Amsterdam Business School
Complexity and
Post Earnings Announcement Drift
Name: Douwe de Vries Student number: 11206098
Thesis supervisor: Dr. Pouyan Ghazizadeh Date: 24 June 2018
Word count: 21,513
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Douwe de Vries who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
Market participants track earnings announcement dates and discuss the contents of earnings releases in great depth because they use this information in their decisions regarding buying and selling of a firm’s shares, and because they attempt to exploit any market anomalies resulting from these earnings announcements. This study examines the relationship between firm complexity and post earnings announcement drift. Post earnings announcement drift is the tendency for a stock’s cumulative abnormal returns to drift in the direction of an earnings surprise for several weeks and months following an earnings announcement. I measure complexity at two different levels, namely first in terms of the degree of diversification of a firm (number of operating segments), and second in terms of the accounting complexity of the industry (industry-level ranking) in which a firm is active. The theory of the efficient market hypothesis (EMH) proposes that stock prices adjust instantaneously to new information. However, a wide body of research has established that the anomaly of the post earnings announcement drift poses a serious challenge to the EMH theory, since it is regarded as a delayed response to earnings announcements as a result of investors' inefficient processing of earnings information into stock prices. Given that diversification and accounting complexity increase the complexity of earnings information and theorizing that therefore a consequent delay in response may be expected, I hypothesize in this study that complexity increases post earnings announcement drift. However, after conducting my tests, I find (1) no evidence that the extent of post earnings announcement drift increases with the extent of complexity measured in terms of the level of diversification. Moreover, I also find (2) no evidence that post earnings announcement drift increases with the extent of industry-level accounting complexity. Finally, my study (3) does also not provide evidence suggesting that the implementation of accounting standard SFAS 131 on enhanced industry segment disclosure (aiming at increased transparency) decreases the extent of post earnings announcement. Overall, my results suggest that the impact of complexity on post earnings announcement drift is not significant. My study on the other hand does point in the direction of a strong association between market volatility and post earnings announcement drift, which provides avenues for future research.
Contents:
1. Introduction... 5
2. Literature review and hypothesis development... 10
2.1. Efficient market hypothesis... 10
2.2 Post earnings announcement drift... 11
2.3 Diversification... 13
2.4 Accounting complexity... 14
2.5 SFAS 131 Disclosures about segments of an enterprise and related information... 17
2.6 Hypotheses development... 18
3. Data, Variables and Methodology... 22
3.1 Sample selection and data... 22
3.2. Measurement of variables... 23
3.3. Empirical models... 25
4. Empirical results... 29
4.1. Descriptive statistics... 29
4.2 Correlations... 33
4.3 The association between diversification and post earnings announcement drift... 34
4.4 The association between accounting complexity and post earnings announcement drift... 40
4.5 Impact of SFAS 131 on post earnings announcement drift... 42
4.6 Additional tests... 46
5. Conclusions... 47
References... 49
1 Introduction
Market participants track earnings announcement dates and discuss the contents of earnings releases in great depth in analyst reports, industry publications, and the like (Basu et al. 2013). Market participants’ keen attention suggests that earnings releases convey a lot of new information and indeed, besides dividend, earnings announcement news is the most important piece of financial information used by investors for decisions regarding buying and selling of any firm’s shares (Lonie et al., 1996). For investors, the efficiency of capital markets is of much concern, and they attempt to exploit any market anomalies, including the ones that are related to earnings announcements.
Ball and Brown (1968) were the first to observe and report the post earnings announcement drift phenomenon, when they found for the years 1957-1965, that abnormal stock returns continue to drift significantly in the direction of an earnings surprise. The drift starts on the first day after an earnings announcement and continues through to the sixtieth day after the announcement of earnings. This phenomenon may be seen as posing a serious challenge to Fama's (1970) 'efficient market hypothesis', which proposes that 'the primary role of the capital market is the allocation of ownership of the economy's capital stock in a market that is ideal or 'efficient' if security prices at any time 'fully reflect' all available information'. Researchers generally regard this anomaly of post earnings announcement drift as a delayed response to earnings announcements, mainly as a result of information uncertainty and the inability of investors to instantaneously and correctly process all information included in earnings announcements into stock prices. Where some researchers (e.g. Livnat and Mendenhall, 2006) suggest that the drift is attributable to the inaccuracy of earnings expectations and analyst's forecasts by not fully incorporating all past information that is available, another stream of research argues that operational complexity results in increased information uncertainty, thereby making the task of processing earnings into stock prices more difficult for analysts and investors (e.g. Thomas, 1999). From this viewpoint, complexity may be regarded as a capital market friction.
In this paper, I investigate whether complexity, both in terms of the level of operational diversification of the individual firm as well as in terms of the accounting complexity of the industry as a whole in which a firm operates, is related to the phenomenon of the post earnings announcement drift. Specifically, I examine whether firms that operate in multiple operating segments (e.g. firms with a higher degree of diversification) experience a stronger post earnings announcement drift. Additionally, I examine whether industries as a whole, which are associated with a higher accounting complexity score (a composite measure which ranks industries with
regard to ten different sub-proxies), experience a stronger post earnings announcement drift than do industries with a lower accounting complexity score. Finally, I assess whether Statement of Financial Accounting Standards No. 131 (SFAS 131), Disclosures about Segments of an Enterprise and
Related Information, issued in 1997 with the objective of providing increased transparency to
investors by increasing the requirements for segment disclosure, mitigates the hypothesized impact of complexity resulting from diversification on post earnings announcement drift.
As noted above, the efficient market hypothesis (EMH) is a theory in financial economics proposing, that 'a market is ideal or efficient if security prices at any time fully reflect all available information'. Regarding this theory, academic researchers agree, that one important potential source of market inefficiency is the disagreement on implications of current information on future prices of a security. One phenomenon, academically generally regarded to be seen as an outcome of this disagreement on future stock prices between investors, is post earnings announcement drift, which was first observed by Ball and Brown (1968), as mentioned earlier. Based on the above, it may be argued, that complexity (operational and accounting complexity) is a potential source of market inefficiency because it impacts on the level of disagreement on the implications of current information (e.g. earnings information) for future stock prices, which consequently may delay the processing speed of this information. Therefore, in this study, I hypothesize that complexity, both in terms of the level of a firm's diversification, as well in terms of the degree of accounting complexity that is associated with a specific industry, is positively related to post earnings announcement drift. Besides, I hypothesize that the expected association between post earnings announcement drift and diversification will be less pronounced shortly after the implementation of SFAS 131. I expect to find this effect, because SFAS 131 aims to provide increased transparency in segment disclosure to investors, thereby reducing complexity and uncertainty of earnings announcements, which consequently results in a more efficiently and more timely processing of relevant information into stock prices.
I test my hypothesis on a sample of 300,519 firm-quarter observations (4,957 firms) in the United States. My sample period is 1 January 1990 until 31 December 2017. Following Barinov et al. (2014), I measure a firm's level of diversification by using three similar proxies, namely (1) the number of segments a firm operates in, (2) whether or not a firm is a conglomerate and finally (3) the extent of concentration of sales activities of a firm. I measure a firm's accounting complexity by using the industry accounting complexity ranking, which has been developed by Seavey in his study in 2011 on the relationship between accounting complexity and audit quality. In case a firm operates in multiple industries, I calculate an average accounting complexity, which is based on the proportion of sales within the individual industries and Seavey's accounting complexity ranking
for these industries. Post earnings announcement drift is measured as the cumulative abnormal return for a stock during the second day after an earnings announcement until the sixtieth day after an earnings announcement.
In this study, I do not find evidence for Barinov et al.'s (2014) proposed positive relationship between diversification and post earnings announcement drift. This may be the result of my use of a larger sample and/or different sample period. On the other hand, I should not rule out other plausible explanations. For example, in academic research, it is generally agreed that the drift anomaly is normally stronger for smaller, illiquid and more volatile firms. Conglomerates may be regarded as being exactly the opposite of that. Therefore, from a viewpoint that more diversified firms are generally larger, more liquid, less volatile and more sophisticated, it may be argued here that the drift-enhancing effects of complexity as a result of the slower processing efficiency of more complex earnings information of conglomerates is offset by the drift-decreasing effects of these larger, more diversified firms. This seems plausible, since these larger firms may be considered firstly, as being able to produce a higher quality of earnings information, secondly as experiencing a higher accuracy of earning forecasts and thirdly as experiencing a higher market liquidity.
Also, I do not find evidence that accounting complexity is positively related to post earnings announcement drift. In particular, I find no statistically significant increase in the extent of post earnings announcement drift with firms that operate in industries associated with a higher accounting complexity, after controlling for size and volatility. A plausible explanation for not finding evidence here for the relationship between (1) drift and diversification and (2) between drift and accounting complexity may be that market volatility as a variable has a far more important impact on drift than complexity. It might well be, that this impact of volatility more than offsets the effects of complexity on drift. Taking into account my findings with respect to the very strong effect that market volatility has on post earnings announcement drift, I need to concede that my study with regard to the impact of complexity on drift has its limitations. When examining the pattern during the years from 1990 until 2017 of both positive and negative post earnings announcement drift on the one hand, and market volatility on the other hand, measured as the level of the CBOE Volatility Index (VIX), I come to the conclusion that both diversification and accounting complexity are playing only a small, subordinate role, when compared to the effect that market volatility has on investors' response to earnings information. Testing the validity of this impact of market volatility on drift is however beyond the scope of this study.
Finally, I also do not find evidence for the hypothesized mitigating effect of SFAS 131 on post earnings announcement drift. Also, here there may be a plausible explanation for not finding
this effect. First, as a result of the enforcement of SFAS 131, investors in 1998 were suddenly confronted with additional and novel earnings information. It may be argued that although the providing of this additional information to investors regarding a firm's diversification did intrinsically enhance transparency to a certain level, it also led, contrarily and at the same time, initially to more inefficiency of information processing because of the necessity of processing novel and additional earnings information, thus causing a delay in response. In later years this process may have reversed when investors got accustomed to the new information layout and they may have ultimately judged this information to be more valuable in their allocation of capital thereby reducing drift. However, also this hypothesis goes beyond the reach of this study and can therefore not be corroborated here; future research may shed more light on this hypothesis. Also, here we need to note the phenomenon of market volatility, since SFAS 131 was introduced in a period with strongly increasing market volatility. Again, a plausible explanation for not finding the hypothesized mitigating effect of the enforcement of SFAS 131 may also here be that the drift-decreasing effect of increased transparency remains unnoticed because it may have been offset by the drift-enhancing effects of market volatility. This hypothesis also is beyond the scope of this study and may also be seen as a possible avenue for future research.
This study makes several contributions. First, my study is related to the study of how operational diversification of a company is related to post earnings announcement drift. Some researchers (e.g. Barinov et al., 2014) find that post earnings announcement drift is stronger for more diversified firms (e.g. conglomerates), attributing this finding to slower information processing about complex firms. I contribute to this stream of literature, by testing Barinov et al.'s proposed relationship while using a different, novel and very large sample containing very recent data. Second, I contribute to the empirical studies on how the issuance of SFAS 131 affects investors' behavior. Although there are prior studies which have examined the effect of issuance of SFAS 131 on post earnings announcement drift within the context of international, geographical diversification (e.g. Kang et al., 2017), I am the first, as far as I am aware, to investigate the effect of SFAS 131 on post earnings announcement drift on a national-level (e.g. United States). Third, to the best of my knowledge, my paper is the first empirical study on the relationship between industry accounting complexity as a proxy for complexity on the one hand and post earnings announcement drift on the other hand. There have been some studies with regard to the relationship between complexity and post earnings announcement drift using other proxies for complexity, for example the number of operational segments a firm is active in (Barinov et al., 2014) and the word count of the 10-K filing as a measure of information complexity (You and Zhang, 2009). This study is novel however and contributes to this stream of literature in the sense
that it empirically examines this relationship while using a novel proxy for complexity, namely industry accounting complexity (a composite measure which ranks industries with regard to ten different sub-proxies).
The results of this study may be regarded as being important both for researchers in extending their academic studies as well as for investors in their decisions about allocation of capital and in their attempts to exploit market anomalies. The novel use here of accounting complexity as a proxy for studying complexity may be exploited by other researchers in different contexts in future research. Also, the relationship between post earnings announcement drift and market volatility, which is hinted at in this study, may prove to offer fertile ground for further research on this subject. For investors, by showing how volatility is related to the efficiency of processing earnings information into stock prices, this study may prove to be useful in their efforts to exploit the phenomenon of the post earnings announcement drift as an investment opportunity. The remainder of this paper is organized as follows: the next section contains a literature review and hypotheses development. Section 3 explains the data, variables and methodology and Section 4 presents the results of the tests. Finally, Section 5, presents the conclusions of this study.
2
Literature review and hypotheses development
2.1 Efficient market hypothesis
In our modern, capitalist society, capital markets operate and serve as a bridge between fund providers and fund users. Either through direct financing (e.g., issuing financial instruments to the public) or through indirect financing (e.g., borrowings via financial intermediaries), collective funds are made available to business firms, and then channeled into productive uses. Because of the economic relevance of this process, the efficiency of capital markets is a constant source of concern. Historically a lot of resources have been devoted and are still being devoted to accounting research in order to identify and propose solutions to frictions with regard to this allocation efficiency. More specifically, it is the capability of the financial markets to process information into stock prices, that captures the attention of scholars, practitioners and regulators (Lee, 2008).
In his theory on efficient markets, often labelled the 'efficient market hypothesis', Fama (1970) proposes that 'the primary role of the capital market is the allocation of ownership of the economy’s capital stock' and that 'a market is ideal or efficient if security prices at any time fully reflect all available information'. Fama (1970) mentions three potential sources of market inefficiency, namely (i) transaction costs (ii) the fact that all available information may not be costlessly available to all market participants and (iii) the possibility of disagreement on the implications of current information on future prices of each security. Fama (1970) also formally defines three different levels of market efficiency, distinguishing one from the other by the degree of capability of processing information into stock prices in an ascending order. First, there is the 'weak form' market efficiency, where past information relevant to the firm under analysis will be fully and quickly reflected in its present stock price. The second form of market efficiency is labelled 'semi-strong form', which reflects the situation where public information relevant to the firm under analysis will be fully and quickly reflected in its present stock price. Finally, there is the level where all information - publicly available or privately held - relevant to the firm under analysis will be fully and quickly reflected in its present price, which is known as the 'strong-form' market efficiency.
The efficient market hypothesis (EMH) has become an integral part of today's world of accounting and since the publication of Fama's work, a tremendous large number of academic articles have been written, with the aim of either countering, supporting or testing certain assertions by using the theory. Fifty years later, despite the ongoing debate, the theory still stands and is still applied in accounting practice regularly.
A part of the academic community however, is showing increasing dissatisfaction with the EMH, swayed partly by evidence that prices underreact to large earnings changes, ratios of prices to fundamentals, and other statistics derived from fundamental accounting analyses (Bloomfield, 2002). Nevertheless, as Bloomfield (2002) notes, the EMH is still influential because there is no alternative theory that explains why we observe the inefficiencies we do. For example, why should the market underreact to large earnings changes, rather than overreact? It is Fama himself who acknowledges that Bernard and Thomas (1990) identify a significant challenge to his efficient market hypothesis, in the way stock prices fail to adjust efficiently to earnings announcements (Fama, 1991). Bernard and Thomas argue that this challenge to Fama's hypothesis is related to the fact that the market does not fully understand the information incorporated in quarterly earnings, which consequently leads to a delayed stock-price response. This phenomenon is commonly referred to as 'the post earnings announcement drift'.
2.2. Post earnings announcement drift
Ball and Brown (1968) were the first to observe and report the post earnings announcement drift phenomenon. They found for the years 1957-1965, that abnormal stock returns continue to drift significantly in the direction of an earnings surprise, starting on the first day after an earnings announcement and continuing through to the sixtieth day. Foster et al. (1984), while also documenting a similar drift over sixty trading days, examine post earnings announcement drift further and find that the drift is inversely related to the size of a firm and positively related to the sign and magnitude of unexpected earnings. Bernard and Thomas (1989) also reaffirm the previous findings of Ball and Brown and Foster et al., when they examine stock reactions to earnings announcements disclosed by companies listed on the NYSE and NASDAQ for the years 1974-1986. However, Bernard and Thomas also notice that a disproportionately large part of the sixty-day drift occurs within five days after the earnings announcement date. They conclude that their findings can not be reconciled with arguments of risk mismeasurement, but that they are on the other hand consistent with a delayed price response, which according to them and others (Ashtana, 2003; Bartov, 1992) is the result of the fact that markets fail to recognize fully the implications of current earnings on future earnings. Nichols and Wahlen (2004) provide further evidence on the existence of the anomaly of the post earnings announcement drift. In their study, they show that Bernard and Thomas’ findings extend to their data for the years 1988-2002. Also, Abarbanell et al. (1992) and Bernard and Thomas (1997) conclude that the literature on both underreaction and overreaction includes evidence to indicate that the anomalous stock price
behavior around earnings announcements may be rooted in a failure by market participants to appreciate what current earnings imply about future earnings (market inefficiency). Lee and Rui (2010) note that despite repeated attempts, prior research has failed to provide a satisfactory explanation for the drift and they believe that, of the presented explanations for the drift in academic research, investors’ misperception of the earnings process is considered to be the most important.
Bartov et al. (2000) further study the issue of misperception of earnings. The authors replicate Bernard and Thomas' research while at the same time examining the relationship between drift and investor sophistication and they find that post earnings announcement drift is a manifestation of inefficient processing of quarterly earnings. They notice that the degree of inefficient pricing of abnormal returns subsequent to quarterly earnings announcements is negatively correlated with the proportion of a firm’s stock held by institutions. These results suggest that trading activities of institutional investors moderate the earnings-processing problems that cause post earnings announcement drift and increase the degree to which earnings information is efficiently priced (Chung and Hrazdil, 2011). In a similar vein, Hou and Moskowitz (2005) find that firms that experience a high drift, as a result of a delayed price response to information, are generally small, volatile, and neglected by most market participants. Together, the studies mentioned above suggest that the primary cause for the existence of the phenomenon of post-earnings-announcement drift in stock prices is the fact that investors misperceive the process that underlies earnings.
Despite the fact that in academic research the anomaly of post earnings announcement drift is one of the most studied phenomena, and notwithstanding the long-standing analytical debate, 'post earnings announcement drift remains one of the most puzzling anomalies in accounting which has defied all attempts at rational explanation' (Nichols and Wahlen, 2002, p. 284). Some feel that this post earnings announcement drift poses a serious threat to Fama's efficient market hypothesis (Kothari, 2001) since the nature of this market reaction is contrary to the logic of the efficient market hypothesis, which predicts stock prices to adjust to new information instantaneously. In one of his later studies, Fama (1998, p. 304) himself also acknowledged that the anomaly of post earnings announcement drift is above all suspicion. He notes that 'the anomaly has survived robustness checks, including extension to more recent data (Bernard and Thomas, 1990; Chan et al., 1996)'.
Concluding, it may be argued that, despite the lack of a rational explanation, academic researchers nevertheless do generally agree on the nature of the drift: post earnings announcement drift, in essence, is to be regarded as a delayed response to earnings announcements, mainly as a
result of uncertainty and the inability of investors to instantaneously and correctly process all information included in earnings announcements into stock prices.
2.3. Diversification
US firms have substantially expanded operations in the last thirty years, whether across industries (operational diversification), across different geographical areas (geographical diversification), or both. Montgomery (1994) reports that in 1992 two-thirds of Fortune 500 US public companies were involved in more than one industry. These organizationally complex firms accounted for nearly 50 percent of US employment and represented about 60 percent of the total assets of publicly traded companies (Martin and Sayrak, 2003). Since then the process of diversification has intensified as a result of globalization and increased merger and acquisition activities. By diversifying across operational segments, firms can expand products markets (Penrose, 1959) and reduce idiosyncratic risk (Ahimud and Lev, 1981). Regarding the value-enhancing aspects of diversification, Lewellen (1971) argues that diversified firms have greater debt capacity and higher firm value, while Chandler (1977) posits that diversification can lead to greater operating efficiency by enhancing economies of scope.
On the other hand, diversification may lead to value destruction, which is ultimately attributable to information asymmetries (Bens and Monahan, 2004). Hill (2015, p. 128), studying the phenomenon of geographical diversification, notes that a downside of diversification is that it can make a firm's operations more complex, because 'the organization's structure becomes intricate, and coordination among operations becomes more difficult, often leading to increased information uncertainty, miscommunication and conflict'. Gilson et al. (2001) find that by combining diverse operations, information aggregation problems are created that can result in substantial information asymmetry between managers and outside investors. Investors can observe aggregate earnings but not individual segment earnings, and as a result the mapping of individual divisional earnings into consolidated earnings is not transparent and reported earnings may convey less value-relevant information (Bushman et al, 2004). Duru and Reeb (2002) show that greater corporate geographic diversification is associated with less accurate and more optimistic forecasts, suggesting that as firms become more geographically diversified, forecasting their earnings becomes more complex (Lai and Liu, 2012). Demirkan et al. (2011) find that industrially diversified firms have poorer earnings quality than focused firms and Bushman et al. (2004) find that the organizational complexity due to industry and geographic diversification might impair earnings timeliness. Lai and Liu (2012, p. 354), on the basis of their analysis of a large set of prior studies,
suggest that 'organizational complexity due to industry and/or geographic diversification limits transparency of diversified firms' operations and information to outside investors, thus amplifying information asymmetry between diversified firms and outside investors'.
Barinov et al. (2014) study the relationship between diversification and post earnings announcement drift and hypothesize that information about complex (more diversified) firms is harder to process. Therefore, they predict that post earnings announcement drift will be stronger for complex firms. They indeed find that the drift per unit of earnings surprise is twice as large from complex (multiple-segment) firms as it is for single segment firms, due to slower-information-processing.
2.4 Accounting complexity
Organizational complexity can be defined in more terms than just the level of diversification. When examining information asymmetry resulting from complexity between firms and investors, it may be argued that solely using diversification as a proxy for a firm's complexity may be too limited and that therefore in research the use of another, broader measure may be desirable.
Complexity in itself is a very broad term. The degree of complexity of a firm can be measured by using a variety of proxies, other than the number of operational segments an individual firm is active in. As regards a definition, the US Security and Exchange Commission describes complexity as ‘the state of being difficult to understand and apply’ (SEC 2008) and Peterson (2008) further proposes that complex accounting specifically pertains to the difficulty for investors in understanding the mapping of transactions (or potential transactions) and standards into financial statements. These two descriptions provide us with some guidance, but inevitably, the question arises how to empirically measure accounting complexity.
Hoitash and Hoitash (2018) develop and evaluate their own measure of accounting reporting complexity (ARC). ARC is based on the count of accounting items disclosed in eXtensible Business Reporting Language (XBRL) 10-K filings. They find that, compared to commonly used measures of operating and linguistic complexity, ARC is associated with greater likelihood of misstatements and material weakness disclosures, longer audit relay, and higher audit fees. Hobson (2011) examines whether reducing the complexity of accounting information will lead to greater processing efficiency of that information. Hobson focuses on the aspect of complexity with respect to revenue recognition in his study and to this aim he uses empirical proxies emanating from the firms’ revenue recognition disclosures. In line with his expectations,
Hobson finds that reducing complexity in information about fundamental value increases market efficiency indirectly by increasing the frequency by which traders process information about fundamental value. The question whether 'simplification' of information may also lead to an increased level of estimation errors made by investors, is however not addressed by Hobson in his study. Filzen and Peterson (2015), in their study on the relationship between financial statement complexity and meeting analysts' expectations, propose that financial statement complexity represents the increased difficulty in understanding, interpreting, and forecasting financial statements. They develop a proxy for their concept of financial statement complexity by using the abnormal length of accounting policy disclosures found in the notes to the financial statements. Baudot et al. (2017) examine the accounting profession's engagement with complexity in accounting standards. They analyze comment letters submitted to the Financial Accounting Standards Board (FASB) over a 12-year period and find that the profession characterizes complexity through three dimensions, namely multiplicity, diversity, and interrelatedness. Baudot et al. conclude that this characterization has consequences for how complexity is thought about and acted upon in accounting standards.
Seavey (2011), in his study on the relationship between industry accounting complexity and the quality of audit outcomes, follows Peterson (2008), who defines complexity as the amount of uncertainty related to the mapping of transactions or potential transactions and standards into the financial statements. From this viewpoint Seavey argues that accounting complexity arises from two related sources, namely firstly from operational complexity, which in its turn generates varying degrees of transactional complexity, and secondly from the application of accounting and regulatory standards. Noting that, on an industry level, different operational and regulatory characteristics give rise to differing levels of transaction-related complexity (Danos et al., 1989; Simunic, 1989; Solomon et al., 1999; Cahan et al., 2008), Seavey argues that accounting complexity is not homogenous across industries, but rather a continuum whereby some industries can, on average be considered more complex than other industries. Along this continuum are industries that have relatively few unique accounting and reporting concerns (e.g. the retail industry), and other industries that have increasingly complex accounting (e.g. chemical industry). Francis and Gunn (2015) underscore the centrality of industry accounting complexity in understanding earnings properties. These authors illustrate how the incorporation of industry accounting complexity can give a more nuanced understanding of accounting phenomena. Exploring the mechanism by which asset prices are sensitive to the complexity of information processing, Cohen and Lou (2012) show that, while it is straightforward to incorporate industry-specific information into a firm operating solely in that industry (i.e., standalone firms), it generally requires more
complicated analyses to incorporate the same piece of information into the price of a firm with operating segments in multiple industries (i.e., conglomerate firms). Cohen and Lou find, by exploiting a setting with two sets of firms that require easy vs. complicated analyses to reflect the same piece of information, that asset prices are sensitive to the complexity of information processing,
In order to establish the level of complexity, Seavey (2011) develops a composite measure of industry accounting complexity, built from ten sub-measures which have been established in prior research. These ten measures will now be presented here below. Seavey follows Boone et al. (2007) and Doyle et al. (2007) by using (1) firm size and the (2) number of firm segments as proxies for complexity, on the premise that larger, more diverse firms have increased organizational complexity. Cahan et al. (2008) find that complex accounting and auditing issues are associated with the nature and level of industry-specific investment opportunity set (IOS). From this view, Seavey proposes that (3) IOS is positively related to more complex accounting and can, on this basis, be employed as a proxy for complexity. Dechow and Dichev (2002) find that the larger, or more complicated a firm's accruals are, the greater the propensity for estimation error in the accrual process, and by implication, the greater the complexity. On the basis of Dechow and Dichev's (2002) findings, Seavey derives four variables as proxies for complexity: firstly a firm's (4) total accruals (operating income minus operating cash flows), secondly the (5) volatility of a firm's operations (variance of cash flows), thirdly the length of a firm's (6) operating cycle (time between acquiring inventory and receiving cash for its sales) and finally the (7) capital intensity of a firm (gross property, plant and equipment scaled by sales). Skinner (2008) poses that, by their very definition, intangibles require subjective assessment and estimation, and that they thus give rise to accounting complexity. Seavey concurs with Skinner's view and argues that the same line of reasoning is applicable to the accounting of research and development (R&D) and on this basis he includes both the level of (8) intangibles and the extent of (9) R&D of a firm as proxies for accounting complexity. Finally, citing the FASB's concerns with regard to the issue of accounting complexity arising from very complex and hybrid financing activities, Seavey argues that firms with more frequent financing activities in the capital market have more complex accounting. Concurring with this view, Seavey constructs his tenth and final proxy for accounting complexity, which is based on a firm's (10) financing activities, measured as the sum of the proceeds from the issuance of equity and debt during a year. From the industry ranking with regard to these ten proxies, Seavey derives a composite industry level ranking for accounting complexity (see Appendix C), which I will use also in this study.
2.5 SFAS 131 Disclosures About Segments of an Enterprise and Related Information
The Financial Accounting Standard Board of the United States of America issued Financial Accounting Standard no. 131, Disclosures About Segments of an Enterprise and Related Information, (SFAS 131) in June 1997, effective for financial statements for periods beginning after December 15, 1997. This statement establishes standards for the way that public business enterprises report information about operating segments in annual financial statements and requires that those enterprises report selected information about operating segments in interim financial reports issued to shareholders. It also establishes standards for related disclosures about products and services, geographic areas, and major customers. SFAS 131 requires that a public business enterprise report financial and descriptive information about its reportable operating segments. Operating segments are components of an enterprise about which separate financial information is available that is evaluated regularly by the chief operating decision maker in deciding how to allocate resources and in assessing performance. SFAS 131 also requires that a public business enterprise report a measure of segment profit or loss, certain specific revenue and expense items, and segment assets. It requires reconciliations of total segment revenues, total segment profit or loss, total segment assets, and other amounts disclosed for segments to corresponding amounts in the enterprise's general-purpose financial statements. SFAS 131 further requires that a public business enterprise report descriptive information about the way that the operating segments were determined, the products and services provided by the operating segments, differences between the measurements used in reporting segment information and those used in the enterprise's general-purpose financial statements, and changes in the measurement of segment amounts from period to period. SFAS 131 became effective for financial statements for periods beginning after December 15, 1997.
Regarding the effects of the issuance of SFAS 131, there exists a considerable body of research. Berger and Hann (2003) investigate the effect of the new segment reporting standard on the information and monitoring environment and find that SFAS 131 increased the number of reported segments, provided more disaggregated information and induced firms to reveal previously 'hidden' information about their diversification strategies. Cho (2015) examines the effect of segment disclosure transparency on internal capital market efficiency and finds that diversified firms, which have improved their segment disclosure transparency by changing segment definitions upon adoption of SFAS 131, experienced an improvement in capital allocation efficiency in internal capital markets after the enforcement of SFAS 131. Hope et al. (2009) conclude that an increase in the number of reported geographic segments, as a result of the
issuance of SFAS 131, can be interpreted as higher-quality information. Chen and Liao (2013) hypothesize that economic benefits from improved segment-reporting quality are attained through the reduction of cost of debt because the capital market takes into account the enhanced transparency and efficiency of multi-segment firms. Chen and Liao indeed find an increase in segment-reporting quality and moreover their results show that the effect of segment-disclosure quality is non-trivial. Wang (2017) documents that enhanced segment disclosures as a result of the issuance of SFAS 131 mitigates information asymmetry and agency costs of debt arising from international diversification. Kang et al. (2017) find that the issuance of SFAS 131 provides more information about a firm's foreign operation, that it reduces uncertainty associated with predicting the earnings of internationally diversified firms and that it might help analysts form more efficient earnings expectations. They conclude that SFAS 131 can reduce the impact of international diversification on the serial correlation of analyst forecast errors and the therewith associated post earnings announcement drift.
2.6 Hypothesis development
The theory of the efficient market hypothesis, which proposes that a market is efficient if security prices at any time fully reflect all available information, still stands almost fifty years after Eugene Fama described it first in 1970. A part of the academic community however is showing increased dissatisfaction with the efficient market hypothesis, fueled partly by evidence that prices underreact to large earnings changes. Fama himself acknowledges that the way stock prices fail to adjust efficiently to earnings announcements, in the sense that abnormal stock returns continue to drift significantly in the direction of an earnings surprise for several weeks and even months after an earnings announcement date, is posing a serious challenge to his hypothesis.
This phenomenon, labelled the 'post earnings announcement drift', is one of the most studied anomalies in academic accounting research. A conclusive, rational explanation has not been offered yet by academics. What researchers do agree on, on the other hand, is that this drift is consistent with a delayed price response, as a result of the fact that markets fail to recognize fully the implications of current earnings on future earnings. In line with this view, there exists a considerable amount of academic evidence showing that the degree of this inefficient processing of quarterly earnings is correlated with the level of sophistication of investors. This may serve as an indication, that complexity is a factor to take into account when studying drift.
Extant literature addresses the costs and benefits of diversification. As regards the possible benefits of diversification mentioned by researchers there are, among others, greater efficiency by
enhancing economies of scale and scope (Chandler, 1977), increased debt capacity and higher firm value (Lewellen, 1971) and the reduction of idiosyncratic risk (Ahimud and Lev, 1981). However, when diversification is poorly executed or emanates from a poor strategy, the benefits may be offset. Academic literature shows that a downside of diversification is that it can make a firm's operation more complex because coordination among operations becomes more difficult, often leading to increased information uncertainty (Hill, 2015). Research also shows that information aggregation problems may result in substantial information asymmetry (Gilson et al., 2001), reported earnings with less value-relevant information (Bushman, 2004), less accurate earnings forecasts (Lai and Liu, 2012) or impaired earnings timeliness (Bushman et al., 2004).
Academic literature has also studied the issue of accounting complexity, which is described by the SEC (2008) as 'the state of being difficult to understand and apply' and which pertains to the difficulty in understanding the mapping of transactions and standards into financial statements (Peterson, 2008). Regarding the issue of measuring accounting complexity, different proxies have been employed by researchers. Examples of these proxies are the count of accounting items disclosed in XBRL (Hoitash and Hoitash, 2018), proxies emanating from a firm's revenue recognition disclosures (Hobson, 2011), the length of accounting policy disclosures found in the notes to financial statements (Filzen and Peterson, 2015) and the complexity level of comment letters submitted to the FASB (Baudot et al., 2017). Seavey (2011) notes that previous research shows that, on an industry level, different operational and regulatory characteristics give rise to differing levels of transaction-related complexity. On this basis, Seavey (2011) argues that accounting complexity is not homogenous across industries, but rather that on average some industries can be considered to be more complex than other industries. From this viewpoint, Seavey (2011) constructs a composite industry-level ranking for accounting complexity, which is composed of a ranking of ten different sub-proxies for complexity.
From the viewpoint of the efficient market hypothesis, one would expect earnings information to translate into stock prices instantaneously, or at least within a very short number of days after the earnings announcement date. However, there exists a wide body of evidence on the existence of the anomaly of the post earnings announcement drift, thus posing a serious challenge to the efficient market hypothesis. Given the extant literature indicating that this drift is to be seen as a delayed response, mainly resulting from the inability of investors to fully understand complex earnings information, I hypothesize that post earnings announcement drift increases with the level of complexity, whether measured as the degree of a firm's diversification or as the level of accounting complexity of the industry (industries) a firm is active in. The above discussion leads to the following two hypotheses:
H1: Diversification is positively related to post earnings announcement drift H2: Accounting complexity is positively related to post earnings announcement drift
The two hypotheses described above and the empirical methods that I use for testing these hypotheses, will be further explained in Section 3.
In 1997, The Financial Accounting Standard Board of the United States of America issued Financial Accounting Standard no. 131, Disclosures About Segments of an Enterprise and Related
Information, (SFAS 131), with the aim of providing increased transparency to investors. Regarding
the effects of the issuance of SFAS 131, there exists a wide body of research, showing for example that (a) SFAS 131 induced firms to reveal previously 'hidden' information about their diversification strategies (Berger and Hann, 2003), that (b) diversified firms, which have improved segment disclosure, experienced an improvement in capital allocation (Cho, 2015), that (c) an increase in the number of reported geographic segments can be interpreted as higher-quality information (Hope et al, 2009) and also that (d) improved segment-reporting quality may result in the reduction of cost of debt (Chen and Liao, 2013). Kang et al. (2017) further find that the issuance of SFAS 131 provides (e) more information about a firm's foreign operation, that it (f) reduces uncertainty associated with predicting the earnings of internationally diversified firms and that (g) it might help analysts form more efficient earnings expectations.
On the basis of the theory and literature giving evidence of the impact of SFAS 131 on a range of accounting phenomena, I conjecture that the disciplinary effect of SFAS 131 in disclosing increased segment reporting will provide investors with more transparency of earnings information. This in its turn will lead to heightened information processing efficiency and consequently to a lower post earnings announcement drift. From this line of reasoning follows my third hypothesis:
H3: Issuance of SFAS 131 reduces post earnings announcement drift
Hypothesis 3 predicts that the issuance of SFAS 131 is associated with increased transparency for investors regarding the diversification level of a firm and consequently is associated with a decrease of the extent of both positive and negative post earnings drift. Hypothesis 3 and the empirical methods that I use to test it, will be further explained in Section 3.
To conclude, the aim of this study is to gauge the effects of diversification and accounting complexity on post earnings announcement drift, while controlling for firm size and market volatility.
3.
Data, Variables and Methodology
3.1 Sample selection and data
My initial sample includes all firms with business segment data on Compustat and daily returns data on CRSP for the years 1990-2017. In line with previous research (e.g. Fama and French, 1992; Nichols and Wahlen, 2004), I exclude firms from the finance industries. Next, firms with no available announcement dates are deleted. Consistent with prior literature (e.g. Kothari et al., 2005; Landsman and Maydew, 2002), I exclude firms for which less than 41 quarterly observations (corresponding to a period of 10 years) are available. This selection method is designed to exclude observations for which the regression models for estimating cumulative abnormal returns are likely to be imprecise. Also, since I measure post earnings announcement drift as the cumulated abnormal returns between trading day 2 and trading day 60 after an earnings announcement, all observations with missing abnormal daily returns during this event period are also excluded. Finally, because I am interested in the general behavior of the variable that serves as a proxy for post earnings announcement drift, I choose to reduce the influence of the few extreme observations in the sample by trimming the data, which in this case involves the filtering out of the most extreme 1 percentiles of the data. Collectively, these filters yield a sample of 300,519 observations. Table 1 shows the sample reconciliation.
TABLE 1
Sample selection
# of Firm-quarters # of Firms
Firms with business segment data on Compustat and
returns data on CRSP for the years 1990-2017 1,258,168 31,872 After deleting firms from the financial services industries 904,355 22,203 After deleting firms with no available announcement dates 733,186 21,072 After removing observations with less than 41 quarterly observations 483,894 6,863 After deleting observations with missing daily returns
during the event window (day 2-60 after the earnings
announcement date) 306,652 4,957
After deleting 1% top and bottom observations of
3.2 Measurement of variables
Diversification
Diversification, in this study, is defined as the extent to which firms are active in multiple operating industries. Following Bushman et al. (2004) and Barinov et al. (2014), I use three independent variables as proxies for measuring diversification in my tests. My first measure of complexity,
NSEG, represents the number of divisions with different two-digit SIC-codes in which a firm is
active, as reported in Compustat. The second measure, CONGLO, indicates whether a firm is a conglomerate or not, where the firm is deemed to be a conglomerate if it has business divisions in two or more different industries, according to Compustat segment files. Industries are defined here again using two-digit SIC codes. The variable CONGLO is closely related to NSEG and moreover follows from it. As regards its value, CONGLO equals 1 if the firm is a conglomerate and 0 otherwise. The third measure, CONC, is a continuous variable based on the degree of sales concentration of a firm and is calculated by using the Herfindahl-Hirschman index (HHI), better known as the Herfindahl index, which is a statistical measure of concentration. HHI accounts for the number of firms in a market, as well as concentration, by incorporating the relative size (market share) of all firms in a market. It is calculated by squaring the market shares of all firms in a market and then summing the squares. CONC equals 1-HHI, where HHI here is the sum of sales shares of each division squared. Therefore, my third measure for diversification is: CONC = 1 - HHI, where HHI = Si=1 si2
,
and si represents the sales share for each division, being the fraction oftotal sales generated by that division. According to this third measure of diversification, a firm with sales in a single segment would have a complexity measure of 0, whereas a firm with sales in a large number of industries could achieve a complexity score close to 1.
Accounting Complexity
In this study, accounting complexity is defined as the difficulty in understanding the mapping of (potential) transactions and standards into financial statements. I argue that this accounting complexity is not homogenous across industries, but rather a continuum whereby some industries can, on average be considered more complex than other industries. Along this continuum are industries that have relatively few unique accounting and reporting concerns and other industries that have increasingly complex accounting. My measure for accounting complexity, ACC_COMP, is a firm-level complexity ranking, which is based on the Composite Industry-level Measure of Accounting
Complexity from Firm Characteristics, as developed by Seavey in 2011 in his framework. (see Appendix
rank is calculated. A higher ranking indicates higher accounting complexity and the ranking ranges from 15.7 (SIC 54 - Food Stores) to 45.4 (SIC 28 - Chemical and Allied Products). The full ranking can be found in Appendix C. For every firm in my sample, I calculate the average complexity of the industries a firm is active in, whereby the complexity of each individual industry is weighted by the proportion of sales of the firm within that particular industry. Again, also here industries are defined by their two-digit SIC-code.
Post earnings announcement drift
Post earnings announcement drift is the phenomenon that abnormal stock returns continue to drift significantly in the direction of an earnings surprise, starting on the first day after an earnings announcement and continuing through to the sixtieth day after an earnings announcement. In order to detect long-run abnormal stock returns and following prior research (e.g. Kang et al, 2017), I measure post earnings announcement drift using the Cumulative Abnormal Return (CAR) of a stock. My use of CAR as a return metric is consistent with, among others, Bernard and Thomas (1989, 1990). CAR is calculated by cumulating the difference between the raw daily return and the value-weighted daily return for a security over the period two days after the earnings announcement date until the sixtieth day, as available from CRSP. Following prior research (e.g. Ball and Brown, 1968; Bernard and Thomas, 1989; Nichols and Wahlen, 2004), I use the 2-60-day period following a quarters earnings announcement date as the event period to measure post earnings announcement drift. The result of this method is a variable which I label CAR_2_60 in this study. Therefore, my proxy for post earnings announcement drift, CAR_2_60, is measured as follows: CAR_2_60 = S t=2,60 (RETit - VWRETDit), where RETit is the daily return for firm i on
day t, inclusive of dividends and other distributions according to CRSP. VWRETDit, following prior research (e.g. Rangan and Sloan, 1998; Narayanamoorthy, 2006; Cao, 2012), represents the daily return, including all distributions, on a value-weighted market portfolio expected return for firm i on day t, according to CRSP. S t=2,60 cumulates the daily differences between RET and VWRETD for the period from the second day until the sixtieth day after an earnings
announcement.
Control variables
Prior research has shown that post earnings announcement drift persists strongest where arbitrage costs are highest, that is, among small NYSE/AMEX firms, and among firms with little or no analyst following or with low stock prices (Johnson and Schwartz, 2000). Also, Bhushan (1993) and Brav and Heaton (2006) documented that the drift is stronger for smaller, low-priced
stocks. Chudek et al. (2011), when examining the firm characteristics of extreme earnings surprise portfolios, find that extreme earnings surprise stocks in Canada often have low market value. Hong et al. (2000), Bhushan (1989) and Christensen (2004) find that firm size and analyst following are strongly correlated. Besides proxies as total assets and market capitalization, a firm's total sales (Dang and Li, 2015) has also been used extensively in prior academic research as a proxy for firm size. Based on the above, it may be argued that analyst following, total assets, market capitalization and sales are all proxies for measuring a similar construct, namely firm size. Therefore, and for the sake of brevity, in this study I will use sales as a single proxy for firm size. More specifically, to avoid skewness, I will use the natural logarithm of the total sales of a firm as a control variable. Therefore, my measure for sales is: logSALESit = log (total sales of firm i in year
t), where SALES is measured in millions of dollars.
Because post earnings announcement drift is by nature a manifestation of volatility and since prior research (Francis et al, 2007; Mendenhall, 2004; Feldman et al., 2010) establishes that market volatility significantly impacts on daily returns, I will also control for market volatility in this study. As a proxy for market volatility, I will use the value of the CBOE Volatility Index (VIX), which is a popular measure of the stock market's expectation of volatility implied by S&P 500 index options, as calculated and published by the Chicago Board Options Exchange (CBOE). As such, the VIX may be seen as a forward-looking measure for expected volatility that is likely to be incorporated in the decision making of investors when assessing earnings information. In this sense, market volatility may be regarded as an apt control variable for the purpose of this study. To sum up, my measure for volatility is: VOLATt = VIXt , where VIX is the monthly average
value for the VIX in month t.
3.3 Empirical models
In order to establish whether complexity influences the extent of post earnings announcement drift, it is necessary to divide my sample into two subsamples, namely a sample of firms with only positive CAR's and a sample of firms with only negative CAR's. This may be regarded as obvious, since adding the two together will destroy the visibility of the potential effects that complexity may have on the magnitude of post earnings announcement drift in both directions. This study is therefore concerned with the magnitude of CAR's in both upward and downward direction separately and not interested in the sum of CAR's for the market as a whole.
estimate the following model for both the positive as well as the negative cumulative abnormal return sample (firm subscript is omitted):
CAR_2_60t = b0 + b1NSEGt + b2CONGLOt + b3CONCt +b4SALESt + b5VOLATt + e (1)
where, CAR_2_60t is the cumulative abnormal return from the second day until the sixtieth day
following an earnings announcement made at date t, inclusive of dividends and other distributions.
NSEGt represents the number of divisions with different two-digit SIC-codes in which a firm is
active, at the time of an earnings announcement made at date t. CONGLOt is a dummy variable,
indicating whether a firm is a conglomerate or not, at the time of an earnings announcement made at date t. The variable CONGLO equals 1 if the firm is a conglomerate and 0 otherwise. CONCt is
a continuous variable based on sales concentration and is calculated by using the Herfindahl-Hirschman index (HHI), which is a statistical measure of concentration. Here, CONC, equals 1-HHI, where HHI is the sum of sales shares of each division squared, at the time of an earnings announcement made at date t. According to this measurement, a firm with sales in a single segment would have a CONC value of 0, whereas a firm with sales in a large number of industries could achieve a CONC score close to 1. logSALESt is a control variable for measuring firm-size through
the natural logarithm of total annual sales in millions of dollars, at the time of an earnings announcement made at date t. VOLATt serves as a proxy for market volatility, measured as the
monthly value of the CBOE Volatility Index (VIX), at the time of an earnings announcement made at date t. VIX is a common measure of the stock market's expectation of volatility implied by S&P 500 index options, calculated and published by the Chicago Board Options Exchange (CBOE). The coefficient b1 measures the impact on drift resulting from an increase in
diversification by 1 operating segment and coefficient b2 measures the impact on drift depending
on whether a firm is a conglomerate, or not. Further, coefficient b3 measures the impact on drift
resulting from sales concentration across industries (diversification of sales) and coefficient b4
measures the impact of sales as a proxy for firm-size on drift (control variable). Finally, coefficient b5 measures the impact of market volatility on drift (control variable)
Hypothesis 1 predicts that an increase in diversification is associated with a higher extent of both positive and negative post earnings announcement drift. Thus, the coefficients b1, b2 and
b3 are expected to be a) positive in the positive CAR-subsample and b) negative in the negative CAR-subsample, since I expect that diversification increases drift both in an upward direction
surprises. I control here for firm size and volatility, since prior research has shown that firm size has a mitigating effect on post earnings announcement drift and volatility an exacerbating effect on drift. Therefore, I expect the coefficient b4 to be negative in the positive CAR-subsample and
positive in the negative CAR-subsample. Contrarily, I expect coefficient b5 to be positive in the
positive CAR-subsample and negative in the negative CAR-subsample.
To test my second hypothesis: Accounting complexity is positively related to post earnings
announcement drift, I estimate the following model (firm subscript is omitted):
CAR_2_60t = b0 + b1ACC_COMP +b2SALESt + b3VOLATt + e (2)
where, CAR_2_60t cumulative abnormal return from the second day until the sixtieth day
following an earnings announcement made at date t, inclusive of dividends and other distributions.
ACC_COMPt is the composite industry-level measure of accounting complexity from firm
characteristics, at the time of an earnings announcement made at date t (see Appendix C). For every firm in my sample, I calculate the average complexity of the industries a firm is active in, whereby the complexity of each individual industry is weighed by the proportion of sales within that industry. Industries are defined by their two-digit SIC-code. logSALESt and VOLATt are
control variables and are as described earlier with regard to H1. The coefficient b1 measures the
impact on drift resulting from accounting complexity and coefficient b2 measures the impact of
sales as a proxy for firm-size on drift (control variable). Finally, coefficient b3 measures the impact
of market volatility on drift (control variable)
Hypothesis 2 predicts that an increase in accounting complexity is associated with a higher extent of both positive and negative post earnings drift. Consequently, the coefficient b1, which measures the impact of accounting complexity on drift, is expected to be a) positive in the positive
CAR-subsample and b) negative in the negative CAR-subsample, since I conjecture that
accounting complexity increases drift both in an upward direction following positive earnings surprises and in a downward direction following negative earnings surprises. I control here for firm size and volatility, since prior research has shown that these two are correlated with post earnings announcement drift. As with hypothesis 1, I expect the coefficient b2, which measures
the impact of firm size, to be negative in the positive CAR-subsample and positive in the negative CAR-subsample. On the other hand, I expect coefficient b3, which measures the effect of market
volatility, to be positive in the positive subsample and negative in the negative CAR-subsample.
To test my third hypothesis: Issuance of SFAS 131 reduces post earnings announcement drift, I estimate the following model (firm subscript is omitted) for both the period of three years prior to enforcement of SFAS 131 (1995-1997) and also the period of three years following the enforcement of SFAS 131 (1998-2000):
CAR_2_60t = b0 + b1NSEGt +b2SALESt + b3VOLATt + e (3)
where CAR_2_60t, NSEGt, and VOLATt and logSALESt are as described earlier. The coefficient
b1 here measures the impact on drift resulting from an increase in diversification by 1 operating
segment. Coefficient b2 measures the impact of sales as a proxy for firm-size on drift (control
variable) and coefficient b3 measures the impact of market volatility on drift (control variable)
Hypothesis 3 predicts that the issuance of SFAS 131 is associated with increased transparency for investors regarding the diversification level of a firm and consequently is associated with a decrease of the extent of both positive and negative post earnings drift. Thus, the coefficient b1, which measures the impact of diversification across segment on drift, is expected
to have a value closer to zero in the Post-SFAS131 period in both the positive CAR-subsample as well as in the negative CAR-subsample. I control for firm size and volatility, since prior research has shown that these two are correlated with post earnings announcement drift. As with H1 and H2, I expect the coefficient b2 ,which measures the effect of firm size, to be negative in the
positive CAR-subsample and positive in the negative CAR-subsample. As regards the control variable for measuring the effect of market volatility, I expect coefficient b3 to be positive in the
4.
Empirical results
4.1 Descriptive statistics
The sample of this study includes 300,519 firm-quarter observations for NYSE and NASDAQ firms with segment and stock return data on Compustat and CRSP for the period 1990-2017. Table 2 presents descriptive statistics of the selected variables by fiscal year. N represents the number of firm-quarter observations per fiscal year. NSEG, representing the number of divisions with different two-digit SIC-codes in which a firm is active, ranges between 1.27 and 1.47. A sharp increase of NSEG can be noted in the years following the implementation of SFAS 131 in December 1997: from 1.27 in 1997 to approx. 1.40 on average in the following three years. The mean value of the dummy variable CONGLO, indicating whether a firm is a conglomerate or not, is also experiencing a strong increase from 0.194 in 1997 to 0.254 in 1998 after implementation of SFAS 131, in line with the aforementioned increase of NSEG. Regarding my statistical measure of concentration, I employ a continuous variable based on sales concentration, which is calculated by using the Herfindahl-Hirschman index (HHI), where HHI is the sum of sales shares of each division squared. This proxy for concentration CONC, equals 1-HHI. According to this measurement, a firm with sales in a single segment would have a complexity measure of 0, whereas a firm with sales in a large number of industries could achieve a complexity score close to 1. From Table 2 we can see that the mean value of CONC, also experiences a very strong increase from 0.0729 in 1997 to 0.0915 in the year 1998, following the increased segment reporting demanded by SFAS 131, and consistent with previous statistics for NSEG and CONGLO. Concerning my measure for accounting complexity, ACC_COMP, the composite industry-level ranking of accounting complexity derived from ten different firm characteristics, it can be noticed that the average value for all firms is not very volatile throughout the sample period, but instead fluctuates only slightly between 36.3 and 37.4. As for my proxy for firm size, logSALES, we see a consistent increase throughout the years. With regard to the proxy for market volatility, VOLAT, we can see that the average yearly value ranges between 11.306 and 32.993, which may be labelled as a considerably volatile pattern.
