The effect of the summer doldrums on earnings announcement returns and ERCs

... price and/ or volume reaction is affected by the timing of the earnings announcement, such as the day of the week of the announcement (Dellavigna and Pollet 2009) or the time of day of the announcement. .. on the relative summer absence of noise traders and its effect on earnings announcement price reactions and earnings response coefficients (ERC’s), or the market price reaction to a unit of earnings. .. primarily on the characteristics of the audience of investors for the announcement The differential beliefs and behavior of these investors can help dictate the nature of the market response to the announcement

THE FLORIDA STATE UNIVERSITY COLLEGE OF BUSINESS

THE EFFECT OF THE SUMMER DOLDRUMS ON EARNINGS ANNOUNCEMENT RETURNS AND ERC’S

By GREGORY B GAYNOR

A Dissertation submitted to the Department of Accounting

in partial fulfillment of the requirements for the degree of Doctor of Philosophy

Degree Awarded:

Fall Semester, 2011

All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent on the quality of the copy submitted.

In the unlikely event that the author did not send a complete manuscript

and there are missing pages, these will be noted Also, if material had to be removed,

a note will indicate the deletion.

All rights reserved This edition of the work is protected against

unauthorized copying under Title 17, United States Code.

ProQuest LLC.

789 East Eisenhower Parkway

P.O Box 1346 Ann Arbor, MI 48106 - 1346

UMI 3502846 Copyright 2012 by ProQuest LLC.

UMI Number: 3502846

Gregory B Gaynor defended this dissertation on September 15, 2011

The members of the supervisory committee were:

Richard Morton Professor Directing Dissertation

Thomas Zuehlke University Representative

Bruce Billings Committee Member

Tim Zhang Committee Member

The Graduate School has verified and approved the above-named committee

members, and certifies that the dissertation has been approved in accordance with university requirements.

ACKNOWLEDGEMENTS

I would like to acknowledge Rick Morton (chair) for his extensive help throughout this process, as well as my committee members for their helpful comments and guidance All errors are my own

TABLE OF CONTENTS LIST OF TABLES Error! Bookmark not defined.

ABSTRACT vi

1 INTRODUCTION 1

2 BACKGROUND 6

2.1 Noise vs Sophisticated Traders 6

2.2 Investor Inattention and Delayed Price Response 9

3 HYPOTHESIS DEVELOPMENT 12

3.1 Noise Traders’ Effect on Returns 12

3.2 Noise Traders’ Effect on the ERC 12

3.3 Pre-Announcement Period Returns 14

3.4 The Effect of the Online Period 14

3.5 Post-Earnings Announcement Drift 15

3.6 Trading Volume 17

3.7 Investor Interest 17

4 RESEARCH DESIGN 19

4.1 Sample 19

4.2 Measuring Return 20

4.3 Measuring Earnings Surprise 21

4.4 Measuring Trading Volume 21

4.5 Models 21

4.5.1 Testing of H1-H3 21

4.5.2 Testing of H4-H6 23

4.5.3 Testing of H7 24

4.5.4 Testing of H8 25

4.5.5 Test of H9 & H10 25

5 RESULTS 28

5.1 Descriptive Statistics 28

5.2 Analysis 28

5.2.1 Test of H1 28

5.2.2 Test of H2 29

5.2.3 Test of H3 30

5.2.4 Test of H4 - H6 31

5.2.5 Test of H7 33

5.2.6 Test of H8 35

5.2.7 Test of H9-H10 35

6 CONCLUSION 61

REFERENCES 64

BIOGRAPHICAL SKETCH 67

LIST OF TABLES

Table 1: Sample and Descriptive Statistics 37 Table 2: Descriptive Statistics of Summer vs Non-Summer and Test of

Differences in Mean 38 Table 3: Test of H2: Regression of Announcement-Period CAR 41 Table 4: Test of H3: Regression of Pre-Announcement-Period {-10,-1} CAR 43 Table 5: Test of H4: The Effect of the Online Period on Announcement-Period

{0, 2} CAR 45 Table 6: Test of H5: The Effect of the Online Period on ERC’s 47 Table 7: Test of H6: The Effect of the Online Period on Pre-Announcement-

Period {-10,-1} CAR 48 Table 8: Test of H7: Regressions of Post-Announcement-Period CAR 49 Table 9: Test of H8: Regression of Summer/Non-Summer Differences in

Announcement-Period CAR 56 Table 10: Test of H9 and H10 57

ABSTRACT

Conventional wisdom, as well as recent research (Hong and Yu 2009), suggest that trading activity and returns decrease during the summer months, possibly due to decreased market participation by net-buying noise traders

I extend previous research by specifically testing for differences in returns

in the period surrounding both summer and non-summer earnings announcements

I document lower abnormal returns surrounding summer earnings announcements compared to non-summer announcements My results suggest that this difference in abnormal returns is greater in the online-trading period an era characterized by increased noise trading However, I do not find this difference between summer and non-summer announcement-period returns to be related to a firm’s analyst following, market-to-book ratio, or the summer

vs non-summer difference in a firm’s announcement-period trading volume In addition, I do not find evidence that the summer vs non-summer difference in announcement-period returns is affected by the level of unexpected earnings revealed in the earnings announcement

CHAPTER 1 INTRODUCTION

It is a widely-held belief that trading activity decreases during the summer months, spawning the term ―summer doldrums‖ to describe this time period.1 Hong and Yu (2009) confirm the existence of significantly lower trading volume and returns on U.S exchanges during the summer (measured as the months of the 3rd calendar quarter (July through September)) They find that trading volume decreases by 8.9% and monthly returns decrease by 1% during the summer They also document lower summer returns for other countries with summer decreases in trading volume They attribute their findings of lower summer returns and trading volume to the relative inattention/absence of both institutional investors and noise traders

Much accounting literature has examined the market reaction to earnings announcements (Ball & Brown 1968, Beaver 1968, among many others) Some research has focused on how this price and/ or volume reaction is affected by the timing of the earnings announcement, such as the day of the week of the announcement (Dellavigna and Pollet 2009) or the time of day of the announcement (Doyle and Magilke 2009) However, to my knowledge, my study represents the first to examine how the market reaction to earnings announcements is affected by the pervasive and predictable summer slowdown in trading activity I extend the work of Hong and Yu (2009) by examining the effect that this summer decrease in investor attention may have on the market reaction to U.S earnings announcements made during the summer My study focuses primarily on the relative summer absence of noise traders and its effect on earnings announcement price reactions and earnings response coefficients (ERC’s), or the market price reaction to a unit of earnings surprise

Though noise traders are thought to be unsophisticated, a considerable amount of research suggests that their trading can affect stock returns Barber and Odean (2008) find that noise traders tend to trade in stocks that

1 Abundant references to the typical summer slowdown in trading activity can be found

in the popular literature using a key word search Examples include:

away/

http://blogs.wsj.com/financial-adviser/2010/05/03/the-truth-about-sell-in-may-and-go-http://seekingalpha.com/article/221113-four-tech-titans-with-cash-to-spend

http://www.marketwatch.com/story/sp-500-is-on-uncertain-footing-2010-06-01

http://www.marketwatch.com/story/us-stocks-slump-on-european-bank-worries-2010-09-07 http://beginnersinvest.about.com/od/beginnerscorner/qt/summerdoldrums.htm

catch their attention The personal preferences of noise traders can dictate which attention-grabbing stocks they will buy The authors suggest that contrarian investors may choose to buy out-of favor stocks that catch their eye, while momentum investors may chase recently high-performing (glamour) stocks Due to noise traders’ aversion/ inability to sell short, they act as net-buyers of these attention-grabbing stocks Consistent with this belief, Lee (1992) finds that noise traders are net buyers subsequent to both positive and negative earnings surprises This can have the effect of magnifying reactions to positive surprises and tempering reactions to negative surprises Frazzini and Lamont (2010) show that stock prices rise around earnings announcements and suggest that this earnings announcement premium is driven by small (noise) investor buying when the announcement catches their attention Huo, Peng, and Xiong (2009) suggest that individual investor attention can both increase price overreactions in up markets as well as attenuate underreactions to events such as earnings reports Lamont and Thaler (2003) suggest that the mispricing caused by noise traders may not

be fully corrected by arbitragers Other research argues that increased online trading has lead to greater noise trader participation and greater ERC’s (Ahmed, Schneible, and Stevens 2003)

Given this body of research suggesting that the presence of noise traders can affect announcement-period price reactions, it stands to reason that differences in attention levels among these investors between summer and non-summer earnings announcements may produce differences in price reactions around these announcements as well Specifically, newsworthy events, such as earnings surprises, may catch the attention of fewer net-buying noise traders during the summer Therefore, I hypothesize and find that there is a less positive price reaction around the time of a summer earnings announcement relative to a non-summer announcement These results hold for both my full sample as well as sub-samples of positive and negative earnings surprises

I also consider how noise trader inattention and an absence of net buying might affect the market reaction to the earnings news itself Research suggests there is significant noise-trader buying following both positive and negative earnings surprises (Lee 1992) that catch the attention

of noise traders (Barber and Odean 2008) It is reasonable to suggest that the magnitude of an earnings surprise is directly related to the ability of

an earnings announcement to catch investors’ attention Therefore, the relative inattention of net-buying noise traders during the summer may result

in a smaller positive price reaction to a unit of positive earnings surprise,

especially for large positive surprises However, this summer decrease in net-buying noise trading may result in a larger downward price reaction to a unit of negative earnings surprise, especially for large negative surprises Therefore, I hypothesize that summer ERCs will be less positive following a positive earnings surprise, but more positive following a negative earnings surprise However, my results do not support this view since I do not find evidence of a differential summer vs non-summer reaction to unexpected earnings This is consistent with the view that a noise-trader’s decision to trade following an earnings announcement is based primarily on the event itself, as opposed to the level of unexpected earnings

Based upon the predicted and/or actual behavior of noise traders, other types of investors may help create differences between summer and non-summer earnings announcement price reaction and ERC’s Frazzini and Lamont (2010) provide evidence of a general increase in institutional-investor buying just before earnings announcements However, this buying is followed by institutional selling beginning just after the announcement, as noise-trader buying emerges Therefore, Frazzini and Lamont (2010) conclude that institutional investors front-run noise traders Noise-trader-induced price appreciation immediately following the announcement can make institutional-investor pre-announcement buying profitable even if the institutional investors choose not to sell the shares immediately following the announcement In such situations, these institutional investors act as net-buyers during the combined pre-announcement and announcement periods Because they may anticipate it to be less profitable, institutional investors may decrease the amount of their pre-announcement buying in the summer since there are fewer net-buying noise traders to boost the stock price following the announcement I find evidence consistent with this view as I document lower pre-announcement-period abnormal returns in the summer period compared

to those of the non-summer period I conclude that this difference has increased in the online period, perhaps, due to the increase in online (noise) trading Therefore, the behavior of other investors may contribute

to the effects under examination based upon the perceived and/or actual behavior of noise traders

In addition to examining price reaction within the shorter-term announcement-period event window, I test for possible differences in the longer-term price reaction to summer announcements compared to that of non-summer announcements Research suggests that in some cases decreased investor attention at the time of an earnings announcement may cause a

reduced immediate price reaction followed by increased price drift (Dellavigna and Pollet 2009) Consistent with this notion, I investigate whether decreased investor attention during the summer also causes differences in the delayed price reaction to summer vs non-summer earnings announcements I find a direct relationship between announcement-period

cumulative abnormal returns (CAR’s) and longer-term, post-announcement CAR’s

In addition, my results suggest that, on average, mean abnormal returns over the longer post-announcement period are lower during the summer than during the non-summer period, even after controlling for announcement-period returns

or unexpected earnings I find some evidence of a summer vs non-summer difference in post-announcement price drift Taken together, these results support the belief that post-announcement returns are also affected by the summer absence of net-buying noise traders

Even though the summer slowdown may produce differential effects regardless of whether a firm announces good or bad news, it is reasonable to suggest that not all firms will be affected equally Specifically, the stocks that experience the largest noise trader participation in the non-summer period may be associated with the largest change in price reaction and ERC’s during the summer when those net-buying noise traders tend to be absent All else equal, a decrease in noise trader participation translates

to an overall decrease in net-buying trading volume and, consequently, an overall decrease in returns Therefore, I hypothesize that the difference in announcement-period abnormal returns is directly related to the difference in summer vs non-summer announcement-period trading volume; however, my results

do not support this prediction Barber and Odean (2008) suggest that noise traders are more likely to trade in stocks that catch their attention I use analyst following and market-to-book ratio (MTB) to proxy for investor

interest I control for unexpected earnings (UE) because it is reasonable to

suggest that the magnitude of an earnings surprise is directly related to the ability of an earnings announcement to catch investors’ attention However, after controlling for unexpected earnings, I do not find a significant relationship between the summer vs non-summer difference in a firm’s announcement-period abnormal returns and either its MTB or analyst following Similarly, I do not find a significant relationship between either MTB or analyst following and the summer vs non-summer difference in announcement-period trading volume It may be the case that overall trading volume is driven, in large part, by forces other than those under consideration (e.g high-frequency trading)

My study contributes to the literature by examining how stock prices impound the information contained in accounting disclosures based upon both the size and composition of the audience for the disclosure It stands to reason that inattention on the part of sophisticated investors who trade according to estimates of fundamental value may cause the price discovery period to be prolonged However, the inattention of net-buying noise traders, as appears to be the case in the summer, may actually alleviate upward pressure on prices While previous research (Dellavigna and Pollet

2009, Patell and Wolfson 1982, among others) addresses the relationship between firm disclosures and investor inattention in other specific situations, little attention has been paid to the differences in trading environment between the summer and non-summer period Prior research has addressed noise-trader behavior (Barber and Odean 2008; Lee 1992; Ahmed et

al 2003, among others) as well as the effects of the typical summer slowdown

in trading activity (Hong and Yu 2009) However, to my knowledge, mine is the first study to examine the effects that the summer inattention of noise traders has on earnings announcement returns and trading volume Because the summer slowdown is a recurring, predictable phenomenon affecting the vast majority of publicly-traded firms, my results may help investors make better trading decisions, especially during the summer In addition, my findings may encourage accounting researchers to control for summer effects to better test for other earnings announcement phenomena

The rest of this paper is organized as follows The next chapter provides a discussion of background literature Chapter 3 contains the theoretical background for the hypotheses that I test Chapter 4 provides a discussion of my research design In Chapter 5, I analyze my results Chapter 6 concludes

CHAPTER 2 BACKGROUND 2.1 Noise vs Sophisticated Traders

Due to their information content, earnings announcements cause a significant market reaction (Beaver 1968, Ball & Brown 1968) Accounting research has extensively studied this reaction in an effort to determine the factors affecting ERC’s (Collins and Kothari 1989; Easton and Zmijewski 1989), stock price (Atiase 1985), and trading volume (Bamber et al 1995) following the announcement While much of the literature focuses on identifying firm characteristics that help determine the nature of the market reaction, my study focuses primarily on the characteristics of the audience

of investors for the announcement The differential beliefs and behavior of these investors can help dictate the nature of the market response to the announcement

According to Beaver (1968), the change in price reflects the average change in investors’ beliefs whereas trading volume reflects the sum of the differences in investors’ reactions to the earnings announcement Several trading models involving price change and volume have since been constructed (Kim and Verrecchia 1991, 1994, Abarbanell et al 1995, Kandel and Pearson

1995, among others) An important drawback of many early models of trade in speculative markets is both the assumption that agents interpret public information identically as well as the simplifying prediction that price change is directly related to trading volume With such assumptions, some models are unable to explain the often-observed situation of heavy trading volume accompanied by little or no price change or vice-versa (Bamber and Cheon 1995) Kandel and Pearson (1995) improve upon much of the prior literature by modeling the scenario in which agents are heterogeneous in their interpretation of news as well as their prior beliefs Their model allows for the presence of ―noise‖ traders whose nạve trading can significantly alter the price move that would otherwise be caused by the trading of informed investors (Kyle (1985), Hasbrouck (1991)) The more that these noise traders take the opposite side of a trade with an informed investor, the less an informed investor’s trades will change the price of a stock and reveal private information Consequently, noise trading can both distort the assumed positive relationship between trading volume and price

change as well as cause price movement away from fundamental value I suggest that noise-trader behavior helps drive the effects under consideration in my study

Though the noise trading of small investors is thought to be uninformed, evidence suggests it could introduce an upward price bias By analyzing intraday transaction data, Lee (1992) finds that noise traders (those placing orders of less than $10,000) are net buyers subsequent to both positive and negative earnings surprises Consistent with these results, Barber and Odean (2008) suggest that individual investors are net buyers of attention-grabbing stocks following events such as earnings announcements

By analyzing stock transactions in individual brokerage accounts, the authors find that individual investors are net buyers on high volume days, days when stocks are in the news, and days following both extremely negative and extremely positive one-day returns Huddart, Lang, & Yetman (2009) document increased noise-trader buying when examining the increased trading volume that occurs when stock prices cross either the upper or lower limit of a key trading range—another example of an attention-grabbing event Barber and Odean (2008) cite the time and resource constraints that noise traders face

in selecting stocks to buy These individuals cannot thoroughly screen the thousands of possible selections; thus, they will more likely purchase a stock that has grabbed their attention An earnings announcement itself can prompt noise-trader buying, regardless of the earnings news Stocks that miss an earnings forecast may be favored by bargain-hunting noise traders who take a contrarian view to that of the market Stocks that meet or beat a forecast may attract buying from momentum investors chasing high-performing (glamour) stocks

Noise traders do not face the same daunting selection task when selling stock since they will most likely sell one of the stocks they already own instead of selling short Lamont and Thaler (2003) suggest that noise traders are especially burdened by short-sale constraints through the mechanical impediments administered by regulators This aversion or inability to sell short tends to cause noise traders to be net-buyers who exert upward price pressure on stocks following both positive and negative earnings surprises This net-buyer effect of noise traders is consistent with Miller (1977), who suggests the holders of a stock will tend to be those who are most optimistic about its prospects and that, given institutional (or self-imposed) constraints on short-selling, any increase in the set of potential owners (potential buyers) should result in a price increase

Lamont and Thaler (2003) suggest that in the presence of short-sale constraints and/or liquidity risk, stock prices can be mispriced because arbitragers are unable or unwilling to correct them They discuss examples

of such clear mispricing, including the well-documented spin-off of Palm by its parent company 3Com in 2000 In this case, despite the fact that there was a simple, relatively risk-free arbitrage opportunity, the shares of the two companies remained wildly mispriced for months because of the short-term constraints on selling short

Frazzini and Lamont (2010) provide additional evidence that noise trading can introduce an upward price bias They show that stock prices rise around earnings announcements and that this ―quantitatively substantial‖ earnings announcement premium appears consistently since 1927 Using intraday transaction data to address which set(s) of investors are responsible for the premium, the authors use small trades (less than $5,000)

as a proxy for individual investors and big trades (over $50,000) as a proxy for institutional investors They conclude that the earnings announcement premium is driven by noise-trader buying when the announcement catches their attention They find evidence that large investors are aware of the premium and trade in anticipation of it They document abnormal net-buying by large investors in the two-week period preceding the earnings announcement This large-investor buying activity reverses on announcement day and on the two trading days subsequent to the announcement, when noise-trader buying is most intense Large investors appear to be front-running noise traders and diminishing what would otherwise be an even-larger earnings announcement premium However, the presence of abnormal returns immediately following announcements indicates that large investors have not eliminated the premium The premium may continue to exist because of market frictions such as transaction or holding costs (Lamont and Thaler 2003)

Evidence suggests that, in addition to affecting returns, noise traders can also affect ERC’s Ahmed et al (2003) address the assertion that the advent of online trading has increased the ratio of nạve (noise) traders to sophisticated traders Because sophisticated investors are thought to have more precise information than do noise traders, the online period is thought

to be associated with a decrease in the average precision of investor information prior to earnings announcements Since investors with less precise prior information will rely more on the earnings announcement information, this decreased average precision of prior information in the online period would translate to larger revisions in investor beliefs post-

announcement and, hence, larger ERC’s Consistent with these views, Ahmed et

al (2003) cite increased noise trading for their finding of larger ERC’s in the online period (1996-99) than in the pre-online period (1992-95) The authors combine positive and negative earnings surprises in their analysis, implicitly assuming a symmetric relation However, it is possible that noise-trader effects on the ERC may depend upon the sign of the earnings surprise

Prior research finds that noise traders are net-buyers following both positive and negative earnings surprises (Lee 1992) in stocks that catch their attention (Barber and Odean 2008) It stands to reason that the likelihood that a stock catches a noise trader’s attention is directly related to the magnitude of the earnings surprise for both positive and negative surprises This would suggest there may be significant noise-trader buying following both large positive and negative earnings surprises Stocks that miss an earnings forecast can attract bargain-hunting, contrarian noise traders Stocks that meet or beat a forecast may induce buying from momentum investors chasing high-performing (glamour) stocks

Evidence indicates not only that noise traders are net-buyers around earnings announcements, but also that this behavior can have a significant effect on returns and/or ERC’s because it is not fully counteracted by sophisticated investors It follows, then, that returns and/or ERC’s are likely to vary according to the attention levels of noise traders The next section provides a discussion of evidence supporting this view

2.2 Investor Inattention and Delayed Price Response

Research suggests that the timing of an earnings announcement can affect the size and/or composition of the investor audience for the announcement In turn, this can affect the price and/or volume reaction to the announcement Doyle and Magilke (2009) examine the difference in trading volume reaction for announcements made after-the-market close (AMC) and before-the-market-open (BMO) It has been suggested that AMC announcements attract more investor attention because of the larger amount of time, compared to that of BMO announcements, that elapses between the announcement and the resumption of trading in the stock the following morning They find evidence consistent with increased noise-trader participation for AMC announcements as they document larger abnormal trading volume following an AMC announcement Dellavigna and Pollet (2009) suggest that investor-

attention levels for earnings announcements also vary according to the day of the week the announcement is made Consistent with the traditional view that there is less investor attention on Fridays compared with other weekdays, they document less immediate price response and trading volume following a Friday earnings announcement

If low investor attention levels at the time of an earnings announcement result in a lower immediate price response to the announcement,

it is reasonable to believe that the stock price may continue to ―drift‖ as investors revisit the information contained in the announcement and correct the initial mispricing Indeed, this reasoning has been used to explain the phenomenon of post-earnings announcement drift (PEAD) PEAD describes the well-documented tendency of post-announcement stock prices to continue to move in the direction of an earnings surprise (Ball and Brown 1968, Bernard and Thomas 1989).2 While Dellavigna and Pollet (2009) find less of an immediate price response to Friday earnings announcements, they find that Friday earnings announcements are associated with more price drift in the post-announcement period (up to 75 trading days after the announcement) They suggest that PEAD represents a delay in the price discovery process caused by investor inattention at the time of the announcement followed by a price drift as investors continue to process the earnings information well after the announcement This explanation is supported by the evidence of Huo, Peng, and Xiong (2009), who suggest that increased noise-trader attention can mitigate the drift associated with initial earnings announcement underreaction

I build upon the existing literature on investor inattention and PEAD

in my study of summer earnings announcements In addition, I extend research indicating that investor attention is lower during the summer period Hong and Yu (2009) find that, for the period 1962-2005, monthly share turnover (trading volume divided by shares outstanding) is 8.9% lower during the summer than during the rest of the year They use intraday transaction data

to determine which set(s) of investors are responsible for the decrease in summer trading activity Consistent with the technique of Frazzini and Lamont (2010), the authors use small trades (less than $5,000) as a proxy for individual investors and big trades (over $50,000) as a proxy for institutional investors They find a summer decrease in trading activity for

2 Instead of attributing PEAD to investor inattention at the time of the announcement, Bernard and Thomas (1989) suggest that it is caused primarily by an apparent inability

of the market to understand the implications of current quarterly earnings for future earnings

both sets of investors In addition, Hong and Yu (2009) document a decrease

in summer returns for a total of 51 countries which also experience a significant decrease in summer trading volume The authors conclude that the decrease in summer returns is related to the decrease in summer trading volume and that both are caused by inattention on the part of investors, including noise traders

Research indicates that inattention on the part of net-buying noise traders can affect returns and trading volume following earnings announcements In addition, existing evidence suggests that there is less investor attention during the summer However, mine is the first study, that

I know of, to examine if and how this general summer slowdown affects both the short and long-term reaction to summer earnings announcements Because the summer slowdown is both pervasive and predictable, insights from my study should be of use to researchers and capital market participants both inside and outside the firm

CHAPTER 3 HYPOTHESIS DEVELOPMENT 3.1 Noise Traders’ Effect on Returns

Because of their inability or unwillingness to sell short, noise traders are normally net-buyers of stocks Due to time and resource constraints, they tend to buy stocks that catch their attention through a newsworthy event such as an earnings announcement (Barber and Odean 2008) Evidence suggests they are net-buyers following both positive (―meet-or-beat‖ forecasts) and negative (―miss‖ forecasts) earnings surprises (Lee 1992) Sophisticated traders may rationally anticipate noise-trader behavior and nullify its effects But if this noise-trader buying at the time of the announcement is not fully counteracted by the trading of sophisticated investors, the upward pressure leads to an earnings announcement premium (Frazzini and Lamont 2010) Thus, it stands to reason that inattention on the part of noise traders will have a negative effect, on average, on stock returns and trading volume following an earnings announcement (Dellavigna and Pollet 2009)

Hong and Yu (2009) suggest that investor attention is lower during the summer for both institutional investors and noise traders Institutional investors are willing to both buy stocks as well as sell short according to their more sophisticated beliefs regarding fundamental value; hence, their inattention should not have a pronounced asymmetric effect on returns However, for noise traders, earnings announcements by themselves can trigger buying regardless of the earnings news Therefore, I suggest that inattention on the part of net-buying noise-traders during the summer results

in lower returns immediately following the announcement for both positive and negative earnings surprises This leads to my first hypothesis:

H1: Announcement-period abnormal returns are lower during the summer period than during the non-summer period

3.2 Noise Traders’ Effect on the ERC

Hypothesis 1 predicts an overall shift in the level of period returns, irrespective of the earnings news In addition, I examine

announcement-how differential levels of investor attention impact the pricing of the underlying earnings information Ahmed et al (2003) suggest that an increase in noise trading during the online-trading era has increased the ratio of noise traders to sophisticated investors participating in the stock market Since noise traders are assumed to have less precise information than do sophisticated traders, they argue that this has resulted in a decrease in the average precision of investor information prior to earnings announcements The authors conclude that this translates to larger ERC’s as investors revise their beliefs to a larger degree based on the earnings surprise because of their less precise information pre-announcement Thus, they suggest that noise traders are responsible for their finding that ERC’s have increased in the online period (beginning in 1996) for a combined sample

of both positive and negative earnings surprises

If net-buying noise traders tend to increase ERC’s for all earnings surprises, then one might expect that summer noise-trader inattention may produce lower ERC’s for all summer announcements However, it is possible that noise-trader effects on the ERC depend upon the sign and/or magnitude of the earnings surprise This is because there may be significant noise-trader buying following both positive and negative earnings surprises (Lee 1992), especially for large surprises in either direction that catch noise traders’ attention (Barber and Odean 2008) It follows that the relative inattention

of net-buying noise traders during the summer may result in a smaller positive price reaction to a unit of positive earnings surprise, especially for large positive surprises However, this summer decrease in noise trader net-buying may result in a larger downward price reaction to a unit of negative earnings surprise, especially for large negative surprises Therefore, while a strict extension of Ahmed et al (2003) would predict that summer ERC’s, compared to those of non-summer, are smaller for negative earnings surprises, I expect that the absence of net-buying noise traders causes summer ERC’s to be larger for negative earnings surprises Thus, I hypothesize that summer earnings announcements, compared to non-summer announcements, are associated with a smaller ERC for positive earnings surprises and a larger ERC for negative surprises This forms my second hypothesis:

H2: For positive (negative) earnings surprises, the ERC is smaller (larger) for summer announcements than for non-summer earnings announcements

3.3 Pre-Announcement Period Returns

The relative summer inattention of noise traders may prompt changes in institutional investor behavior Sophisticated investors may rationally decide to profit from this noise-trader behavior by buying in the pre-announcement period and then selling into the noise–trader buying following the announcement Consistent with this view, Frazzini and Lamont (2010) find increased institutional-investor buying pre-announcement along with increased institutional-investor selling pressure following the announcement However, this selling does not completely eliminate the earnings announcement premium This could be due to market frictions that prevent arbitragers from fully correcting the mispricing caused by noise traders (Lamont and Thaler 2003)

If institutional investors front-run anticipated buying of noise traders, then, during the summer, fewer institutional investors may be buying in the pre-announcement period This could be because they expect fewer noise traders to be buying in the announcement period and, consequently, less profit opportunity to exploit This would be consistent with the findings of Hong and Yu (2009), who suggest that trading activity decreases in the summer for both noise traders as well as institutional investors This leads to my third hypothesis:

H3: Pre-announcement-period abnormal returns are lower during the summer period than during the non-summer period

3.4 The Effect of the Online Period

The proliferation in recent years of 24-hour news dissemination and online, low-cost trading may have impacted the traditional differences in the summer vs non-summer periods Prior research (Ahmed et al 2003, Barber and Odean 2002) suggests that the emergence of online trading has led to more overall noise trading, where online investors, compared to professional investors, are thought to be less sophisticated and profitable (Barber and Odean 2002) It is reasonable to suggest that the overall increase in noise trading in the online period has created a larger difference between the summer vs non-summer levels of noise trading This is because the more noise trading there is in the non-summer period, the larger the potential decrease in noise trading that will occur due to the summer inattention of

noise traders Simply stated, the level of inattention on the part of noise traders is relevant only when they have the ability to trade in the first place The innovations of the online period have enhanced this ability My focus is on differences in investor attention between summer and non-summer periods; therefore, I examine whether the effects hypothesized in H1 through H3 are stronger in the online period than in the pre-online period Since I believe the summer vs non-summer difference in noise-trading activity has increased in recent years, I believe the effects that I am already testing have become more pronounced in the online period This leads to my next three hypotheses:

H4: The difference between summer and non-summer announcement-period abnormal returns is greater in the online period than in the pre-online period

H5: The difference between summer and non-summer ERC’s is greater in the online period than in the pre-online period

H6: The difference between summer and non-summer pre-announcement-period abnormal returns is greater in the online period than in the pre-online period

Consistent with my analysis of H1- H3, I test H4- H6 by examining positive and negative earnings surprises separately as well as in a combined sample

3.5 Post-Earnings Announcement Drift

As previously discussed, I believe that an absence of net-buying noise traders in the summer results in less positive returns immediately following the earnings announcement As part of my study, I also examine the longer-term returns that occur in the post-announcement (drift) period Expectations regarding these returns are less clear The relative inattention of net-buying noise traders in the summer is expected to result

in a less upward-biased price immediately after the earnings announcement All else equal, any post-earnings announcement drift would be relatively more symmetric than the drift following non-summer announcements, assuming prices gravitate to fundamental values during the summer post-announcement period as they may in the non-summer However, if there is also relatively less

attention on the part of institutional investors who trade according to fundamental value then there may be a greater delay in the price discovery process during the summer This could exacerbate the price drift in the direction of the earnings surprise (positive and negative) as these investors later revisit the earnings information

Dellavigna and Pollet (2009) find that Friday earnings announcements are associated with significant price drift in both directions as investors re-emerge and begin to correct the initial underreaction as early as the following week While Friday earnings announcements may be characterized by relatively high levels of inattention on the part of both sets of investors, the results of Dellavigna and Pollet (2009) are likely driven primarily by institutional investor inattention at the time of a Friday earnings announcement Relative inattention of noise and sophisticated traders are likely to have distinct effects on the nature of the summer PEAD As noted, institutional investor inattention may cause an initial underreaction in price followed by a symmetrical price drift in the direction of the earnings surprise Thus, prices might be slower to gravitate to fundamental values in the summer Alternatively, noise-trader inattention may cause a lack of upward-biased price reaction to the earnings announcement Assuming that the earnings information is not fully impounded in prices at the time of the announcement, as suggested by prior research, summer announcements might be followed by a more symmetric price drift than for non-summer announcements Announcement-period inattention on the part of both sets of investors may result in price drift in both directions, with the positive price drift being stronger than the negative drift Adding to the difficulty in predicting price drift is the uncertainty regarding the timing and extent of the re-emergence of investors necessary to create the price drift Hong and Yu (2009) find that returns for the entire summer period are generally lower than for the non-summer period This is consistent with there being less summer noise-trader attention during the announcement period as well as the post-announcement period—at least until the end of the summer In other words, investors may not re-emerge before the end of the summer period to correct any initial underreaction I am not sure which of the effects will dominate Therefore, I expect that there is a summer vs non-summer difference in post-announcement price drift This becomes my next hypothesis:

H7: There is a summer and non-summer difference in longer-term, announcement price drift

post-3.6 Trading Volume

Hong and Yu (2009) find that both trading activity and returns are lower during the summer Both effects may be caused, at least in part, by the relative inattention of noise traders and, thus, the relatively less upward price pressure during the summer Because noise trader attention should be positively correlated with total trading volume, I use the difference in total trading volume between the summer and non-summer announcement periods as a proxy for the difference in noise trader attention

I suggest that the greater the difference in summer vs non-summer attention

of noise traders (as proxied by the difference in summer vs non-summer announcement-period trading volume), the greater the difference in summer vs non-summer announcement returns as well However, it may be the case that factors such as high-frequency trading sufficiently distort the relationship between returns and trading volume such that I do not find an association This leads to my next hypothesis:

H8: The difference in summer and non-summer announcement-period abnormal returns is increasing in the difference between summer and non-summer announcement-period trading volume

3.7 Investor Interest

The possibility that the summer slowdown produces an appreciable decrease in noise trader activity is directly related to a stock’s level of non-summer, baseline noise trader participation Simply put, it is those stocks that noise traders normally trade that should experience a greater effect on volume and returns due to less attention Evidence suggests that noise traders will be more likely to trade in stocks that catch their attention (Barber & Odean 2008) Salience of a stock is likely reflected by the number of analysts following the stock since it is reasonable to assume that analyst following is positively related to overall investor interest

(O’Brien and Bhushan 1990) Therefore, analyst following for a stock should

be directly related to both noise trader participation in the baseline, summer period as well as the drop-off in noise trader participation during the summer This large reduction in noise trader participation may result in

non-a lnon-arge reduction in both non-announcement-period non-abnormnon-al returns non-and trnon-ading volume This leads to my next hypotheses:

H9A: The difference in summer and non-summer announcement-period abnormal returns is increasing in analyst following

H9B: The difference in summer and non-summer announcement-period trading volume is increasing in analyst following

Another possible proxy for investor interest is the market-to-book ratio (MTB) The MTB has been used to differentiate ―growth‖ or ―glamour‖ stocks from ―value‖ stocks Therefore, all else equal, high-MTB stocks may

be more likely to catch the attention of noise traders (Hong and Stein 2007) Therefore, a stock’s MTB may be directly related to both noise trader participation in the baseline, non-summer period, as well as the drop-off in noise trader participation during the summer This large reduction in noise trader participation for high MTB stocks may result in a large reduction in announcement-period returns and trading volume for these firms My next hypotheses are as follows:

H10A: The difference in summer and non-summer announcement-period abnormal returns is increasing in market-to-book ratio

H10B: The difference in summer and non-summer announcement-period trading volume is increasing in market-to-book ratio

CHAPTER 4 RESEARCH DESIGN 4.1 Sample

I use a sample period of 1990-2009 so as to include data from both the pre-online and online periods Consistent with Ahmed et al (2003), I define the 1990-1995 (1996-2009) period as the pre-online (online) period Consistent with prior literature (Hong and Yu 2009), I define the summer period as the months of the 3rd calendar quarter (July through September) My study focuses on the effect that the summer slowdown has on the reaction to

an earnings announcement Therefore, I define an earnings announcement as a summer (non-summer) announcement if the announcement actually took place from July through September (October- June) The summer slowdown is a widespread and consistent phenomenon; therefore, I am able to use a large sample of U.S stocks subject to data availability Specifically, I use firm-quarter observations that meet the following requirements:

 Earnings announcement dates and analyst forecast data available in First Call

 Price and Trading Volume data available in CRSP

 Financial data available in Compustat

 Share price on the trading day before the earnings announcement of at least $5.003

Evidence suggests that the price and/or trading volume reaction to an earnings announcement may be affected by whether or not the announcement pertains to the fourth fiscal quarter (Mendenhall and Nichols 1988) Thus, I examine earnings announcement observations from only non-fourth fiscal quarters To be able to make summer vs non-summer firm-year comparisons, I also discard non-fourth quarter observations in the cases in which the summer announcement was the fourth-quarter announcement Because I am comparing summer vs non-summer earnings announcements for the same firms, I avoid the potential problems associated with the need to create matched pairs of

3 I impose the stock price requirement because of the potential clientele effects associated with stocks priced below $5.00 (―penny stocks‖) In addition, an abnormally low stock price may be problematic when scaling an earnings surprise by price or computing percentage returns Fluctuations in returns due to the bid-ask bounce can introduce noise in these situations as well, especially in the pre-decimalization, larger-spread portion of my sample

similar firms Also because I am using the same set of firms, my results should not be affected by the differences in trading volume calculation methodology among exchanges.4 My final sample consists of 156,122 firm-

quarter observations

4.2 Measuring Return

To measure the cumulative abnormal return during the earnings

announcement event window, I define a variable, CAR, defined as follows:

CAR = (Raw Return – Expected Return Using the Market Model over the {t= -

300, -46} estimation window) compounded over the event window

Note: For the sake of brevity I suppress subscripts when describing variables Unless otherwise noted, variables are measured for each firm-quarter

I measure CAR across different time periods depending upon which hypothesis is being tested Using {t = 0} to represent the day of the

earnings announcement, I define the pre-announcement and announcement periods

4 Due to the nature of the dealer-oriented market of Nasdaq, compared with the auction-oriented markets of the NYSE and AMEX, Nasdaq has traditionally reported higher trading volume for certain types of transactions See Atkins and Dyl (1997) for

a more complete discussion of this topic

5 There are approximately 250 trading days per calendar year Because firms tend to keep the number of days between announcements roughly constant (250/4= 62.5), I measure 75 days from the previous announcement to include the period following the next quarter’s announcement

4.3 Measuring Earnings Surprise

Consistent with prior research (Ahmed et al 2003), I define an

unexpected earnings variable, UE, as follows:

UE= (Actual EPS – (Latest mean consensus forecast)) / (Closing Price 11

trading days before the announcement)

My pre-announcement period starts 10 trading days before the earnings announcement Therefore, I use the closing price 11 trading days before the

announcement in my computation of UE so that it will not be affected by announcement returns I winsorize UE at the 1% and 99% levels to remove the

pre-effect of outliers

4.4 Measuring Trading Volume

Consistent with prior research (Doyle and Magilke (2009), Hong and Yu (2009), among others), I measure trading volume as follows:

Trading Volume = (Number of shares traded) / SharesOutstanding

Where (Number of shares traded) is the average daily number of shares traded over the 3-day {t=0, 2} announcement-period and SharesOutstanding is the

number of shares outstanding on the day of the earnings announcement I scale the number of shares traded by the number of shares outstanding so as

to not have my results unduly affected by those companies with the most shares outstanding My concept of trading volume relates to the proportion

of a firm’s shares traded as opposed to an absolute number of shares

4.5 Models

4.5.1 Testing of H1-H3 To test H1, I first compute {0, 2}

announcement-period returns, by firm-year, for both the summer and non-summer periods I then test for a difference in mean across the two periods Earnings announcements are attention-grabbing events that spur noise-trader buying regardless of the earnings surprises It follows that the summer inattention of net-buying noise traders will cause, on average, a less positive price reaction to all summer earnings announcements Therefore, as suggested in H1, I predict that the mean of summer announcement-period returns will be significantly lower than that of non-summer returns

H2 examines how the differential summer reaction may be related to the news in the earnings announcement I begin by estimating the following earnings response model for the three-day announcement window:

MODEL 1: CAR = B 0 + B 1 Summer + B 2 UE + B 3 Summer*UE + E

Where:

CAR= cumulative abnormal return over the {0, 2} event window

Summer= 1 if the earnings announcement occurred July-September, = 0 otherwise UE= unexpected earnings scaled by stock price

H2 suggests that non-summer earnings announcements, compared to summer announcements, are associated with a larger ERC for positive earnings surprises and a smaller ERC for negative surprises Consequently, I test H2 using Model 1 on sub-samples of positive earnings surprises (―meet-or-beat‖ firms) and negative earnings surprises (―miss‖ firms) individually Collins and Kothari (1989) examine factors that help explain cross-sectional variation in the ERC, including growth, persistence of earnings, riskless interest rates, and systematic risk I do not control for such variables in Model 1 because I am examining seasonal (summer vs non-summer) variation, as opposed to cross-sectional variation, and I have no reason to believe that there are systematic summer vs non-summer differences in these variables

In order for these variables to cause summer vs non-summer differences in ERC’s, the level of these variables would need to repeatedly shift from non-summer levels, to summer levels, and back again to non-summer levels within the same year This seems inconsistent with the normal, more gradual change

in these variables, which are often most-effectively calculated using annual data These annual measurements would be too coarse for my study, which most often considers 3 firm-level observations within a single 9-month period

B 2 , the coefficient on UE, represents the ERC for non-summer

announcements and is expected to be significantly positive B 3 , the

coefficient on the interaction term Summer*UE, captures the incremental effect on the ERC for summer announcements It follows that, consistent with

H2, I predict that B 3 will be significantly negative (positive) for positive

(negative) earnings surprises B 1, the coefficient on Summer, captures the

difference in returns between summer and non-summer, after controlling for

the earnings surprise Finding B 1 to be significantly negative would be

consistent with H1

H3 suggests that pre-announcement returns are lower during the summer

To test H3, I estimate Model 1 over the {-10, -1} event window after removing

“B 2 UE” and “B 3 Summer*UE” I remove unexpected earnings (UE) from the model

since it can be argued that I should not control for them in this situation This is because the motivation for any single investor to buy in the pre-announcement period is increasing in the degree to which he believes the company will beat the analysts’ estimates and, consequently, the stock price will increase following the announcement Therefore, controlling for unexpected earnings in my test of pre-announcement returns may remove the effect for which I am testing Consistent with H3, I expect the coefficient

on Summer to be significantly negative for the combined sample as well as for

each sub-sample. 6

4.5.2 Testing of H4-H6 To test H4-H6, I add a binary variable,

PostOnline, to Model 1 PostOnline = 1 if the earnings announcement took

place from 1996-2009, = 0 for observations from 1990-1995 I run separate regressions for the combined sample as well as each sub-sample using the following model:

MODEL 2: CAR= B 0 + B 1 Summer + B 2 UE + B 3 Summer*UE + B 4 PostOnline +

B 5 Summer*PostOnline + B 6 UE*PostOnline + B 7 Summer*UE* PostOnline + E

I suggest that the overall increase in noise trading in the online period has created a larger difference between the summer vs non-summer levels of noise trading As stated in H4 (H6), this should manifest in a larger difference in summer vs non-summer announcement-period (pre-

announcement-period) abnormal returns B 5, the coefficient on the interaction

term Summer*PostOnline, represents the incremental difference in summer vs

non-summer returns during the online period, independent of the earnings

news Consistent with H4 (H6), I predict that B 5, will be significantly negative for the combined sample of announcement-period returns (pre-announcement-period returns) as well as for each sub-sample individually

H5 states that the difference in summer vs non-summer ERC’s has increased in the online period Therefore, I predict that the coefficient on

the interaction term Summer*UE*PostOnline will be significant for both

positive and negative earnings surprises However, I expect the sign of the

6 I also consider H3 while controlling for unexpected earnings (UE) since it may be

the case that pre-announcement returns are lower in the summer due to lower unexpected earnings instead of less noise-trader activity

coefficient on the interaction term to be negative for positive earnings surprises and to be positive for negative earnings surprises This follows directly from my prediction that for positive (negative) earnings surprises, the ERC is larger (smaller) for non-summer announcements than for summer earnings announcements

4.5.3 Testing of H7 The presence and magnitude of longer-term

post-announcement price drift may vary based upon which group(s) of investors are paying attention at the time of the earnings announcement Institutional-investor inattention may cause an initial underreaction in price followed by greater price drift in the direction of the earnings surprise, i.e., both positive and negative drifts Alternatively, noise-trader inattention may mean less upward-biased summer announcement returns followed by a more symmetric price drift than following non-summer announcements Announcement-period inattention on the part of both sets of investors may result in price drift in both directions

Adding to the difficulty in predicting price drift is the uncertainty regarding the timing and extent of the re-emergence of investors necessary to create the price drift Hong and Yu (2009) find that returns for the entire summer period are generally lower than for the non-summer period This is consistent with there being less summer noise-trader attention during the announcement period as well as the post-announcement period, at least until the end of the summer In other words, investors may not re-emerge before the end of the summer period to correct any initial underreaction Thus, it remains an empirical question whether there is a summer vs non-summer difference in the longer-term price drift that occurs following the announcement-period I do not expect these possibly conflicting factors responsible for price drift to perfectly cancel themselves out Therefore,

as stated in H7, I expect there to be a difference between summer and summer price drift

non-To test H7, I first run regressions of longer-term post-announcement

CAR’s for the combined sample as well as each sub-sample using the following

models:

CAR{3,50} = B 0 + B 1 Summer + B 2 CAR{0,2} + B 3 Summer*CAR{0,2} + E

CAR{3,75} = B 0 + B 1 Summer + B 2 CAR{0,2} + B 3 Summer*CAR{0,2} + E

In addition, I test the sensitivity of these results by adding in

unexpected earnings (UE) in additional regressions using the following

models:

CAR{3,50} = B 0 + B 1 Summer + B 2 UE + B 3 Summer*UE + E

CAR{3,75} = B 0 + B 1 Summer + B 2 UE + B 3 Summer*UE + E

B 1 , the coefficient on Summer, represents the incremental difference in longer-term CAR between the summer and non-summer periods B 2 represents the

incremental change in longer-term CAR for a given change in either CAR{0,2}

or UE, depending upon the regression in question Finding B 2 to besignificantly positive is consistent with longer-term price drift in the same

direction as announcement-period CAR or UE B 3, the coefficient on the interaction term, represents the incremental difference in this price drift between the summer and non-summer periods Since H7 suggests that there is

a difference between summer and non-summer price drift, I predict that B 3 will

be significant for the combined sample as well as each sub-sample individually

4.5.4 Testing of H8 H8 concerns the possible relationship between a

firm’s summer vs non-summer announcement-period abnormal returns and the difference between its summer vs non-summer announcement-period trading volume To test H8, I run the following regression on the combined sample:

MODEL: CARDiff = B 0 + B 1 VolumeDiff + E

Where: CARDiff = [(Average non-summer announcement-period CAR) – (Summer

announcement-period CAR)] by firm-year

VolumeDiff = [(Average non-summer announcement-period trading volume)

– (Summer announcement-period trading volume)] by firm-year

All else equal, a decrease in trading volume may be directly related to

a decrease in net-buying noise trader activity and, thus, abnormal returns

over the same period Therefore, I expect that B 1 , the coefficient on

VolumeDiff, will be significantly positive

4.5.5 Test of H9 & H10 It stands to reason that it is those stocks

that noise traders normally trade that should experience the most significant effects during their summer absence Evidence suggests that noise traders

trade in attention-grabbing stocks (Barber & Odean 2008) It is reasonable

to assume that MTB and analyst following are positively related to salience and, thus, baseline non-summer noise trader participation These same stocks should experience a larger decrease in noise trader participation and, therefore, a larger decrease in announcement-period abnormal returns and trading volume during the summer

H9 suggests a direct relationship between a firm’s analyst following and the summer vs non-summer difference in both its abnormal returns (H9A) and trading volume (H9B) H10 suggests a direct relationship between a firm’s market-to-book ratio and the summer vs non-summer difference in both its abnormal returns (H10A) and trading volume (H10B) To test H9 & H10, I sort the combined sample into decile ranks, by year, based on both the market-to-book ratio (MTB) and the number of analysts making an EPS estimate for each announcement observation.7 I use decile ranks because I do not expect the proposed relationships in H9 & H10 to be purely linear I test

H9A and H10A using the following regression in which Analysts and MTB are

included both separately and simultaneously:

MODEL: CARDiff = B 0 + B 1 Analysts + B 2 MTB + B 3 UEDiff + E

Where: CARDiff = [(Average non-summer announcement-period CAR) – (Summer

announcement-period CAR)] by firm-year;

Analysts = the decile rank of the number of analysts making an EPS

estimate for the summer announcement in a given year

MTB = the decile rank of the market-to-book ratio measured using the

closing price 11 trading days before the announcement and measured using book value at the beginning of the fiscal year

UEDiff = [(Mean of the non-summer unexpected earnings) – (Summer

unexpected earnings)] by firm-year I include UEDiff because summer vs

non-summer differences in UE can be expected to generate differences in CAR, regardless of Analysts and MTB

Consistent with H9A and H9B, I predict that the coefficients on

Analysts and MTB will be significantly positive

I test H9B and H10B using the following regression in which Analysts and MTB are included both separately and simultaneously:

7 Note that the lowest number of analysts in the sample will be one since an

―unexpected earnings‖ variable could not have been calculated without at least one analyst estimate

MODEL: VolumeDiff = B 0 + B 1 Analysts + B 2 MTB + B 3 AbsoluteUEDiff + E

Where: VolumeDiff = [(Average non-summer announcement-period trading volume)

– (Summer announcement-period trading volume)] by firm-year

Analysts = the decile rank of the number of analysts making an EPS

estimate for the summer announcement in a given year

MTB = the decile rank of the market-to-book ratio measured using the

closing price 11 trading days before the announcement and measured using book value at the beginning of the fiscal year

AbsoluteUEDiff = [(Mean of the absolute values of non-summer

unexpected earnings) – (Absolute value of summer unexpected earnings)]

by firm-year;

In my test of H9B and H10B, I attempt to control for the difference in unexpected earnings because this may affect trading volume Since the ability of an earnings surprise to catch an investor’s attention might be increasing in the absolute value of unexpected earnings for both positive and negative surprises, I use the summer vs non-summer difference in absolute value of unexpected earnings (measured as |UEnon-summer| - |UEsummer|) as a control variable The summer absence of noise traders should most greatly affect the stocks with a large analyst following and/ or market-to-book ratio, causing these stocks to experience the largest difference between summer and non-summer announcement-period trading volume Therefore, consistent with H9B

and H10B, I expect that the coefficients on both Analysts and MTB will be

significantly positive

CHAPTER 5 RESULTS 5.1 Descriptive Statistics

In Table 1, Panel A, I describe the formation of my final sample of 156,122 earnings announcement observations from 1990 – 2009 (26,413 observations from the pre-online period; 129,709 from the online period) In Panel B, I present a correlation matrix For all variables, the Pearson and

Spearman correlation coefficients are qualitatively similar CAR’s over each

of the {-10,-1}, {0,2} , {3,50}, and {3,75} windows are significantly

correlated with UE (CAR’s {-10,-1} and {0,2} are positively correlated with

UE; CAR’s {3,50} and {3,75} are negatively correlated with UE) In addition,

all CAR’s are significantly positively correlated with each other, with the exception of CAR{-10,-1} and CAR{0,2} being significantly negatively

correlated with each other Among many explanations, this negative correlation may be the result of a possible direct relationship between institutional-investor buying in the pre-announcement period and institutional-investor selling in the announcement period

5.2 Analysis

5.2.1 Test of H1 Table 2 provides descriptive statistics of variables

under consideration In addition, it shows the results of a test of the difference in the mean of these variables between summer and non-summer observations Earnings announcements are attention-grabbing events that spur noise-trader buying regardless of the earnings surprises It follows that the summer inattention of net-buying noise traders will cause, on average, a less positive price reaction to all summer earnings announcements Therefore, as suggested in H1, I predict that the mean of summer announcement-period returns will be significantly lower than that of non-

summer returns To Test H1, I test for a difference in mean of CAR {0, 2}

between the summer and summer periods Consistent with H1, the

non-summer mean of CAR {0, 2} is significantly higher than non-summer mean of CAR {0,

2} summer period (.000237 vs -.001503, p-value= 0002) for the combined

sample This difference remains significant for each subsample of Positive Surprises (.010345 vs .008169, p-value <.0001) and Negative Surprises (-

.022232 vs -.025078, p-value= 0009) individually These results are consistent with the belief that there is less noise-trader buying following summer earnings announcements than there is following non-summer

announcements Note that this higher non-summer mean for CAR{0, 2} does not appear to be the result of larger unexpected earnings (UE) during the non- summer period In fact, as shown in Table 2, UE is significantly larger in

the summer period for the combined sample (.000404 vs .000246, p-value= 0004) The difference between the two is not significant for either sub-sample (for positive surprises, 002975 for summer vs .002965 for non-summer, p-value =.8854; for negative surprises, -.00586 for summer vs -.00580 for non-summer, p-value =.5446)

5.2.2 Test of H2 H2 examines how the differential summer price

reaction may be related to the news in the earnings announcement I predict that non-summer earnings announcements, compared to summer announcements, are associated with a larger ERC for positive earnings surprises and a smaller ERC for negative surprises I test H2 by estimating Model 1 over the three-day announcement window {0, 2} for sub-samples of positive and negative earnings surprises individually The results of my test of H2 are displayed

in Table 3 As predicted, the coefficient B2 is significantly positive for the combined sample (coeff = 1.45901, p-value <.0001) and each sub-sample (for positive surprises, coeff = 87416, p-value <.0001; for negative surprises, coeff.= 57360, p-value< <.0001) This suggests a direct

relationship between announcement-period CAR and unexpected earnings H2 suggests that B 3 , the coefficient on the interaction term Summer*UE, will be

significantly negative (positive) for positive (negative) earnings surprises However, as shown in Table 3, Panels B and C, my results do not support H2

B 3 is not negative and is insignificant for positive surprises (coeff.=

.02314, p-value = 7919) In addition, B 3 is positive but insignificant for negative surprises (coeff.= 04688, p-value = 6186) Therefore, I do not find evidence of a differential reaction to unexpected earnings between the summer and non-summer periods It may be the case that the predicted differential reaction in H2 is found primarily in large positive earnings surprises, which may catch the attention of noise traders to a larger degree than do other earnings surprises To test this view, I create a dummy

variable, LargeUE, to represent the largest earnings surprises within the positive surprise subsample Specifically, LargeUE is equal to 1 when the value of UE is in the highest quartile among positive surprise observations

LargeUE is equal to 0 otherwise The results of this test are displayed in

Panel D of Table 3 B7, the coefficient on the interaction term,

Summer*UE*LargeUE, is negative and insignificant (coeff = -1.28141, p-value

= 1186) Therefore, similar to my results for the entire sample, I do not find evidence of a differential reaction to unexpected earnings between the summer and non-summer periods for the sample of the largest positive earnings surprises

B 1, the coefficient on Summer, captures the difference in returns

between summer and non-summer, after controlling for the earnings surprise

B 1 is significantly negative for both the positive and negative sub-samples

(for positive surprises, coeff = -.00225, p-value =.0001; for negative surprises, coeff.= -.00254, p-value= 0064) These results are consistent with H1, and suggest that the summer period is associated with lower

announcement-period CAR’s, even after controlling for unexpected earnings This is consistent with my hypothesis that summer announcement-period CAR’s

are lower due to investor inattention instead of being the result of summer announcements containing more negative news

5.2.3 Test of H3 H3 suggests that pre-announcement returns are lower

during the summer than in the non-summer period Evidence in support of H3

can be found in Table 4 The mean of CAR {-10, -1} is less for the summer

period for the combined sample (coeff = -.00300, p-value <.0001) as well as for each sub-sample individually (for positive surprises, coeff = -.00375, p-value <.0001; for negative surprises, coeff.= -.00186, p-value= 0310) These results are consistent with the belief that pre-announcement returns are lower during the summer than in the non-summer period due to summer noise-trader inattention which, in turn, leads to less front-running I also

consider H3 while controlling for unexpected earnings (UE) since it may be

the case that pre-announcement returns are lower in the summer due to lower unexpected earnings instead of less noise-trader activity To control for

UE, I estimate Model 1 over the {-10, -1} pre-announcement-return window

Results for these regressions are displayed in Table 4, Panels D-F

Consistent with H3, the coefficient on Summer, B 1, is significantly negative for the combined sample (coeff = -.00300, p-value <.0001) as well as for the positive sub-sample individually (for positive surprises, coeff = -.00254, p-value =.0001; for negative surprises, coeff.= -.00114, p-value= 1680)

Note that B 2 , the coefficient on UE, is significantly positive for all sample

groups (for the combined sample, coeff = 43663, p-value <.0001; for positive surprises, coeff = 22013, p-value =.0001; for negative surprises, coeff.= 17917, p-value= 0017) This is consistent with there being a direct