positive surprise in industrial productionduring the depression could indicate the end of the depression andhigher forecasts of firms’ cash flows. Such an announcement wouldlikely be “good news” for the stock market. In late 1969, with lowunemployment and factories running near full capacity, a positivesurprise in industrial production may result in fears of an overheatingeconomy, inflation, and possible efforts by policymakers to increasereal interest rates. Such an announcement could then be “bad news”for the stock market. If the same type of news is considered good insome states of the economy and bad in others, the response coefficienton the surprise in previous studies will be biased toward zero.2The popular press uses this good news/bad news story to interpretdaily stock price movements. For example, on February 4, 1983, after16 months of recession, the Labor Department reported that the unemployment rate fell to 10.4 percent. This represented a rate of 0.2 or0.3 percentage points below what was expected. This news was usedby the media to explain the 13.25-point jump in the Dow Jones Industrial Average, and prompted the Chairman of the Council of EconomicAdvisers, Martin Feldstein, to comment that “a recovery is eitherbeginning or already here” ( Wall Street Journal, February 7, 1983).In contrast, on November 4, 1988, after six years of expansion, theLabor Department reported that the unemployment rate fell to 5.3percent, matching a 14-year low. This represented a rate of 0.1 or 0.2percentage points below what was expected. The media’s interpreta1Chen, Roll, and Ross (1986) also investigate whether monthly stock returns covary with variousmacroeconomic variables. They again find that the explanatory power is low. The main focus oftheir study, however, is whether the covariance of economic variables with stock returns can explainex ante returns.2Several recent studies find significant effects from business conditions on stock returns. Ferson andMerrick (1987). for example, find shifts in consumption-based asset pricing parameters acrossstages of the business cycle measured by recession versus nonrecession. Fama and French(1989) and Fama (1990) consider term-premium and default-risk-premium variables asdeterminants of equity discount rates. They suggest that the term premium is related to NBERbusiness cycles, while the risk premium is related to business conditions over longer periods.684tion in this instance was “bond market investors reacted with gloom,sending interest rates higher on fears of tighter Fed policy. The stockmarket also fell” ( Wall Street Journal, November 7, 1988). The problem with this type of evidence, however, is that it is anecdotal andlargely after the fact.In this article we examine whether the response of stock prices tomacroeconomic news varies over different stages of the business cycle.By allowing the response to vary over different states of the economy,we can test the good news/bad news story and provide unbiasedestimates of the effects of fundamental information about the economy. We study daily percentage changes in closing values of theStandard & Poors 500 Index and several variables related to equitydiscount rates and cash flows. By considering these other variables,we can investigate the sources of any business-condition effect onthe response of stock prices.Following this introductory section, we present in Section 1 a simple theoretical framework to consider how news affects stock pricesand how this effect can vary over different stages of the business cycle.We describe the data in Section 2. In Section 3 we present the empirical results. We consider the robustness of the results in Section 4,and we summarize the main conclusions in Section 5.1. Theoretical FrameworkA common model that links stock prices to information posits thatstock prices equal the present discounted value of rationally forecasted future dividends. This model can be represented aswhere P, is the price of the stock at timedenotes the mathematical expectation conditional on information available at time tis the dividend paid at timeis the stochasticdiscount factor for cash flows that occur at timeEconomic announcements affect daily share price movements if thenew information revealed by announcements affects either expectations of future dividends or discount rates or both. The new information is represented by the difference in the announced value onday t + 1 and the expected value as of day t. Consequently, theunanticipated component of an announcement on day t + 1 is uncorrelated with information available on day t. The information set,includes past announcements of other economic variables, soannouncement surprises are uncorrelated under rational expectationsif they are made on different days. Combining daily stock-price changes685with announcement surprises on different days allows us to isolatethe effects of individual economic variables.1.1 Impact of real economic activity surprisesWe need not expect that real economic activity surprises will affectcash flows and discount rates in the same way across different statesof the economy. As a result, stock prices may well react differently tosurprises of this nature, depending on whether the economy is operating below capacity. When the economy is booming, for example, areal economic activity surprise could result in a larger increase indiscount rates than cash flows, causing stock prices to fall. In thiscase, high capacity utilization and employment may constrain furtherincreases in output and, consequently, cash flow in the absence ofnew investment in plant and equipment.The announcement effect we examine corresponds to the disclosure in month t of production growth that already occurred in montht – 1. Information about the previous month is relevant in that it maychange expectations about the future. That is, consistent with Fama(1990) and Schwert (1990), the information provided by an industrialproduction announcement causes stock prices to respond if this information causes revisions in expected future industrial production.1.2 Impact of other economic announcement surprisesWe also consider possible asymmetric effects of announcement surprises other than those related to real economic activity. This othereconomic information is, however, less closely related to the possiblebusiness-conditions effects discussed above. The announcements weconsider are for foreign trade, inflation, and money. We briefly discusseach in turn.First, foreign trade deficit announcements have at times receivedconsiderable attention in the popular press. For the 1979-1984 period,however, Hardouvelis (1987) does not find any significant effects onstock prices. We update his sample and test for varying effects overdifferent economic states.Second, following the empirical studies of Nelson (1976) and Famaand Schwert (1977), a number of studies estimate a significant negative relationship between inflation and stock returns. Among these,Feldstein (1980) argues that the tax treatment of depreciation andinventories results in lower real after-tax corporate profits and, hence,lower stock prices during times of inflation. Fama (1981), Geske andRoll (1983), and Kaul (1987) explain the negative relationship byappealing to real output effects. In terms of inflation announcementsurprises, the significance of the stock-price response is mixed [e.g.,Pearce and Roley (1985) and Hardouvelis (1987)]. We again extend686these announcement studies by lengthening the sample and by allowing business-condition-dependent responses.Third, Pearce and Roley (1983, 1985), Cornell (1983), and Hardouvelis (1987) find that stock prices respond significantly to moneyannouncement surprises. Varying responses over different monetarypolicy regimes are tested in these studies, but possible businessconditions effects are not considered.3 We estimate the stock-priceresponse not only to money announcements but also to Federal Reservediscount rate changes, over different economic states.2. DataOur sample period begins in September 1977 and ends in May 1988.The start of the sample period coincides with the initial availabilityof survey data from Money Market Services International (MMS). Wediscuss the robustness of the results using alternative sample periodsand expectation measures in Section 4.3.2.1 Asset prices and yieldsWe use daily percentage changes in the closing value of the S&P 500Index to estimate the response of stock prices to new macroeconomicinformation. For economic announcements occurring either beforeor while the stock market is open, we use the percentage change inthe index from the previous business day’s closing price to the closingprice on that day. For announcements made after the stock market isclosed, we use the percentage change in the index from that day’sclosing quote to the next business day’s closing quote. Throughoutthe sample, the stock market closed at 4:00 P.M. EST. (We use ESTfor all closing and announcement times.)To measure the response of equity discount rates to new information, we consider several proxies. These include daily changes inthe three-month Treasury-bill and lo-year Treasury-bond yields. Following Fama and French (1989) and Fama (1990), we also includevariables denoted as the term spread and the default spread as equitydiscount rate proxies. We represent the term spread by Moody’s Aaacorporate bond yield minus the three-month bill yield, and the default3Given the evidence that both short- and long-term interest rates respond differently to moneyannouncement surprises over different Federal Reserve policy regimes [e.g., Roley (1983, 1986).Cornell (1983), and Roley and Walsh (1985)], another potentially interesting hypothesis is thatstock prices respond differently to economic news over these regimes. For the October 1979 andOctober 1982 regimes, however, Pearce and Roley (1983, 1985) and Hardouvelis (1987) find nosignificant difference in the stock market’s response to money surprises. We nevertheless investigatethe effects of the monetary policy regimes in October 1979, October 1982, and February 1984. andthe hypothesis that the stock market’s response is the same across regimes for our set of economicannouncements can be rejected only at the 25 percent significance level. Consequently, we do notexamine the effects of monetary policy regimes further.687spread by Moody’s Baa corporate bond yield minus the Aaa yield.These yield data are from the Federal Reserve’s H.15 release, andthey correspond to yields based on bid prices prevailing at 3:30 P.M.42.2 Economic announcementsVirtually all of the economic announcements are well-publicizedevents with regular schedules. Data on industrial production (IP) areinitially released, seasonally adjusted monthly percentage changes inthe Federal Reserve Industrial Production Index, all items. BetweenJanuary 1979 and October 1985, the announcements were made at9:30 A.M.; since October 1985, at 9:15 A.M. Before 1979, the industrialproduction press releases give no specific announcement time, statingonly “for immediate release.” However, the announcements weremade before the market opened for our sample.Data on the unemployment rate (UNEM) and the percentage changein nonfarm payroll employment (NFP) are based on the initialannouncements by the Bureau of Labor Statistics, and both are seasonally adjusted. We convert the announced nonfarm payroll employment data into percentage changes from the previous month’sannounced level. During our sample period, both the unemploymentrate and payroll employment announcements were made at the sametime, typically the first Friday in the month. Each announcement may,however, contain unique information, since they are based on twodifferent surveys. The unemployment data are collected from a surveyof households, conducted and tabulated by the Bureau of the Censusfor the Bureau of Labor Statistics. The payroll employment data arecollected by state agencies from payroll records of employers and aretabulated by the Bureau of Labor Statistics. These employment datawere announced at 9:00 A.M. through March 1982 and at 8:30 A.M. fromApril 1982 to the present.The merchandise trade deficit (MTD) is announced by the ForeignTrade Division of the Department of Commerce, and it representsthe seasonally adjusted monthly trade deficit in billions of dollars(trade surpluses are negative). For most of the sample period, theseannouncements give information on the preceding month’s deficit.Starting in March 1987, the announcements were delayed severalweeks. So, an announcement in March, for example, would give information on January’s trade deficit. Between February 1979 and November 1983, the announcements were made at 2:30 P.M., and in December 1983 it was made at 9:30 A.M. Since January 1984, theannouncements have been made at 8:30 A.M.4We also use the 10-year Treasury-bond yield in the term and default spreads, replacing the Aaayield. The test results reported in the next section ate qualitatively the same using these alternativedefinitions.688The data on inflation are seasonally adjusted monthly percentagechanges in the Consumer Price Index (CPI) and Producer Price Index(PPI) as announced by the Bureau of Labor Statistics. Beginning inFebruary 1978, we use the CPI-U (all urban consumers), consistentwith the MMS expectations data. The PPI series corresponds to allfinished goods, again consistent with the MMS expectations data. ThePPI and CPI announcements were made on various days near themiddle of each month. The PPI announcement is, however, madeearlier in the month than the CPI announcement. With three exceptions, the inflation announcements were made before the stock marketopened, specifically at 9:00 A.M. before March 1982 and at 8:30 A.M.from April 1982 to the present.5The money stock data consist of seasonally adjusted weekly percentage changes in M1, as announced in the Federal Reserve’s H.6release. We convert the M1 data into percentage changes from theprevious week’s announced level. Before January 31, 1980, theannouncements were made on Thursdays at 4:10 P.M., and they corresponded to changes in “old M1.” Then, the announcements weremade at 4:10 P.M. on Fridays, and they corresponded first to Ml-B andthen to MI, where this latter M1 is equivalent to M1-B.6 Beginningon November 29, 1982, money announcements were made at 4:15P.M. Starting on February 16, 1984, money announcements wereswitched back to Thursdays, and since March 22, 1984, they havebeen made at 4:30 P.M. Changes in the Federal Reserve’s discountrate and surcharge were announced intermittently with no typicalannouncement day or time.2.3 Expected values of announcementsWe use the survey data compiled by MMS International to form measures of the market’s expectation of economic announcements. ForMl, the survey data start on September 27, 1977. The survey data forthe CPI, PPI, and the unemployment rate begin in November 1977.For industrial production, the data begin in December 1977. For themerchandise trade deficit and nonfarm payroll employment, the survey data begin in February 1980 and February 1985, respectively. Nosurvey data are available for discount rate and surcharge announcements. As a consequence, all such changes are treated as unantici-5The PPI announcements in October 1981 and August 1985 were made at 2:00 P.M., and theFebruary 1979 CPI announcement was made at 2:30 P.M.6Old M1 differs from the current definition mainly in that it excludes “other checkable deposits”at depository institutions. Following the introduction of nationwide NOW accounts in 1981, thiscategory became substantial,689pated.7 Finally, we convert the survey data for M1 and nonfarm payrollemployment into expected percentage changes from the previouslyannounced level.Although not reported here, we subject the survey data to unbiasedness and efficiency tests for the entire sample period and overvarious subsamples [e.g., Pearce and Roley (1985)]. The overall resultsof these tests are mixed. While the survey data are not always unbiasedand efficient, they generally have smaller root-mean-square errorsthan autoregressive models. To correct for any systematic biases, aswell as to update the survey data with new information, we formrevised expectations [e.g., Roley (1983, 1985) and Shiller, Campbell,and Schoenholtz (1983)]. Since the survey can be taken as long asfive business days before an announcement, we use the change inthe three-month Treasury-bill rate over the four business days beforean announcement as the new information proxy. We estimate regression equations for each calendar year to form revised expectations.82.4 Classification of economic statesTo test the hypothesis that the stock market’s response to news variesover business conditions, some classification of different levels ofeconomic activity is required. NBER business cycle turning pointsare one possibility, but they classify the direction of economic activity(i.e., expansion or recession) rather than the level. Unfortunately,widely accepted definitions analogous to NBER reference cycles arenot available for relative levels of economic activity.In this article, we define economic states using several alternativeeconomic variables. For most of the reported results, we use theseasonally adjusted monthly industrial production index, all items(1977 = 100), to define economic states. First, we estimate a trendin the log of industrial production by regressing the actual log ofindustrial production on a constant and a time trend from September1977. Then we add and subtract a constant from the trend, creating7Roley and Troll (1984) also make this assumption. Other researchers. however, attempt to forecastdiscount rate changes. See, for example, Smirlock and Yawitz (1985). Batten and Thornton (1984).and Hakkio and Pearce (1988). We do not use these approaches because they cannot isolate thespecific day in which the change is expected to occur. In contrast to these approaches, Cook andHahn (1988) simply classify changes into unexpected and expected categories based on FederalReserve statements.8When an economic announcementis made before the market opens, the revised expectation isthe within-sample fitted value of the equationwhereis the survey measure,is the 3-month Treasury-bill yield at the close of day t – 1.ei is a random error term. and a, b, and care coefficients. We perform the regressions over calendaryears instead of economic states to avoid possible biases in later tests that examine the effects ofbusiness conditions. We include the last few months of 1977 and the first live months of 1988 inthe 1978 and 1987 calendar years, respectively.690F igu re 1Natural log of industrial production, actual and bounds (trend ± .028)the upper and lower bounds illustrated in Figure 1. We choose theconstant 0.028 so that the log of industrial production is above theupper bound, denoted as “high” economic activity, 25 percent of thetime. The log of industrial production is below the lower bound,indicating “low” economic activity, about 25 percent of the time aswell. “Medium” economic activity is represented by the remainingobservations between the bounds. As we discuss in Section 4.1, theempirical results are not very sensitive to moderate changes in thebounds or different series used to classify the states.3. Empirical Results3.1 Response to economic announcementsWe first examine the impact of new economic information on stockprices, interest rates, and other discount rate proxies without conditioning on the state of the economy. The results for interest rates,the term spread, and the default spread are useful because they provide evidence that economic announcements contain relevant information for financial markets. Although there are over 3800 days inour sample period, we estimate how the markets respond to newsonly for the 932 days on which one or more announcements is made.Our initial estimation uses the following specification:691wherepercentage change in stock prices or change in interestrates (measured in basis points) from business dayt – 1 to business day t1 × 9 vector of unanticipated components of economicannouncements, calculated as1 × 9 vector of economic announcements1 × 9 vector of expected economic announcements1 × 4 vector of day-of-the-week dummy variables forMonday through Thursdayerror terma, b = scalar and 9 × 1 vector of coefficients, respectivelyFollowing Pagan (1984), ordinary least-squares (OLS) estimation ofEquation (2) results in consistent estimates of coefficients and standard errors in the absence of heteroskedasticity. In all tables, however,White’s (1980) procedure is used to calculate standard errors to takepossible heteroskedasticity into account [e.g., French, Schwert, andStambaugh (1987) and Schwert (1989))We report the results for Equation (2) in Table 1 for the September1977-May 1988 sample.9 The first row in the table shows, for example,that the S&P 500 Index falls by 0.1 percent in response to an unanticipated increase in industrial production of 1 percentage point. The loyear bond yield and the three-month Treasury-bill yield increase by5.5 and 9.5 basis points, respectively, in response to this sameannouncement. While interest rates exhibit statistically significantresponses to most of the new economic information, stock prices donot. The S&P 500 Index response coefficient is significant at the 5percent level only for unanticipated components of Ml announcements. These unconditional results are similar to those of other studies using much shorter sample periods [e.g., Pearce and Roley (1985)].109In addition to specification (2). we also obtain results for a specification including the expectedvalues of economic announcements The inclusion of these variables has no effect on theestimated response coefficients, 6, since the measures of unanticipated announced changes,are uncorrelated withby construction….