Loss Aversion Discount before Earnings Announcements Shai Levi and

Loss-Aversion Discount before Earnings Announcements Shai Levi and Xiao-Jun Zhang* Haas School of Business, UC Berkeley July 2008 Abstract: This study tests and finds that stock prices before earnings announcements reflect investor aversion to negative news. Specifically, we find that there is a one-sided returns reversal at earnings announcements. Earnings announcements that are preceded by price decreases trigger positive returns reactions, on average, while those preceded by price increases have zero announcement returns. Furthermore, the price decreases before earnings announcements are associated with a higher likelihood of negative news, and reflect an extra discount, one that is consistent with investor aversion to loss in share value. The phenomenon can also reflect, as we show, compensation (discount) that riskaverse market makers require when buying stocks before earnings announcements. Taken together, the evidence suggests that there is asymmetric reaction to negative and positive information on the upcoming earnings announcements, or in specific that negative news looms larger than positive news. * Contact information: Shai Levi can be reached at slevi@haas.berkeley.edu, and Xiao-Jun Zhang at xzhang@haas.berkeley.edu; mailing address for both is 545 Student Services Building, University of California at Berkeley, CA 94720-1900. We thank seminar participants in Barclays Global Investors, Berkeley, Hebrew University, and Interdisciplinary Center in Herzliya for their comments. 1 1. Introduction When investors suspect that the coming earnings will fall short of expectations, they will, in all likelihood, bid the firm’s stock price down. This is not a surprise, since stock prices usually equal the present value of expected earnings (or cash flows), and lower expected earnings should lead to lower prices. But does higher probability of negative earnings news also cause investors to assign stocks higher discount rates? The answer can be yes if investors are loss-averse. We empirically test this possibility. Loss-aversion, which is a feature of Kahneman and Tversky’s (1979) prospect theory, suggests that people are more sensitive to the possibility of losses than to the prospect of gains. In the context of investing, investors are likely to be averse to losses in their total wealth (e.g, Benartzi and Thaler 1995). Aversion to losses in value of individual stocks is also a plausible alternative that might explain return behavior, as, for example, shown by Barberis and Haung (2001). Using this idea, we empirically test whether investors exhibit loss-aversion when they set stock prices before earnings announcements. If investors are loss-averse over a change in value of individual stocks, they may be more sensitive to the possibility of negative earnings news than to the prospects of positive earnings news. We use the days before earnings announcements as the setting to test our hypothesis. We find that price decreases before earnings announcements are a result not only of an increase in the likelihood of a negative earnings surprise, but also of an additional discount that we believe reflects investor aversion to the price consequences of the negative-news possibility. Specifically, we find that stock prices increase at earnings announcements that were preceded by a week of negative stock returns. In contrast, stock prices remain 2 unchanged at announcements that come after a week of positive returns. This returns pattern indicates that investors are averse to the possibility of negative news. As we show, price declines in the week before earnings announcements are associated with a higher probability of negative earnings surprises. To be more precise, the likelihood of missing an analyst forecast is 4% to 5% higher for announcements preceded by a week of negative returns. Therefore, investors seem to bid the stock price down before announcements, when they believe that firms are more likely to miss analyst forecasts. The price decline, however, also reflects an additional discount. This discount is evident from the price increase at the earnings announcements that were preceded by price declines. Namely, the discount is removed and prices increase once earnings are reported and the loss uncertainty is resolved. As further support for our loss-aversion hypothesis, we find that investors use higher discount rates when the expected losses are higher. We assume that investors are loss-averse over changes in the value of individual stocks they own. If indeed investors dread loss, then the discount should also be a function of the potential impact that the negative news will have on the firm’s value. In our research setting, the wealth loss potential is the expected price drop at the earnings announcement. We use market-tobook as a proxy of this expected loss. High market-to-book firms that report lower than expected earnings usually experience big price drops at the announcement (e.g., Sloan and Skinner 2002). Therefore, higher market-to-book implies higher expected loss. Consistent with our hypothesis, we find that among the stocks experiencing a price decline before announcements, higher market-to-book stocks have higher returns on the announcement. In contrast, among the stocks experiencing a price increase before 3 announcements, announcement returns are zero across different market-to-book portfolios. These results indicate that, for firms that can experience greater price declines on negative news, investors use higher discounts in response to an increase in the probability of negative news. Our findings can be described as a one-sided return reversal around earnings announcements. When earnings announcements follow a week of negative returns, the announcement returns will be positive. When announcements follow a week of positive returns, the announcement returns will be, on average, zero. Short-term return reversals have been documented in prior studies (e.g., Jagadeesh 1990; Lehmann 1990), and are not unique to our earnings announcement setting. However, we show that the one-sided return reversal at earnings announcement is about twice the size of the reversal on other days.1 We also show that our one-sided returns reversal is not the result of the wellknown bid-ask bounce. Predictable stock return patterns, as we document in this paper, naturally raise the possibility of inefficiency in the stock market. In terms of returns magnitude, buying stocks that experienced a price decline before earnings announcements would yield a return of approximately 0.5% per announcement, when buying the stock a day before the announcement and selling it a day or two after the announcement. A portfolio that systematically follows this buying strategy would have had average abnormal (alpha) returns of 4.6% per month, after controlling for the three factors of Fama and French 1 Our analysis is unique, in that it demonstrates one-sided returns reversals that can be a result of lossaversion. This one-sided reversal, however, can exist on other days that are unrelated to the earnings announcement. After all, the two ingredients necessary for loss-aversion – uncertainty and a possible loss in stock value – exist on other days. Nonetheless, the reversal is stronger at earnings announcements, as should be expected. The resolution of uncertainty is usually more complete at earnings announcements, and, consequently, greater mitigation of loss-aversion and larger increase in price are expected then. 4 (1993) and a momentum factor (Carhart 1997). The question remaining is whether these returns leave profit opportunity for diversified investors, who are less averse to losses. The answer hinges on transaction costs. Taking into account the effective bid-ask spreads, we find that traders probably could not have exploited the predictable positive returns pattern that we report. The returns pattern that we document only suggests that the average investor who sets prices before earnings announcements is loss-averse. Our results can be interpreted differently when taking the view that stock prices are determined by investors who seek liquidity and market makers who provide it (e.g., Grossman and Miller 1988), instead of by trades between willing buyers and sellers. In this case, prices can decrease before earnings announcements when liquidity traders seek to sell their stocks. Risk-averse market makers will ask for compensation (discount) when buying the stock and increasing their position before announcement periods, which are usually more volatile. The discount will be alleviated once earnings are announced, and consequently prices will increase, as we observe.2 To further test this alternative explanation, we examine the effects of price pressures before earnings announcements on the returns reversal at the announcements. Similar to Campbell, Grossman, and Wang (1993), we define sell pressures as price decreases on high trading volumes, and assume that market makers will buy the stocks to satisfy liquidity needs. We find that stronger sell pressures before the announcements are followed by higher announcement returns (reversal). This result supports the idea that the discount in price before the announcement is the result of the compensation that market 2 Campbell, Grossman, and Wang (1993), for example, provide a model predicting that liquidity demand by traders will cause short-term returns reversal, such as the one we observe. 5 makers demand for increasing their stock position before earnings announcements. Buy pressures, on the other hand, are not followed by returns reversals, and the announcement returns are zero across our different buy pressure groups. Note that, even when taking a microstructure approach to pricing, loss-aversion can still play a role. We cannot observe the motives of traders who sell the stocks before earnings announcements. One can assume that the sellers have pressing liquidity needs, as inventory models usually assume (e.g., Grossman and Miller 1988), but sellers may also unload the stock because they dread the possibility of negative news. The fact that the reversals are one-sided, and do not exist after buy pressures, casts doubt on whether these inventory models can by themselves explain the phenomenon.3 This paper contributes to the literature that considers the impact of loss-aversion on stock prices. Theoretical studies have demonstrated that aversion to losses in total wealth can explain the return behavior of the overall market. For example, Benartzi and Thaler (1995) show that the premium puzzle (the fact that stocks have outperformed bonds by a surprisingly large margin) might be due to loss-averse investors who are reluctant to allocate much of their wealth to stocks. Our paper, however, focuses on investors’ aversion to losses on individual stocks. Barberis and Haung (2001) show that a model with investors who are averse to losses in individual stocks can explain observed regularities, such as the time-series predictability of returns. Yet, as Barberis and Huang 3 Note that inventory models as the one in Grossman and Miller (1988) predict that buy pressures, and not only sell pressures, will cause short-term return reversals. Our findings on lack of reversals after buy pressures can, for example, indicate that specialists carry excessive inventories of the stock before earnings announcements - so they do not mind selling but are averse to buying more stocks. Alternatively, it is possible that it is easier for investors who act as market makers, and provide liquidity before earnings announcements, to find assets that are comparable to the ones they sell than to diversify the risk associated with buying a stock before its earnings announcement. Finally, the phenomenon could be due to loss-aversion of the liquidity traders. Namely, traders are more eager to sell (than to buy) before earnings announcement and are willing to complete the sell transaction at lower than usual bid prices. 6 point out, other models that do not have a loss-aversion feature can also explain these phenomena (e.g., Barberis, Shleifer, and Vishny 1998, and Berk, Green, and Naik 1999). Our findings contribute to this literature by demonstrating that stock prices are affected by investor aversion to losses in individual stocks. This paper also contributes to the literature that documents the asymmetric investor reaction to negative earnings news versus positive earnings news. For example, Skinner and Sloan (2002) find that when earnings miss analyst forecasts, the stock price drops more than the stock price rises when firms beat forecasts by an equivalent magnitude of earnings surprise. Sloan and Skinner assign this asymmetric reaction to investors’ biased expectations, or more specifically over-optimism regarding high-growth firms that reverses at the negative earnings surprise. In a different setting, we demonstrate the effect that expectations for negative surprise have on the discount that investors use is also asymmetric. Examining stock prices before earnings announcements, we show that price declines are associated with a discount that reflects investor aversion to negative news. Any premium (or discount) is not detected in the case of price increases before announcements. Lastly, the phenomenon we document also has implication for accounting. If bad news tends to hurt investors more, as we show, then it may be a sensible practice to inform investors about bad news in a more timely fashion (Watts 2003). This lends additional support for the documented asymmetric timeliness in recognizing good versus bad news in accounting (Basu 1997). The rest of this paper is organized as follows: Section 2 analyzes the association between price declines before earnings announcements, and the probability of a negative 7 earnings surprise; Section 3 contains the results of our main hypothesis (loss-aversion); Section 4 examines the relationship between the loss-aversion discount and potential wealth loss; Section 5 compares the return reversal around earnings announcement to returns reversals on other days; Section 6 controls for the effects of bid-ask bounce and provides a robustness check; Section 7 analyzes the transaction costs associated with trading on the predictable returns pattern that we find; Section 8 tests an alternative explanation to the phenomenon: risk-averse market makers ask for compensation (discount) to ease sell pressures and buy stocks before earnings announcements; and Section 9 concludes. 2. Decrease in stock prices before earnings announcements Our main research question is whether the belief that negative earnings news is more likely to lead investors to demand higher expected returns, a phenomenon consistent with the effect of loss-aversion on stock prices (e.g, Benartzi and Thaler 1995; Barberis and Haung 2001). Specifically, we test the existence of a loss-aversion discount by examining the pricing of stocks that had decreases in price before earnings announcement. We focus on price decreases occurring about a week before the earnings announcement. In this short time window, the price decreases are more likely to be driven by negative revisions in investor beliefs about the coming earnings announcement. Specifically, in this section we show that price decreases in the week before earnings announcements are related to a higher likelihood of negative earnings surprises. In the next section, we will test whether these price decreases before the earnings 8 announcements also reflect an increase in the discount rate, due to investor aversion to the possibility of negative news (loss-aversion discount). Investors have the opportunity and incentives to collect private information before earnings announcement, and there is evidence suggesting that investors in fact attain information in the days prior to announcements. Bagnoli, Beneish and Watts (1999), for example, find that unofficial forecasts of earnings, referred to as ‘whisper’ numbers, circulate among traders and investors in the days preceding earnings announcements. These whisper numbers reflect private information that market participants have attained either through discussions with stockbrokers, financial analysts, and investor relations departments of companies, or as a result of private forecasting techniques. Bagnoli et al. show that whisper forecasts are, on average, more accurate than the official analysts’ forecasts, and that profits can be made by trading on the information. This evidence illustrates the incentives that investors have to acquire private information on earnings announcements, and also the fact that investors obtain information in the days before the announcements. The information that investors collect in the days before earnings announcement is mostly unobservable; therefore, we use an indirect test to examine whether decreases in prices before earnings announcement are associated with negative information about the coming announcements. Specifically, we test whether the likelihood that earnings will fall short of analyst forecasts (bad news) is greater in announcements preceded by negative returns than in those preceded by positive returns. The sample includes all the quarterly earnings announcements made between 1984 and 2004: (a) whose earnings announcement date is available on the Industrial 9 Quarterly Compustat; (b) made by firms with corresponding stock return data on CRSP, for a period starting 6 trading days before the earnings announcement and ending 2 days before the earnings announcement; and (c) for which there is at least one analyst earnings forecast 90 calendar days before the earnings announcement on I/B/E/S detailed tape. We also require that firms have a positive book value of equity, in order to attain a meaningful market-to-book variable for our analysis.4 168,683 quarterly earnings announcements answer these criteria. As shown in Table 1, earnings announcements that are preceded by negative returns fall short of analyst forecasts (bad news) more frequently than announcements preceded by positive returns. Specifically, 83,922 earnings announcements were made by firms that experienced negative cumulative size-adjusted returns in a five-day period, from the 6th trading day through the 2nd trading day before the earnings announcement. The remaining 83,274 earnings announcements had positive (non-negative) size-adjusted returns in the same five-day period. We find that 40.4% of earnings announcements in the negative pre-announcement returns group fell short of analyst forecasts, while only 34.7% of the announcements in the positive pre-announcement returns group missed analyst forecasts. As shown in the table, the difference between the proportions of negative earnings surprises of the two groups is statistically significant (the two samples’ t-test is 24.10). The earnings surprise is calculated from the I/B/E/S detailed tape, and it is 4 The requirement that firms have positive book value of equity results in the exclusion of 2,982 earnings announcements from the sample. Note that the results (which do not require a market-to-book variable) stay similar when these announcements are included. 10 the difference between actual earnings and the median of forecasts issued less than 90 days before the announcement.5 The univariate results above indicate that negative returns in the week before earnings announcements are associated with higher likelihood of negative earnings surprises. This indicates that returns before the announcements are driven by information about the coming earnings announcement. To further check the association between negative returns before earnings announcement and negative earnings surprises, we conduct a multivariate (logistic) analysis. This analysis controls for differences in firm characteristics that might also affect the frequency of negative earnings surprises. As shown in Table 1, the negativereturns-before-earnings group includes firms with slightly smaller market value and lower market-to-book than the positive-returns-before-announcement group. Smaller firms with lower market-to-book are less likely to meet analyst forecasts (e.g. Matsumoto 2002), even before the factor of negative returns before announcement is taken into consideration. Therefore, we test, below, the marginal association of negative preannouncement returns with negative earnings surprise after controlling for size, marketto-book, and industry membership. We assess the following logistic regression: Pr ob( Miss = 1) = F ( β 0 + β1 D Neg Re t it + β 2 Sizeit + β 3 MBit + ∑ δ j Industry _ Dummyit + ∑ λ k QtrYear _ Dummyit + ε it ) , j k (1) where the dependent variable equals 1 when announced earnings miss the analyst consensus forecast, which we calculate as the median of forecasts issued less than 90 5 When using a 30-day window, instead of a 90-day window, for the calculation of a consensus forecast, we get similar results. Announcements preceded by negative returns fall short of expectations 33.8% of the time, and announcements preceded by positive returns fall short 28.9% of the time; i.e., there is still a difference of 5% between the groups. The sample size, however, drops to about half its current size. 11 days before the announcements; DNetRet is a dummy variable that equals 1 when sizeadjusted returns are negative for the five-day period from the 6th trading day through 2nd trading day before the earnings announcement; Size is the market value at the 6th trading day before the announcement; and M/B is the market value, above, divided by the book value of equity in the previous quarter. The M/B variable is winsorized at the top and bottom one percentiles. Following Matsumoto’s finding that industry membership is correlated with the frequency of meeting analyst forecasts, equation (1) includes dummies for industries that are based on the two-digit SIC industry membership. Dummies are also included for each calendar quarter to control for time-series changes and for cross-sectional correlation effects. The results of the estimation are presented in Table 2. Negative preannouncement returns are associated with higher likelihood of negative earnings news, after size, market-to-book, and industry membership are taken into account. Specifically, the coefficient of DNegRet (β1) is positive and significant, and its marginal effect on the probability of missing analyst forecast is 0.0416. This means that, after controlling for size, market-to-book, and industry membership, announcements that were preceded by negative returns are still 4.16% more likely to miss analyst forecast than announcements preceded by positive returns. The coefficients of M/B and Size are negative, as expected, and significant. In sum, our findings in this section indicate that negative stock returns in the week before earnings announcements are associated with greater likelihood of negative 12 earnings surprise. This result indicates that these price decreases are driven, at least in part, by negative information that investors attain about the coming announcement.6 3. Loss-aversion and earnings announcement returns In this section, we examine whether the decrease in prices before earnings are also caused by an increase in discount rates. We form two portfolios, one containing the stocks that had negative returns, and the other with the stocks that had positive returns, in the week before earnings announcements. We then examine the subsequent returns of each portfolio at the earnings announcements. Positive announcement returns for firms that experienced a price decline before the announcement will support the existence of the loss-aversion discount. If investors discounted stocks in response to an increase in the likelihood of lower earnings, once earnings are announced and the uncertainty is resolved, this loss-aversion discount will fade, and stock prices will increase.7 Our above portfolio test assumes that markets are efficient. Under this assumption, investors have unbiased expectations, and price changes after the portfolios’ formation (in our case, at the time of earnings announcement) can only be a result of shifts in discount rates. The portfolio test is performed using the calendar-time approach (see, e.g., Mitchell and Stafford 2000; Fama 1997), which mitigates possible biases introduced by 6 7 Note that we proxy for the content of the announcement using the standard earnings surprise, where earnings announcements can contain other information that can impact stock prices. Additionally, decrease in returns before the announcement can be a result of information unrelated to the announcement. Still, we believe that our results indicate that the price declines are associated with negative information on the coming earnings announcement. A well-specified pricing model would have allowed us to directly breakdown the price decline into: a) a decrease in expectations for future cash flows or earnings; and b) an increase in the discount rate. Unfortunately, existing pricing models are noisy and imprecise (e.g., Fama and French 1997), and such a test is not likely to yield reliable results. We therefore use the indirect test that is described above. 13 clustering in times of earnings announcements. We constructed two portfolios. One portfolio holds the stocks that had negative returns in the week before earnings announcements, or specifically negative cumulative size-adjusted returns between the 6th and 2nd trading days before the announcement. The other portfolio holds the stocks that had positive (non-negative) returns during the same five-day period before the announcement. In each portfolio, stocks are bought at the close of trading day before the earnings announcement date, and sold two trading day later at the close of trading, which means that we use the returns reported by CRSP for the day of the announcement (day 0) and the trading day following it (day 1). In each of the two portfolios, the value-weighted returns are calculated for each calendar day, and monthly returns are then calculated by compounding the daily returns. Finally, in order to control for market-wide effects, the monthly returns of each portfolio are regressed on the three factors of Fama and French (1993), and a momentum factor (Carhart 1997): rt = α + β 1 MRKTt + β 2 SMBt + β 3 HMLt + β 4UMDt + ε t (2) The intercept (α), which we call alpha returns, is an estimate of the monthly portfolio returns that are not explained by the market-wide factors. The sample used for the estimation of (2) is described in Table 1 and section 2, above, and includes 252 monthly returns, from January 1984 through December 2004. The results, which are presented in Table 3, support our loss-aversion hypothesis. For the stocks that had price declines before the announcement returns, the monthly alpha returns are 4.6% with a t-statistic of 4.91; whereas, for the stocks that had a price increase before the announcement, the monthly alpha returns are insignificantly different than zero (α=0.2% with t-statistic of 0.32). 14 Controlling for forecast error To further support the loss-aversion hypothesis, we control for earnings forecast errors and test whether returns are still positive at announcements that come after price declines. We use the following regression: Ret(0,1) it = α + β1 D Neg Re t it + β 2 FEit + β 3 D Neg Re t it × FEit + β 4 MBit + β 5 LogSizeit + ε it (3) where Ret(0,1) represents the size-adjusted returns over a two-day window that includes the earnings announcement date and the subsequent trading day; DNegRet is a dummy variable that equals 1 for firms that experience price declines before the earnings announcement, or, specifically, negative size-adjusted returns for a window, starting 6 trading days before and ending 2 trading days before, the earnings announcement; and FE is actual earnings minus the median of all earnings forecasts issued by analysts 90 days before the announcement. The FE variable is deflated by the stock price on the 6th trading day before the announcement, and winsorized at the top and bottom one percentiles. We expect that the returns at announcements that come after stock price declines will be positive; i.e. that the coefficient of DNegRet (β1) will be positive. The estimation results of (3) are presented in Table 4. The sample used is described above (see section 2). The estimation is performed independently for each calendar quarter in the sample, and the coefficients and significance levels are calculated as in Fama and MacBeth (1973). The R-square reported in the Table is the average of Rsquares of per-quarter regressions. 15 The results support our loss-aversion hypothesis. After controlling for earnings forecast errors, market-to-book, and size, the returns at announcements that come after price declines is 0.5% and statistically significant (t-statistic of 10.52). 4. Loss-aversion discount as a function of expected loss As further support to our loss-aversion hypothesis, we test whether announcement returns are higher when the expected loss is larger. We assume that investors are loss-averse over changes in the value of individual stocks they own. Consistent with this idea, investors seem to use higher discount rates when they believe that the probability of a negative news announcement is higher, as shown above. But if investors dread loss, they are also expected to be concerned about the impact of negative earnings surprise on the value of the firm. In our research setting, the wealth loss potential is the expected price drop at the earnings announcement. In this section, we test whether investors apply larger discounts when the potential price drop on negative earnings news is larger. We use market-to-book as a proxy for expected wealth loss. High market-to-book firms that miss earnings expectations usually experience big price drops at the announcement (e.g., Sloan and Skinner 2002; Barth et al. 2001). Therefore, higher market-to-book implies higher expected loss. Investors are accordingly expected to use higher discount rates when they believe that high market-to-book firms face a higher probability of reporting negative earnings news. We examine whether the price decreases before earnings announcement of firms with higher market-to-book lead to higher announcement returns. As discussed above, price declines before earnings announcements are associated with a higher probability of 16 a negative earnings surprise; i.e., prices decline when investors attain negative information on the coming announcement. Reacting to this information, loss-averse investors can use higher discount rates for stocks of high market-to-book firms, because of the sharp price decline that these stocks experience on negative news. Once earnings are announced, the loss uncertainty is alleviated, and the loss-aversion discount fades. Consequently, we expect that among the firms that had price declines in the week before the announcements, the higher market-to-book firms will experience higher returns at the earnings announcement. To test this hypothesis, we form portfolios, based on the sign of returns before earnings announcement and firms’ market-to-book, as follows. In each calendar quarter we sort stocks, based on their market-to-book, and assign them to three same-sized groups: low, medium, and high. We then construct 6 portfolios, based on whether the returns before announcement were positive or negative, and based on membership in the three market-to-book groups. Returns before earnings announcements are, as above, the size-adjusted returns over five trading days, from the 6th trading day through the 2nd trading day before the announcement. Market-to-book is market value at the 6th trading day before the earnings announcement, divided by the book value of equity in the previous fiscal quarter. Consistent with our hypothesis, we expect that among the three portfolios with negative pre-announcement returns, the high market-to-book portfolio will have the higher announcement returns. Among the three market-to-book portfolios with positive pre-announcement returns, we expect no systematic difference in announcement returns. 17 The accumulation and measurement of returns of the six portfolios is done in the following manner. In each portfolio, stocks are bought at the close of trading the day before the earnings announcement date, and sold two trading days later at the close of trading, which means that we use the returns reported by CRSP for the day of the announcement (day 0) and the trading day that follows it (day 1). The returns of each portfolio are accumulated, using a value-weighted time calendar approach, meaning that the returns for each calendar day are the value-weighted returns of the stocks in the portfolio. Calendar monthly returns are then calculated for each portfolio by compounding the daily returns. Finally, in order to control for market-wide factors, the monthly returns of each portfolio are regressed on the three factors of Fama and French (1993) and a momentum factor (Carhart 1997), as described in equation (2). The results are in Table 5. Panel A presents the estimation of equation (2) over the 252 monthly returns, from January 1984 through December 2004, of each of the 6 portfolios. Panel B provide the descriptive statistics of the six portfolios. Consistent with the hypothesis, the announcement returns for earnings that were preceded by negative returns (negative-returns-before-earnings group) systematically change across the three market-to-book groups. As can be seen in Panel A, the high M/B group has alpha returns of 6.9% per month (t-statistic of 5.56), the medium M/B group has monthly returns of 4.1% (t-statistic of 4.65), and the low M/B has monthly returns of 1.9% which are statistically insignificant (t-statistic of 1.78). A hedge portfolio that buys the stock in the high M/B group, and sells the stocks in the low M/B group, attains monthly returns of 4.7% (t-statistic of 4.25), and therefore indicates that the difference in the returns of the two groups is statistically significant. 18 For earnings announcements that were preceded by positive returns, on the other hand, the announcement returns do not change systematically across the three market-tobook groups. The monthly alpha returns are -0.6% for the low M/B group, -0.5% for the medium M/B group, and 0.2% for the high M/B group, and they are all statistically insignificant. 5. Return reversals on days other than the earnings announcement Our findings can also be described as one-sided return reversal around earnings announcements. When the announcements follow a week of negative returns, the announcement returns will be positive. Short-term return reversals have been documented in prior studies (e.g., Jagadeesh 1990; Lehmann 1990), and are not unique to our earnings announcement setting. In this section, we compare the return reversal around the earnings announcement to the return reversals on days other than the earnings announcement. Starting 30 trading days before the earnings announcement, we perform our main portfolio test, which is presented in section 3, every two trading days, from day -30 to day -4, and from day 2 to day 30; i.e., 29 times in total. For each trading day, the analysis is performed in the same way as in the original test. For trading day t, we constructed two portfolios. One portfolio holds stocks that experienced a week of negative returns, or, specifically, negative size-adjusted cumulative returns for the five-day period starting at day t-6 and ending at t-2. The other portfolio holds the stocks that had positive returns for the same five-day period. In each portfolio, stocks are bought at the close of trading day t-1 and sold at the close of trading 19 day t+1, which means that we use the returns reported by CRSP for the days t and t+1. The returns in each of the two portfolios are accumulated using a time-calendar approach, as in the original test, and the time-series of monthly returns is regressed on the Fama and French (1993) and momentum (Carhart 1997) factors. For generality, we use the entire sample available on CRSP and Compustat from 1971 to 2004, which includes 519,961 quarterly earnings announcements.8 Table 6 compares the alpha returns at the earnings announcement date to the alpha returns on the other trading days. Following negative returns, the returns reversal (alpha returns) at earnings announcements are about twice the size of the reversal on the other days surrounding the announcement. Specifically, the alpha returns at earnings announcements are 4.14%. The median alpha returns on other days is 1.79%, and the maximum is 2.68%. The difference between the returns at the earnings announcements and on other days is statistically significant (t-statistic of 5.16). Following positive returns, return reversal at the earnings announcements are not different than the reversals on other days. The returns for the earnings announcement portfolio is 0.46%, and the median returns for the portfolios of the other days is 0.07%. The difference, however, is not statistically significant (t-statistic of 1.28). In sum, the one-sided return reversal at earnings announcement is about twice the size of the reversal on other days. Our analysis is unique, in that it shows that returns reversals are one-sided and can be a result of loss-aversion. This one-sided reversal, however, can exist on other days that are unrelated to the earnings announcement. After 8 The sample selection requirements are: (a) earnings announcement date is available on the Industrial Quarterly Compustat; and (b) the announcement was made by firms with corresponding stock return data on CRSP for a period starting 6 trading days before the earnings announcement and ending 2 days before the earnings announcement. Note that the results are similar when using the original sample. 20 all, the two ingredients necessary for loss-aversion – uncertainty and a possible loss in stock value – exist on other days. Nonetheless, the reversal is expected to be stronger at earnings announcements. The resolution of uncertainty is usually more complete at earnings announcements, and, consequently, a greater mitigation of loss-aversion and a larger increase in price are expected then. 6. Robustness check Short term return reversals can, in general, be a result of a bid-ask bounce (e.g. Jegadeesh and Titman 1995). In our setting, bid-ask bounce should not be a concern. We build a trading strategy based on the sign of returns, and do not focus on stocks with extreme returns, on which reversal trading strategies are often based and for which bid-ask bounce might be a problem. In any event, in this section we control for the possible effect of the bid-ask bounce on our results. Bid-ask bounce can induce negative serial correlation in returns on adjacent days. Therefore, similar to Jegadeesh (1990) and Lehmann (1990), portfolios are formed a day after the dates when the returns used to form the portfolios are measured. Namely, in our main tests, above, the pre-announcement returns are accumulated up to day -2, and the trading position is taken at day 0. Another standard way to deal with bid-ask bounces is the use of the midpoint between bid and ask spread, instead of the actual price, for the calculation of returns (e.g., Subrahmanyam 2005). We use closing bid and closing ask prices reported by CRSP for NYSE and AMEX stocks from 1993 through 2004.9 Midpoint prices are calculated as the 9 The reliability of the closing bid and closing ask prices reported by NYSE and AMEX is higher after the end of 1992, according to CRSP’s manual. 21 average of the closing bid and closing ask. Returns before the announcement are calculated as the change in closing midpoint prices for the same five-day period as in our main test (from the end of trading day -7 to end of trading day -2), and any dividends and distributions are added. Announcement returns are calculated as the change in closing bid-ask midpoints from day -1 to day 1, plus all distributions. These midpoint-based returns correspond to the sum of CRSP returns for days 0 and 1, which we use as announcement returns in the main test. The sample of firms for which returns can be calculated includes 94,566 earnings announcements. In the sample, 45,371 earnings announcements came after negative preannouncement returns, which are calculated based on midpoint prices. For these announcements, the midpoint-based announcement returns are 0.51%. For the remaining 49,195 earnings announcements that came after positive (non-negative) returns, the midpoint-based announcement returns are 0.13%. For both groups, the midpoint-based announcement returns are similar to the announcement returns calculated based on CRSP data. In sum, the one-sided returns reversal that we observe is not the result of a bid-ask bounce. At earnings announcements that follow price declines, the returns are 0.51%, based on the midpoint of closing bid and ask prices, compared with 0.46% in the main test (see Table 1). 22 7. Transaction costs Predictable stock return patterns, as we document in this paper, naturally raise the possibility of inefficiency in the stock market. In this section we explore this issue by comparing the magnitude of the average return to the transaction costs. Buying stocks that experienced a price decline before earnings announcements would yield a return of approximately 0.5% per announcement; that is, when buying the stock a day before the announcement and selling it a day or two after the announcement (see Table 1). A portfolio that systematically follows this buying strategy would have had average abnormal (alpha) returns of 4.6% per month, as reported in Table 3. The question is whether this leaves room for diversified investors who are less averse to losses to take advantage of this trading opportunity. The most direct estimate of transaction costs is the bid-ask spread plus commission fees paid to brokerage firms to execute the buy and sell orders. We first estimate the return net of bid-ask spread costs, and then discuss the issue of commission fees. The quoted bid-ask spread may overstate the true spread because trades are often executed inside the quoted spread.10 To avoid this problem, an estimate of the effective bid-ask spread is used.11 Our effective spread measure is based on data from TAQ. For each stock, we match the prices during the day with the preceding bid and ask quoted on NYSE, given that it was made less than five minutes before the transaction and quoted between 9:30am and 4pm. The effective spread is then calculated for each pair of bid-ask and price as an absolute difference between price and the bid-ask midpoint, divided by 10 Lee and Ready (1991), for example, provided evidence that many trades are inside the quoted bid-ask spread. 11 Discussion of bid-ask spread and other components of transaction costs is provided, for example, by Keim and Madhavan (1998). 23 bid-ask midpoint. Each stock’s effective spread for the day is the the average of effective spreads during the day. We use the earnings announcements that were made by NYSE firms for which TAQ database has sufficient data to calculate the effective bid-ask spreads. The sample includes 56,619 earnings announcements from 1993 through 2004, of which 28,542 announcements were preceded by a week of negative returns, and 28,077 announcements were preceded by positive pre-announcement returns. The results are presented in Table 7. For stocks that had a price decline in the week before the earnings announcement, a strategy of buying the stock a day before the announcement and selling it two days after the announcement, where the buy and sell are at the opening prices, yields average returns of 0.55% per earnings announcement. However, when trying to execute these transactions, a trader would incur average transaction costs of about 0.5%, due to bid-ask spreads. As seen in the table, the average effective bid-ask spreads are about 0.25% on the days in which our strategy suggests the buy and sell should occur. In addition to the bid-ask spread costs, we need to consider the commission fee, the payment to the broker for the execution of the trade. Keim and Madhavan (1997), for example, found that commission costs overall are about 0.20% of trade value, and Stoll (1995) reported that commissions in 1992 averaged 0.24% of the market value of the trade. In sum, traders probably could not have exploited the predictable positive returns pattern that we report in the previous sections, due to high transaction costs. The return 24 pattern we document only suggests that the average investor who sets prices before earnings announcements is loss-averse. 8. Alternative Explanation When taking a view that stock prices are a product of the interaction between investors who seek liquidity and market makers who provide it (e.g., Grossman and Miller 1988), instead of the dealings of sellers and willing buyers, our results can be interpreted differently. In this case, prices will decrease before earnings announcements when liquidity traders seek to sell their stocks. This is a result of the discount that risk-averse market makers will ask when buying the stock and increasing their position before announcement periods, which are usually more volatile. The discount will be alleviated once earnings are announced, and consequently prices will increase at the earnings announcements, as we observe.12 To test this alternative explanation, in this section we examine the effects of sell pressures on returns reversals around earnings announcements. If market makers are the ones buying stocks to ease the sell pressures, as, for example Campbell, Grossman and Wang (1993) assume, then sell pressures will lead to a higher discount before announcements and, consequently, to the higher return reversal; i.e., higher positive returns at the announcement.13 We define sell pressures as price decreases on high trading volumes, and buy pressures as price increases on high trading volumes. Similar to Campbell, Grossman, 12 For a model predicting that liquidity demand by traders will cause short-term returns reversals, see, for example, Campbell, Grossman, and Wang (1993). 13 Note that inventory models such as the one in Grossman and Miller (1988) predict that buy pressures, and not only sell pressures, will cause short-term return reversals. Details are provided below. 25 and Wang (1993), we assume that market makers will buy or sell the stocks to ease the sell or buy pressures and to satisfy liquidity needs. To measure whether prices decreased/increased on higher than usual trading volume (sell/buy pressures), we run a regression of trading volume in the 365 calendar days prior to the earnings announcement on the absolute returns: Vol t = α + β Abnret t + ε t , (4a) where Vol is trading volume of a firm at day t, divided by the outstanding shares, and |Abnret| is the absolute value of size-adjusted returns. Vol is winsorized at the top and bottom one percentile. The residual, εt, is an estimate of the trading pressure; it indicates that prices decreased or increased on higher than usual trading volumes. We use the sum of the standardized residuals from day -6 through day -2 before the announcement as a measure of the abnormal volume or trading pressures: Volmit = t = −6 ∑ε −1 it σi 5 (4b) Earnings announcements with less than 50 observations available for the estimation of (4a); i.e., less than 50 trading days before the announcement, are excluded. Note that the results of an alternative measure of abnormal trading volume, which are based on the increase in trading volume before the earnings announcement, regardless of price change, are similar to those that are based on (4b).14 14 The alternative abnormal trading volume measure we used is as follows: Volmit = t = −6 ∑ (V −1 it − Vi ) σ i 5 Daily volume, Vit is shares of firm i traded during day t, divided by the shares outstanding of firm i during day t. V and σ are the mean and standard deviation in daily trading volume for firm i from 365 days before the earning announcement to 11 trading days before the announcement. 26 To examine the impact of sell and buy pressures on the announcement returns, we form portfolios based on the sign of returns before earnings announcement and the magnitude of the abnormal trading volume, as follows. In each calendar quarter, we sort stocks based on their abnormal trading volume and assign them to three same-sized groups: low, medium, and high. We then construct 6 portfolios, based on whether the returns before announcement were positive or negative, and based on membership in the three trading volume groups. The higher the trading volume, the higher is the trading pressure. For negative returns before announcement, the high trading volume indicates a sell pressure, and for positive returns before announcement it indicates a buy pressure. Returns before earnings announcements are, as above, the size-adjusted returns over five trading days, from the 6th trading day through the 2nd trading day before the announcement. Abnormal trading volume is defined above. We expect that among the three portfolios with negative pre-announcementreturns, the high trading volume portfolio; i.e., the stocks with higher sell pressures will have the higher announcement returns. After all, if market makers ask for a discount to buy the stock to ease the sell pressure before the announcement, at the announcement the discount will be alleviated and the stock price will increase. The accumulation and measurement of the returns of the six portfolios is done in the following manner. In each portfolio, stocks are bought at the close of the trading day before the earnings announcement date, and sold two trading day later at the close of trading, which means that we use the returns reported by CRSP for the day of the announcement (day 0) and the trading day that follows it (day 1). The returns of each portfolio are accumulated using a value-weighted time calendar approach, meaning that 27 the returns for each calendar day are the value-weighted returns of the stocks in the portfolio. Calendar monthly returns are then calculated for each portfolio by compounding the daily returns. Finally, in order to control for market-wide factors, the monthly returns of each portfolio are regressed on the three factors of Fama and French (1993) and a momentum factor (Carhart 1997), as described in equation (2). The results are in Table 8. Panel A presents the estimation of equation (2) over the 252 monthly returns, from January 1984 through December 2004, of each of the 6 portfolios. Panel B provide the descriptive statistics of the six portfolios. Consistent with our expectations, the announcement returns for earnings that were preceded by price decreases (the negative-returns-before-earnings group) systematically change across the three trading volume groups. As can be seen in Panel A, the high trading volume group has alpha returns of 6.3% per month (t-statistic of 5.94), the medium trading volume group has monthly returns of 4.7% (t-statistic of 4.58), and the low trading volume group has monthly returns of 2.5% (t-statistic of 2.36). A hedge portfolio that buys the stock in the high trading volume group and sells the stocks in the low trading volume group attains monthly returns of 2.9% (t-statistic of 3.10), which indicates that the difference in the returns of the two groups is statistically significant. Across the different levels of buy pressures, the announcement returns are zero. Namely, when earnings announcements are preceded by positive returns, announcement returns do not change across the three trading volume groups. The monthly alpha returns are -0.5% for the low trading volume group, 0.0% for the medium trading volume group, and 0.2% for the high trading volume group, which are all statistically insignificant. 28 Based on the results above, we conclude that higher sell pressures before earnings announcements drive market makers to ask for higher discounts in prices before earnings announcements, a discount that is alleviated once earnings are announced, and leads to the positive announcement returns that we document. Note that inventory models such as the one in Grossman and Miller (1988) predict that buy pressures, and not only sell pressures, will cause short-term return reversals. Our findings on lack of reversals after buy pressures can, for example, indicate that specialists carry excessive inventories of the stock before earnings announcements - so they do not mind selling but are averse to buying more stocks. Alternatively, it is possible that it is easier for investors who act as market makers, and provide liquidity before earnings announcements, to find assets that are comparable to the ones they sell than to diversify the risk associated with buying a stock before its earnings announcement. Finally, the phenomenon could be due to loss-aversion of the liquidity traders. Namely, traders are more eager to sell (than to buy) before earnings announcement and are willing to complete the sell transaction at lower than usual bid prices.15 9. Conclusion The evidence in this study suggests that there is asymmetric reaction to negative and positive information on upcoming earnings announcements, or in specific that negative news looms larger than positive news. In particular, we document a one-sided returns reversal around earnings announcements. Earnings announcements that are preceded by price decreases trigger positive returns reaction. In contrast, those preceded by price 15 We cannot observe the motives of traders who sell the stocks before earnings announcements. One can assume that the sellers have pressing liquidity needs, as inventory models usually assume. But sellers may unload the stock also because they dread the possibility of negative news. 29 increases have zero announcement returns, on average. The price decreases before earnings announcements are shown to be associated with a higher probability of negative earnings news. In reaction to an increased probability of negative news, loss-averse investors seem to assign the stock an additional discount, which is alleviated once earnings are announced and the uncertainty is resolved. The phenomenon we document can be interpreted alternatively as a result of the compensation (discount) that market makers ask for buying stocks and easing sell pressures just before announcement periods. The discount is alleviated once earnings are announced, and prices increase at the earnings announcements, as we observe. Note that even when taking this microstructure approach to pricing, loss-aversion can still play a key role. We cannot observe the motives of traders who sell the stocks before earnings announcements. One can assume that the sellers have pressing liquidity needs, as inventory models usually assume (e.g., Grossman and Miller 1988). But sellers may unload the stock also because they dread the possibility of negative news. The fact that the reversals are one-sided and do not exist after buy pressures casts doubt on whether these inventory models, which predict reversals after both sell and buy pressures, can by themselves explain the phenomenon. Finally, the two ingredients necessary for loss-aversion, uncertainty and a possible loss in stock value, exist also on days unrelated to the earnings announcement. Also, as we show, the one-sided returns reversals happen on days other than the earnings announcement day, although they are smaller in magnitude. By setting our main tests around earnings announcements, we are able to link the drop in price to an increase in the probability of bad news – we do that by comparing analyst consensus forecast and actual 30 earnings – and then to examine whether prices increase when the negative uncertainty is resolved, i.e., at the earnings announcement. We believe that the implications of our results, nonetheless, extend beyond the earnings announcement setting. 31 References Bagnoli, M., Beneish, M., Watts, S., 1999. Whisper forecasts of quarterly earnings per share. Journal of Accounting and Economics 28, 27-50 Ball, R., Kothari, S.P., Wasley, C., 1995. Can we implement research on stock trading rules? Journal of Portfolio Management 21(2), 54-63 Basu. S., 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting and Economics 24, 3-37. Benartzi, S., Thaler, R., 1995. Myopic loss aversion and the equity premium puzzle. Quarterly Journal of Economics 110, 73-92 Barberis, N., M. Huang, 2001. Mental accounting, loss aversion, and individual stock returns. Journal of Finance 56(4), 1247-92 Barberis, N., Shleifer, A., Vishny, R., 1998. A model of investor sentiment. Journal of Financial Economics 49, 307-343 Berk, J., Green, R., Naik, V., 1999. Optimal investment, growth options, and security returns. Journal of Finance 54, 1553-1607 Campbell, J., Grossman, S., Wang, J., 1993. Trading volume and serial correlation in stock returns. The Quarterly Journal of Economics 108, 905-939. Chambers, A., Penman, S., 1984. Timeliness of reporting and the stock price reaction to earnings announcements. Journal of Accounting Research 22, 21–47. Cohen, D., Dey, A., Lys, T., Sunder, S. 2005. Earnings announcement permia and the limits to arbitrage. Working paper Northwestern University. Dechow, P., Sloan, R., 1997. Returns to contrarian investment strategies: Tests of naive expectations hypotheses. Journal of Financial Economics 43, 3-27. Fama, E., French, K., 1992. The cross section of expected stock returns. Journal of Finance 47, 427-465. Fama, E., French, K., 1993. Common risk factors in the return on bonds and stocks. Journal of Financial Economics 33, 3–53. Frazzini, A. Lamont, O., 2006. The earnings announcement premium and trading volume. Working paper Chicago University and Yale School of Management 32 Grossman, S., Miller, M., 1988. Liquidity and Market Structure. Journal of Finance 43, 617-633. Jegadeesh, N., 1990. Evidence of predictable behavior of security returns, Journal of Finance 45, 881–898. Jegadeesh, N. and S. Titman, 1995. Short-horizon return reversals and the bid-ask spread, Journal of Financial Intermediation 4, 116–132. Kanhneman, D., and A. Tversky, 1979, Prospect theory: An analysis of decision under risk, Econometrica 47, 263-291. Kasznik. R., Lev, B., 1995. To warn or not to warn: management disclosures in the face of an earnings surprise. The Accounting Review 70, 113-134. Keim, D., Madhavan, A., 1998. The cost of institutional equity trades. Financial Analyst Journal, 50-69. Keim, D., Madhavan, A., 1997. Transaction costs and investment style: an inter-exchange analysis of institutional equity trades. Journal of Financial Economics 46, 265-292 Kim, O., Verrecchia, R.E., 1991. Trading volume and price reactions to public announcements. Journal of Accounting Research 29, 302–321. Kothari, S.P., 2001. Capital markets research in accounting. Journal of Accounting and Economics 31, 105–231. Lehmann, B., 1990. Fads, martingales, and market efficiency, Quarterly Journal of Economics 105, 1–28. Lee, C., 1992. Earnings news and small traders: an intraday analysis. Journal of Accounting and Economics 15, 265–302. Lee, C., Ready, M., 1991. Inferring trade direction from intraday data. Journal of Finance 46, 733–747. Matsumoto, D., 2002. Management’s incentives to avoid negative earnings surprises. Accounting Review Mitchell, M., Stafford, E., 2000, Managerial decisions and long-term stock price performance. Journal of Business 73, 287-329. 33 Richardson, S., Teoh, S.H., Wysocki, P., 2004. The walk-down to beatable analyst forecasts: the role of equity issuance and insider trading incentives. Contemporary Accounting Research 21, 885-924 Shapira, Z., Venezia, I., 2001. Patterns of Behavior of Professionally Managed and Independent Investors. Journal of Banking and Finance 25 (8), 1573-87. Skinner, D., Sloan, R., 2002. Earnings surprises, growth expectations, and stock returns, or don’t let an earnings torpedo sink your portfolio. Review of Accounting Studies 7, 289–312. Subrahmanyam, A., 2005. Distinguishing between rationales for short-horizon predictability of stock returns. The Financial Review 40, 11—35. Trueman, B., Wong, F., Zhang, XJ, 2003. Anomalous stock returns around internet firms’ earnings announcements. Journal of Accounting and Economics 34, 249-271. Watts, R., 1978, Systematic “abnormal” returns after quarterly earnings announcements, Journal of Financial Economics, 6, 127-150. Watts, R., 2003. Conservatism in accounting part I: explanations and implications. Accounting Horizons 17, 207–221. 34 Table 1 Univariate analysis of association between negative returns before earnings announcements and the probability of negative earnings news Negative Returns Before Earnings (days -6 to -2) 1 83,922 -4.41% 0.46% $2,561 mil. ($384 mil.) 2.82 (1.92) 40.4% Positive Returns Before Earnings (days -6 to -2) 1 84,761 5.16% -0.05% $2,722 mil. ($404 mil.) 3.11 (2.09) 34.7% 13.44** -2.58 * -17.71** 24.10** T-test of Difference Number of Observations Returns before Earnings (days -6 to -2) 1 Earnings Announcement Returns (days 0 & 1) 2 Size 3 mean (median) M/B 3 mean (median) % Negative Earnings Surprises 4 * and ** statistical significance at the 5% and 1% levels, respectively The sample includes 168,683 quarterly earnings announcement from 1984 to 2004. 1 The Negative Return before Earnings group includes the announcements preceded by negative sizeadjusted returns in a five-day period, from the 6th trading day through the 2nd trading day before the earnings announcement. The Positive Returns before Earnings group includes announcements preceded by positive returns in that five-day period. Returns at the earnings announcements are the cumulative raw return at the earnings announcement date (day 0) and the subsequent trading day. Size is the market value of equity 6 trading days before the earnings announcement. M/B is this market value divided by the previous quarter’s book value of equity, and the variable is winsorized at the top and bottom one percentiles. % Negative Earnings Surprises is the proportional number of earnings announcements that were below the median of analyst forecasts issued less than 90 days before the announcements. 2 3 4 35 Table 2 Logit analysis of relationship between negative returns before earnings announcements and negative earnings surprises Pr ob( Miss = 1) = F ( β 0 + β1 D Neg Re t it + β 2 Sizeit + β 3 MBit j k + ∑ δ j Industry _ Dummyit + ∑ λ k QtrYear _ Dummyit + ε it ) (Expected sign) (+) Variables DNegRet (dummy for negative returns before earnings days -6 to -2) LogSize Estimates Coefficient P-value Marginal effect Coefficient P-value Marginal effect Coefficient P-value Marginal effect 0.2174 (<0.0001) 0.0416 -0.1661 (<0.0001) -0.0317 -0.0678 (<0.0001) -0.0129 168,683 63,302 0.057 (-) MB (-) Total number of observations Number of observations with miss = 1 R-Square (McFadden’s LRI) The sample includes 168,683 quarterly earnings announcements from 1984 to 2004. The dependent variable equals 1 when reported earnings were lower than consensus analyst forecast, which is defined as the median of the forecasts issued less than 90 days before the announcement. The marginal effects are based on the mean values of the independent variables. For brevity, the estimates of the industry dummies and quarter-year dummies are not presented. DNetRet is a dummy variable that equals 1 when size-adjusted returns were negative for the five-day period from the 6th trading day through the 2nd trading day before the earnings announcement. LogSize is the log of market value at the 6th trading day before the announcement. MB is the market value at the 6th trading day before the announcement, divided by book value of equity at the previous quarter. The variable is winsorized at the top and bottom one percentiles. Industry_Dummy represents dummy variables that are based on the two-digit SIC industry membership. QtrYear_Dummy represents dummy variables for the calendar quarters in the sample period. 36 Table 3 Testing loss-aversion on announcement returns This table tests whether returns at earnings announcements that are preceded by price declines are positive, a pattern that will support the loss-aversion hypothesis. The portfolio test is performed using the calendartime approach. We constructed two portfolios. One portfolio holds the stocks that had negative cumulative size-adjusted returns for the period between the 6th and 2nd trading days before the announcement. The other portfolio holds the stocks that had positive (non-negative) returns during the same five-day period before the announcement. In each portfolio, stocks are bought at the close of the trading day before the earnings announcement date and sold two trading days later, at the close of trading. Value-weighted calendar monthly returns are calculated by compounding the daily returns, and are regressed on the three factors of Fama and French (1993) and a momentum factor (Carhart 1997). The sample consists of 168,683 quarterly earnings announcements from 1984 to 2004. rt = α + β 1 MRKTt + β 2 SMBt + β 3 HMLt + β 4UMDt + ε t β1 (MRKT) 0.924** (4.01) 1.130** (6.73) β2 (SMB) 0.781** (2.73) 0.170 (0.82) β3 (HML) -0.263 (-0.77) -0.159 (-0.64) β4 (UMD) 0.247 (1.24) -0.051 (-0.35) Returns before earnings (days -6 to -2) 1 Negative Positive * α 0.046** (4.91) 0.002 (0.32) R-Squared 0.158 0.231 and ** statistical significance at the 5% and 1% levels, respectively 1 The Negative Return before Earnings group includes the announcements preceded by negative sizeadjusted returns in a five-day period, from the 6th trading day through the 2nd trading day before the earnings announcement. The Positive Returns before Earnings group includes announcements preceded by positive returns in that five-day period. 37 Table 4 Announcement returns when controlling for forecast error The sample consists of 168,683 quarterly earnings announcements from 1984 to 2004. The equation below is estimated for each calendar quarter, and the reported coefficients and significance levels are calculated as in Fama and MacBeth (1973). The R-square reported is the average of R-squares of per-quarter regressions. Ret(0,1) it = α + β1 D Neg Re t it + β 2 FEit + β 3 D Neg Re t it × FEit + β 4 MBit + β 5 LogSizeit + ε it (Expected Sign) Intercept DNegRet FE DNegRet *FE MB LogSize 0 (+) (+) 1 0.0003 (0.45) 0.0049** (10.17) 0.492** (8.77) (?) (+) (?) -0.010 (-0.29) 2 -0.0008 (-0.59) 0.0050** (10.52) 0.490** (8.87) -0.011 (-0.31) 0.0003* (2.55) -0.000 (0.28) 168,683 1.79% 168,683 2.04% Number of observations R-Square * ** and statistical significance at the 5% and 1% levels, respectively The dependent variable, Ret(0,1), represents the size adjusted returns over a two-day window that includes the earnings announcement date and the subsequent trading day. DNegRet is a dummy variable that equals 1 for firms that experience negative returns before the earnings announcement,or, specifically, negative returns for a window starting 6 days before the earnings announcement and ending 2 days before the earnings announcement. FE is actual earnings, minus the median of all earnings forecasts issued by analysts 90 days before the announcement, where the variable is deflated by the stock price at the 6th trading day before the announcement. FE is winsorized at the top and bottom one percentiles. 38 Table 5 Loss-aversion as a function of potential loss This table tests whether loss-aversion discount is affected by the size of potential loss, using market-tobook as a proxy. Higher market-to-book firms usually experience greater price drops on negative earnings surprises, and therefore they have higher expected loss. In each calendar quarter, we sort stocks based on their market-to-book and assign them to three same-sized groups: low, medium, and high. We then construct 6 portfolios, based on whether the returns before announcement were positive or negative, and based on membership in the three market-to-book groups. Returns before earnings announcements are the size-adjusted returns over five trading days, from the 6th trading day through the 2nd trading day before the announcement. Market-to-book is market value at the 6th trading day before the earnings announcement, divided by the book value of equity in the previous financial quarter. In each of the six portfolios, stocks are bought at the close of the trading day before the earnings announcement date, and sold two trading day later, at the close of trading. Value-weighted calendar monthly returns are calculated by compounding the daily returns, and are regressed on the three factors of Fama and French (1993) and a momentum factor (Carhart 1997). The sample consists of 168,683 quarterly earnings announcements from 1984 to 2004. rt = α + β 1 MRKTt + β 2 SMBt + β 3 HMLt + β 4UMDt + ε t Panel A: Announcement returns portfolios Returns before Earnings (days -6 to -2) 1 Market to-Book Low Medium Negative High HighLow Low Medium Positive High HighLow * α 0.019 (1.78) β1 (MRKT) 0.889** (3.34) β2 (SMB) 1.074** (3.25) β3 (HML) -0.015 (-0.04) β4 (UMD) -0.707** (-3.06) RSquared 0.152 0.132 0.142 0.066 0.170 0.189 0.141 0.055 0.041** (4.65) 0.897** (4.11) 0.522 (1.94) -0.168 (-0.52) -0.096 (-0.51) 0.069** (5.56) 0.758* (2.48) 0.821* (2.17) -0.994 (-2.19) 0.298 (1.13) 0.047** (4.25) 0.092 (0.34) 0.307 (0.91) -0.945* (-2.34) 0.363 (1.54) -0.006 (-0.63) 1.244** (5.55) 0.391 (1.40) 0.197 (0.59) -0.413* (-2.12) -0.005 (-0.56) 1.197** (5.95) 0.312 (1.26) 0.345 (1.16) -0.500** (-2.88) 0.002 (0.25) 0.972** (4.54) 0.036 (0.14) -0.406 (-1.28) 0.073 (0.40) -0.001 (-0.08) 0.300 (1.49) 0.113 (0.45) -0.531 (-1.77) 0.124 (0.71) and ** statistical significance at the 5% and 1% levels, respectively 39 Panel B: Descriptive statistics of portfolios Returns before earnings announcements 1 Market value [$ mil.] 2 Mean (median) 922 (196) 1,988 (437) 5,053 (692) 920 (200) 1,984 (423) 4,990 (682) M/B 2 Mean (median) 1.07 (1.08) 2.06 (1.99) 5.64 (4.19) 1.09 (1.10) 2.07 (2.01) 5.87 (4.30) Returns before Earnings (days -6 to -2) 1 -4.6% -4.1% -4.5% 5.0% 4.7% 5.7% % Negative Earnings Surprises 3 49.0% 40.0% 30.9% 44.4% 34.5% 26.4% M/B Low Negative Medium High Low Positive Medium High 1 The negative returns before earnings group includes the announcements preceded by negative sizeadjusted returns in a five-day period, from the 6th trading day through the 2nd trading day before the earnings announcement. The positive returns before earnings group includes announcements preceded by positive returns in that five-day period. Market value is based on the stock price 6 trading days before the earnings announcement, and marketto-book is market value, divided by the book value of equity in the previous financial quarter. % Negative Earnings Surprises is the proportional number of earnings announcements that were below the median of analyst forecasts issued less than 90 days before the announcements. 2 3 40 Table 6 Comparing the returns reversal at the earnings announcement to the returns reversal on other days This table compares the return reversal around the earnings announcement to the return reversals on other days around the earnings announcement. The return reversal is examined every two trading days from day 30 to day -4, and from day 2 to day 30; i.e., 29 times in total (where day 0 is the earnings announcement date). For each trading day t, two portfolios are constructed. One holds stocks that experienced a week of negative returns, or, specifically, negative size-adjusted cumulative returns for the five day period starting at day t-6 and ending at t-2. The other portfolio holds stocks that had positive returns for the same five day period. In each portfolio, stocks are bought at the close of trading day t-1 and sold at the close of trading day t+1. The returns in each of the two portfolios are accumulated using a time-calendar approach, and the time-series of value-weighted monthly returns is regressed on the Fama and French (1993) and momentum factors (Carhart 1997). For generality, we use the entire sample available on CRSP and Compustat from 1971 through 2004, which includes 519,961 quarterly earnings announcements. For brevity, we report only the intercept of the regressions (alpha returns). rt = α + β 1 MRKTt + β 2 SMBt + β 3 HMLt + β 4UMDt + ε t Alpha returns (intercept of regression) Trading days preceded by a week (t-6 to t-2) of Negative Returns Returns reversal on earnings announcement (day 0) 4.14% 0.46% Trading days preceded by a week (t-6 to t-2) of Positive Returns Returns reversal on other days around the earnings announcement (days -30 to 30) Minimum 25th percentile Median 75th percentile Maximum 1.07% 1.51% 1.79% 2.16% 2.68% -0.64% -0.30% 0.07% 0.24% 0.66% Difference between return reversal on earnings announcement and on other days T-test ** 5.16** 1.28 statistical significance at the 1% level. 41 Table 7 Controlling for transaction costs This table presents estimates of the transaction costs that a trader would incur if trying to exploit the onesided return reversal at earnings announcements. TAQ data is used, and the sample includes 56,619 earnings announcements made by NYSE firms between 1993 and 2004. Number of observations Returns before earnings 1 Market value 2 Market-to-book 2 Negative Returns Before Earnings 1 (days -6 to -2) 28,542 mean (median) mean (median) mean (median) mean (median) mean (median) mean (median) mean (median) mean -2.90% (-2.08%) 4,128 (931) 2.63 (1.80) 0.25% (0.16%) 0.24% (0.16%) 0.55% (0.30%) 0.58% (0.34%) 0.10% Positive Returns Before Earnings 1 (days -6 to -2) 28,077 3.61% (2.62%) 4,250 (951) 2.81 (1.90) 0.24% (0.16%) 0.24% (0.15%) 0.29% (0.00%) 0.26% (0.04%) 0.13% T-test of Difference -1.06 -6.47** 1.85 2.36* 5.22** 6.44** -2.27* Effective bid-ask spread one day 3 before the announcement (day -1) Effective bid-ask spread two days 3 after the announcement (day 2) Returns for buying at the open of day -1 and selling at open of day 2 4 Announcement returns (CRSP) 5 Size-decile returns 6 * 1 and ** statistical significance at the 5% and 1% levels, respectively The Negative-Return-before-Earnings group includes the announcements preceded by negative sizeadjusted returns in a five-day period, from the 6th trading day through 2nd trading day before the earnings announcement. The Positive-Returns-before-Earnings group includes announcements preceded by positive or zero returns in that five-day period. Market value is based on the stock price 6 trading days before the earnings announcement, and marketto-book is market value, divided by the book value of equity in the previous financial quarter. Effective bid-ask spread is the daily average of the absolute price deviation from midpoint of the bid and ask spread, scaled by the bid-ask midpoint. Announcement returns (open-to-open) are the returns from open of trade at the trading day before the earnings announcement (day -1) until the opening price two trading days after the announcement (day 2). Announcement returns (CRPS) are the cumulative raw returns at the earnings announcement date (day 0) and the subsequent trading day (day 1) that are reported by CRSP. Size-decile returns are returns attained by firms in the same size-decile as the reporting firm at the earnings announcement date (day 0) and the subsequent trading day (day 1), as provided by CRSP. 2 3 4 5 6 42 Table 8 Discount before earnings announcements as a function of sell pressures This table tests whether higher sell pressures before earnings announcements lead to more positive announcement returns, a pattern that is consistent with market makers demanding a discount to buy stocks and ease sell pressures before earnings announcements. We define sell pressures as decreases in prices (negative returns) on high trading volume. In each calendar quarter, we sort stocks based on their abnormal trading volume and assign them to three same-sized groups: low, medium, and high. We then construct 6 portfolios, based on whether the returns before announcement were positive or negative, and based on membership in the three trading volume groups. In each of the six portfolios, stocks are bought at the close of the trading day before the earnings announcement date, and sold two trading day later at the close of trading. Value-weighted calendar monthly returns are calculated by compounding the daily returns, and are regressed on the three factors of Fama and French (1993) and a momentum factor (Carhart 1997). The sample consists of 168,683 quarterly earnings announcements from 1984 to 2004. rt = α + β 1 MRKTt + β 2 SMBt + β 3 HMLt + β 4UMDt + ε t Panel A: Announcement returns portfolios Returns before Earnings (days -6 to -2) 1 Abnormal Trading Volume Low Medium Negative High High-Low Low Medium Positive High High-Low * α 0.025* (2.36) β1 (MRKT) 0.634* (2.42) β2 (SMB) 0.610 (1.88) β3 (HML) -0.186 (-0.48) β4 (UMD) 0.158 (0.69) RSquared 0.069 0.181 0.070 0.005 0.131 0.129 0.197 0.023 0.047** (4.58) 1.056** (4.12) 0.931** (2.94) -0.320 (-0.84) -0.462* (-2.08) 0.063** (5.94) 0.797** (3.05) 0.177 (0.55) -0.274 (-0.71) 0.048 (0.21) 0.029** (3.10) -0.072 (-0.31) -0.169 (-0.58) -0.102 (-0.29) -0.189 (-0.94) -0.005 (-0.53) 1.103** (5.16) 0.367 (1.39) 0.408 (1.29) -0.265 (-1.44) 0.000 (0.06) 0.827** (4.20) 0.099 (0.40) -0.354 (-1.20) -0.082 (-0.48) 0.002 (0.32) 0.864** (4.71) 0.385 (1.70) -0.440 (-1.62) -0.232 (-1.46) -0.001 (-0.20) 0.257 (1.52) 0.253 (1.21) 0.126 (0.50) -0.152 (-1.04) and ** statistical significance at the 5% and 1% levels, respectively 43 Panel B: Descriptive statistics of portfolios Market value [$ mil.] 3 Mean (median) 2,324 (353) 2,135 (326) 3,305 (505) 2,778 (384) 2,451 (363) 2,914 (465) % Negative Earnings Surprises 4 41.0% 40.4% 39.7% 36.0% 35.0% 33.3% Returns before earnings announcements 1 Abnormal Trading Volume 2 Low M/B 3 Mean (median) 2.70 (1.87) 2.75 (1.88) 3.03 (2.04) 2.95 (2.01) 3.01 (2.02) 3.36 (2.23) Returns before Earnings (days -6 to -2) 1 -4.14% -4.05% -5.12% 4.72% 4.53% 6.12% Negative Medium High Low Positive Medium High 1 The negative-return-before-earnings group includes the announcements preceded by negative sizeadjusted returns in a five-day period, from the 6th trading day through the 2nd trading day before the earnings announcement. The positive-returns-before-earnings group includes announcements preceded by positive returns in that five-day period. Abnormal trading volume measures the increase/decrease in trading volume in a five-day period, from the 6th trading day through the 2nd trading day before the earnings announcement, relative to average trading volume in the preceding year. The details are described in Section 8. Market value is based on the stock price 6 trading days before the earnings announcement, and marketto-book is market value, divided by the book value of equity in the previous financial quarter. % Negative Earnings Surprises is the proportional number of earnings announcements that were below the median of analyst forecasts issued less than 90 days before the announcements. 2 3 4 44