Gender, accounting conservatism and financial analysts’ forecast accuracy

(1)

Gender, Accounting Conservatism and

Financial Analysts’ Forecast Accuracy

Master Thesis

Name: Emily Mao Student number:11364262 Date: 25th_{of June 2018} Word count: 16,468 Supervisor: W.H.J. Janssen

MSc Accountancy & Control, variant Accountancy Amsterdam Business School

(2)

Statement of Originality

This document is written by student Emily Mao 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.

(3)

Abstract

This study investigates whether an analysts issues more accurate earnings forecasts when he or she has a better understanding of accounting conservatism in general. Additionally, this study investigates whether female analysts moderate this expectation. Prior research that investigated the effect of accounting conservatism on analysts’ forecast accuracy mostly examined whether analysts incorporate accounting conservatism in their forecasts on average. This research on the other hand, will investigate this association by taking each individual analyst into account. Furthermore, the conclusions of past research are contradictory, and a few of these studies have received criticism regarding their research methods. This study was conducted over a sample of 24,898 observations that consists of 700 firm-years and 875 unique analysts. This research provides empirical evidence that when an analyst has a better understanding of accounting conservatism in general, that this results in more accurate earnings forecasts. Providing more accurate forecasts helps to mitigate agency problems and reduce information asymmetry. There was no sufficient statistical evidence found that the gender of an analyst moderates the association between an analyst’s understanding of accounting conservatism in general and his or her earnings forecast accuracy.

(4)

1. Introduction

During the past few decades, accounting conservatism has become a popular

accounting practice, and it has remained relevant despite the ongoing discussion on whether accounting conservatism does more harm than good (Watts, 2003). There have been

numerous studies that have investigated the advantages of accounting conservatism. Ahmed and Duellman (2007) for example, found that accounting conservatism helps the board of directors reduce agency costs within a firm. Furthermore, aside from assisting directors in reducing agency costs, it proves its importance in mitigating bondholder and shareholder disagreements regarding dividend policies and the firms’ debt costs (Ahmed et al., 2002). Prudence was implemented as a principal in IASB’s conceptual framework. This principle was also known as the conservatism principle. However, it was removed from its framework in 2010 as IASB mentioned in a discussion paper that prudence (conservatism) was

considered undesirable to counteract uncertainty because it would be in conflict with neutrality. Nevertheless, many stakeholders, the UK Shareholders’ Association for example (2012), raised their concerns about removing prudence (conservatism) from the conceptual framework. As a result, IASB reintroduced this in 2015 omitting the conservative basis (Măciucă et al., 2015).

This study investigates whether a better understanding of accounting conservatism leads to an improvement in the financial analysts’ forecast accuracy, and if the gender of an analyst moderates this relationship. More specifically, I examine whether analysts issue more accurate earnings forecasts when they have a better understanding of accounting

conservatism in general. This means that an analyst recognizes conservative accounting practices of a firm and is able to process and act upon this when forecasting a firm’s earnings. This is measured by matching the general level of accounting conservatism of an analyst’s issued forecasts in this study’s sample with that of each firm that he or she followed in the sample. Consequently, the following research question is central throughout this study:

Does a better understanding of accounting conservatism by financial analysts lead to an improvement in the financial analysts’ forecast accuracy?

Furthermore, it is expected that female analysts are more likely to have a better understanding of accounting conservatism as women are generally more risk-averse and less overconfident than men (Jianakoplos and Bernasek, 1998; Niederle and Vesterlund, 2007;

(6)

2 Beyer and Bowden, 1997). According to prior literature, this also explains their findings that female analysts issue on average more accurate forecasts than their male counterparts

(Kumar, 2010). Moreover, Green et al. (2007) found that women provide less optimistic forecasts than men.

The reason why this research should be considered important is because financial analysts’ earnings forecasts are an important element of market efficiency, the external information environment, and the valuation of a company (Sohn, 2012). Aside from that, regulators and other market participants see the activities of analysts as increasing the

information efficiency of equity prices and/or returns as a result of their skills and knowledge in firm valuation (Frankel et al., 2006). Frankel et al. (2006) also claim that forecasts are being used to identify systematic and aggregate errors that lead to increased agency costs and declined informational efficiency within capital markets.

Beaver and Ryan (2005) make a distinction between conditional conservatism and unconditional conservatism. Unconditional conservatism is defined as an accounting bias toward recognizing the book values of net assets that are below the expected market values during their lives at inception; whereas conditional conservatism is defined as the

accountant’s tendency to demand stricter requirements when recognizing earnings than when a company recognizes loss. In other words, firms recognize losses in a timelier manner than earnings (Basu, 1997). Since there are various market participants who value sell-side analysts’ forecasts, it is important that these be accurate by using all of the available public and private information, therefore giving analysts an incentive to decode and incorporate various levels of conditional conservatism in their earnings forecasts (Jung et al., 2017).

LaFond and Watts (2008) argue that accounting conservatism leads to a reduction in information asymmetry by increasing the speed of recognizing expected losses compared to earnings. Additionally, Li (2008) suggests that accounting conservatism expands the

predictability of earnings by anticipating all losses, thereby reducing uncertainty for stakeholders. Furthermore, Li (2008) found a negative relation between accounting conservatism and absolute financial analysts forecast errors, suggesting that conservatism increases analysts’ forecast accuracy. Previous studies (Sohn, 2012; Jung et al., 2017; Mensah et al., 2004; Heflin et al., 2014) have investigated whether financial analysts incorporate accounting conservatism in their forecasts. However, these studies have

(7)

3 these studies. Therefore, I am going to counterpart this criticism by conducting an empirical investigation that provides new insights into the effects of accounting conservatism on financial analysts’ forecast accuracy.

This study’s methodology is based on Jung et al.’s (2017) research method. However, due to this study’s sample size, a few requirements that Jung et al. (2017) applied have been adjusted to make their research method more suitable for this study. The adjustment that has led to the biggest difference between these two studies is the measure of the analysts’ asymmetric forecast timeliness. Jung et al. (2017) measured this at analyst-firm-year level. This research on the other hand measures this at analyst level. This means that the

asymmetric forecast timeliness will be measured per individual analyst, without taking each firm and fiscal year the analyst follows into account. Thus, this research measures an analyst’s understanding of accounting conservatism in general.

The findings of this study are consistent with the expectations as this research concludes that the accuracy of analysts’ earnings forecasts improve when they have a better understanding of conservatism. This is calculated by the absolute difference between the overall level of asymmetric forecast timeliness of an analyst with each firm’s level of asymmetric timeliness that the analyst has followed in the sample. The conclusion of this study is consistent with prior research (Jung et al., 2017; Sohn, 2012; Hugon and Muslu, 2010) that concluded that accounting conservatism has a positive influence on analysts’ forecast accuracy. Thus, the results of this study reject prior studies (Heflin et al., 2014; Mensah et al., 2004) that concluded that conditional conservatism does not lead to more accurate analysts’ forecasts. The studies that rejected the positive influence of accounting conservatism mentioned that accounting conservatism leads to an increase in forecasts errors and a decrease in persistence and informativeness. Bushee et al. (2010) mentioned that

financial analysts act as intermediaries in the equity market where a firm’s management is the agent and the debtholders and shareholders the principals. I find that when the understanding of accounting conservatism of analysts is high, this results in more accurate forecasts for those analysts. By issuing accurate forecasts, financial analysts provide useful information for both debtholders and shareholders. Therefore, my findings support Ahmed et al. (2002) who concluded that accounting conservatism helps mitigate agency problems and reduce

information asymmetry. There is not sufficient statistical evidence found in this research to support the expectation that female analysts have a better understanding of conservative

(8)

4 accounting practices than male analysts, and thus issue more accurate earnings forecasts (hypothesis 2).

This study contributes to the existing literature by applying a relative new research methodology by Jung et al. (2017). Their research methodology distinguishes itself from previous studies as other studies’ research methods (Helbok and Walter, 2004; Sohn, 2012; Mensah et al., 2004; Pae and Thornton, 2010) only examined to what extent analysts apply conditional conservatism, and if this leads to more accurate earnings forecasts on average. Unlike prior research, this study takes the differences between individual analysts into account in its research methodology. García Lara et al. (2014), for example, only focused on whether firms that use conservative accounting have lower information asymmetry than those who do not use conservative accounting. Moreover, no prior research investigated whether the gender of an analyst moderates the relationship between the understanding of accounting conservatism and the analysts’ forecast accuracy.

The remainder of this study consists of the literature review, the hypothesis

development based on the literature review. Furthermore, the research methodology that is applied in this study will be explained together with the sample size in chapter four. The results are provided in chapter five which also gives an overview of the final sample. Lastly, the final chapter contains the discussion and conclusion.

(9)

5

2. Literature review and theory

2.1 Conservatism

2.1.1 Definition of conservatism

Conservatism has globally become a very important principle for the financial accounting standards because of its various benefits and effects. One important reason to support accounting conservatism is that it supports the information needs of creditors. Not only in German accounting it is recognized as a central accounting principle, according to Haller (2003), but also in the United States accounting conservatism has developed into a highly influential accounting principle (Sterling, 1967). The increased use of conservatism can also be found in the accounting standards by the Financial Accounting Standards Board (hereafter; FASB), which required the expense of R&D costs, recognize pension costs, and recognition when impairing assets but forbids an ascending revaluation of assets.

Watts (2003) defines conservatism as follows: “Conservatism is defined as the differential verifiability required for recognition of profits versus losses.” Its extreme form is the traditional conservatism adage: “Anticipate no profit but anticipate all losses.” Basu’s (1997) definition of conservatism is the accountant’s tendency to require a higher degree of verification for recognizing good news in earnings as opposed to bad news; this is also referred to as asymmetric timeliness of earnings. This means that bad news will be recognized in a timely matter, whereas good news is recognized gradually which is not timely. Even though Basu’s (1997) measure of asymmetric earnings timeliness is widely used ever since it was introduced, there is some criticism as this only captures one source, and there are also limitations as an asymmetric timeliness measure itself (Givoly et al., 2007). More specifically, the piecewise linear regression by Basu (1997) presumes that the firms that are in the same industry are identical since it does not take firm-specific measurements into consideration.

Using Beaver and Ryan’s (2005) distinction between conditional and unconditional conservatism, the measure of Basu (1997) would be defined as conditional conservatism. Conditional conservatism is defined by Beaver and Ryan (2005) as the writing down of the book value of assets under unfavorable circumstances. Beaver and Ryan (2005) define unconditional conservatism as the understatement of the book value of a firm’s assets due to the preset aspects of its accounting procedure. Besides Beaver and Ryan’s (2005) distinction of conservatism, Richardson and Tinaikar (2004) distinguish ex ante and ex poste

(10)

6 conservatism, where ex ante is related to unconditional conservatism, and ex poste to

conditional conservatism. An example of ex ante, or unconditional conservatism, is the immediate expenditures for R&D and the depreciation of property, plant and equipment which leads to a higher Market-To-Book ratio (Roychowdhury and Watts, 2007). Another difference between unconditional conservatism and conditional conservatism is that

unconditional conservatism is based on rules, whereas conditional conservatism is based on discretion, which also explains why there is some criticism about accounting conservatism. Do note that conditional and unconditional conservatism are not mutually exclusive. As explained by Pope and Walker (2003) and Pae et al. (2005), they found that there is an association between an increase of unconditional conservatism measured by the Market-to-Book ratio and decrease of lower conditional conservatism that was measured by the asymmetric earnings-return response.

2.1.2 Effects of conservatism

Accounting conservatism has several effects on different factors. One of the reasons why conservative accounting was introduced is because the earnings’ response to negative returns is stronger than positive returns (Basu, 1997). Furthermore, managers may have clashing incentives with their stakeholders, which stems from career and reputation concerns, earnings- or equity-based compensations contracts, and the value creation of its organization (Ball, 2009; Core, et al., 2003). Because of the above-mentioned incentives, it was found that managers therefore have the incentives to withhold bad news, which will be stockpiled for a longer period and then released all at once causing a stock price crash. However, Kim and Zhang (2016) found that conditional conservatism is negatively associated with the likelihood of future stock price crashes since accounting conservatism requires a firm to recognize loss in a timely manner. Because of conservatism’s timely loss recognition characteristic,

managers are not able to withhold bad news, since they need to anticipate all losses, and therefore inform their stakeholders beforehand.

García Lara et al. (2015) investigated the association between accounting conservatism and firm investment efficiency. They concluded that conservative firms’ investments and debt issuance increased, leading to a reduction in underinvestment and overinvestment because of the reduction of information asymmetry. Furthermore, Biddle et al. (2010) did a study on US listed firms and concluded that conservative accounting leads to a reduction in the bankruptcy risk by reducing the information asymmetry between the firm (borrower) and debtholders, additionally through the creation of cushions during bad times.

(11)

7

2.1.3 Criticism regarding conservatism

Despite the above mentioned positive effects of conservatism on various issues, accounting conservatism has received criticism from several parties. In the discussion paper of the FASB and the IASB regarding its improved conceptual framework that was released in 2006, they wrote that prudence (conservatism) was considered to be undesirable (IASB, 2006). FASB and IASB explained that it was considered undesirable because prudence (conservatism) leads to a bias of consistent understatement of the net assets and profits, which is an inadequate solution when dealing with uncertainty according to them. As a result, it was removed from the conceptual framework in 2010 as a qualitative characteristic of accounting information. However, various stakeholders were concerned about the removal of prudence (conservatism) for the Conceptual Framework, the UK Shareholders’ Association (2012) for example who wrote papers about this matter. The concerns of various stakeholders eventually led to a reintroduction of the concept of prudence but omitting the conservative basis, as they concluded that this clashes with neutrality (Măciucă et al., 2015).

Furthermore, Penman and Zhang (2002) also raised their doubts about the advantages of accounting conservatism. Their study argues that when a firm uses accounting

conservatism this can affect the quality of its earnings by its changes in the amount of their investments. They also explain that reduction in reported earnings occur as a result of an increase in investments which eventually results in the creation of reserves. When the investments are decreased by the firm, this releases the created reserves, which increases the earnings. Furthermore, when the modifications in the investments are temporary, then the current earnings will become temporary as well. Consequently, this decreases the

predictability of a firm’s future earnings. By analyzing the forecast stock returns, they found that investors disregard the changes in investments, thus possibly the earnings as well because of conservative accounting. Therefore, Penman and Zhang (2002) found that

applying accounting conservatism might raise questions about the quality of a firm’s reported earnings. However, to prevent the creation of such hidden reserves, the FASB and the

International Accounting Standards Committee (IASC) explicitly mentioned that prudence (conservatism) in the conceptual framework does not allow this creation through deliberately undervaluing assets or income or exaggerating the liabilities of a firm (Mora and Walker, 2015).

(12)

8 2.2 Financial analysts’ forecast accuracy

Financial analysts have a very important role in the financial markets as expert financial intermediaries, therefore it is not unusual to identify their earnings forecasts as useful to investors (Covrig and Low, 2005). LaFond and Roychowhury (2008) mention that these earnings forecasts, also called street earnings, that are provided by analysts, help

investors in predicting future recurring earnings. Frankel et al. (2006) argue that the activities of analysts play a crucial role for regulators and other market participants regarding the increase of information efficiency of equity prices and returns resulting from their knowledge and skills in firm valuation as these forecasts can be used to identify systematic and aggregate errors. More specifically, the increase of information efficiency mentioned by Frankel et al. (2006) is explained in greater detail by Liu and Thomas (2000). They concluded that the expected abnormal earnings, which are acquired from analysts’ forecasts, deliver the contribution of book value in clarifying the abnormal returns.

An analyst’s forecast is considered accurate when it contains few forecast errors between the actual Earnings per Share (hereafter; EPS) and the analyst’s forecasted EPS. Financial analysts are known to be bias and overreact when there is positive information, but they underreact to prior negative information (Easterwood and Nutt, 1999). Furthermore, Gu and Wu (2003) argue that the forecasts of analysts seem to be biased because of the

difference between the average and median earnings, and that analysts attempt to forecast (close to) the median earnings in order reduce the mean absolute error. Additionally, according to Francis and Philbrick (1993), analysts’ earnings forecasts are also driven by their relationship with the management. As a less favorable stock recommendation, which is in this case provided by a different analyst than the one who issues the forecasts, creates an incentive for the analyst who did not provided the stock recommendation, to provide some advantageous performance information to maintain his or her relationship with the

management.

Prior studies have investigated determinants of forecast accuracy as according to Clement (1999) it is of importance that accounting researchers understand these determinants when using financial analysts’ forecasts as a proxy for the capital markets’ estimates of earnings. By using cross-sectional regression, Clement (1999) concluded that analysts’ experience (proxy for ability) and employer size (proxy for resources) is positively related to analysts’ forecast accuracy. However, there is a negative correlation with the number of firms

(13)

9 and industries an analyst follows. These results are also supported by Jacob et al. (1999) who indicated that the aptitude and the brokerage house characteristics of an analyst influences the forecast accuracy. Besides the characteristics of an analyst, including his or her experience and employer size as described by Clement (1999), it was also concluded by Brown (2001) that past forecast accuracy is associated with the future forecast accuracy of an analyst. 2.3 Gender differences in an economic setting

Prior research has done various examinations on the behavior differences between men and women in an economic setting where they concluded that women are more risk-averse than men. The effect of this characteristic of women has been the explanation for various differences between male and female. Jianakoplos and Bernasek (1998) for example argue that women being risk-averse leads to them having a less risky asset portfolio

compared to men. As in their research, on average, women hold an equal amount of stocks and bonds, whereas men hold double the amount of stocks compared to bonds. Jianakoplos and Bernasek (1998) argue that being risk-averse also may be an explanation for why men have a higher level of wealth compared to women. Niederle and Vesterlund (2007) and Eckel and Grossman (2002) conducted laboratory experiments and concluded that women tend to be more risk-averse in their decision making, and are consequently less likely to be willing to participate in competitive environments. Niederle and Vesterlund (2007) examined their laboratory experiment through a real task where the participants had to solve mathematical problems and could choose between a piece rate compensation or a tournament. Seventy-three percent of the male participants chose the tournament, whereas this was only 35% for the female participants. Eckel and Grossman (2002) conducted their research by giving the participants the choice between five alternative gambles with substantial financial stakes. Psychology literature explained that women’s tendency to be more reluctant to competition than men is already being developed by way of how men are raised differently than women. Ruble et al. (2006) mention how parents, teachers, and peers encourage boys in activities that centralize being assertive, while girls are encouraged to be supportive of one another and emphatic.

Past research has a few explanations as to why women are more risk-averse than men. First, this can be explained by the fact that women have different emotional reactions to risk than men. Psychology research indicates that women experience emotions stronger than men (Harshman and Paivio, 1987). More specifically, when negative results are anticipated,

(14)

10 women report more extreme nervousness and fear than men (Fujita et al., 1991). So, it is natural for women to be more risk-averse in negative situations if their emotional reactions are stronger than men’s reaction to it. Furthermore, the emotions of the gender also play a role in the way they perceive probability. Grossman and Wood (1993) concluded that in negative situations, women are more likely to feel fear whereas men feel anger. In particular, Lerner et al. (2003), argue that when an individual is angry, he or she perceives a given gamble as less risky as opposed to when he or she is afraid.

Secondly, overconfidence is also a difference between the genders which may lead to different reactions of risk. Beyer and Bowden (1997) concluded that men tend to be more overconfident than women, and this association is more pronounced when applying it to masculine tasks. Consistent with this conclusion, Barber and Odean (2001) found that because of men’s overconfidence, they make more extreme trades than women in the financial markets. Additionally, Estes and Hosseini (1988) conducted their research by having the participants examine financial statements of a fictional company and advise how much to invest in the fictional company. Afterwards, the participants were asked to tell how confident they were of their investment decision. The conclusion drawn from their

investment decisions was that women were more uncertain of their decision than men. Thus, Estes and Hosseini (1988) provide evidence that because of men’s overconfidence, they perceive situations less likely to be negative and are therefore less risk-averse.

Furthermore, there are three more gender differences of preferences in an economic setting. The social preferences of men and women also differ from each other as women seem to be more sensitive to social cues when they determine appropriate behavior (Kahn et al., 1971). Various research has been done in different experimental settings to test the social preferences of women and men (Eckel and Grossman, 2002; Houser and Schunk, 2009; Croson and Buchan, 1999). Nevertheless, in general these studies concluded that women are more sensitive to the conditions of the experiment and that women’s decisions are more context-specific than men’s. They explained that women’s decisions were more sensitive to the conditions of the experiment, and the available information and actions of the other party (Gilligan, 1982). This behavior can also refer to the previously mentioned explanation why women are more risk-reverse since their emotional reactions are stronger than men’s.

Secondly, women are more ethical than men. Galbraith and Stephenson (1993)

(15)

11 as they also have different attitudes towards codes of ethics (Ibrahim and Angelidis, 2009). Women are found to be more ethical than men as they show a more ethical attitude and behavior in the workforce (Bernardi and Arnold, 1997). Women are more likely to speak up when they detect unethical behavior (Vermeir and Kenhove, 2008). According to Ho et al. (2015) an ethical work climate, which stems from an ethical attitude and behavior,

encourages integrity in financial reporting and discourages earnings management.

At last, another difference between men and women is their behavior in competitive situations as women are more reluctant compared to men (Vandegrift and Brown, 2005). Furthermore, men’s performance improves in competitive interactions unlike women’s (Gneezy et al., 2003). Babcock and Laschever (2009) found that women are less likely to involve themselves in competitive negotiations as opposed to men.

2.4 Agency Theory

A theory that is related to this research is the agency theory. The agency theory distinguishes the concept of an organization between the agent and the principal (Jensen and Meckling, 1976). The agent is hired by the principal to carry out a task on their behalf. The responsibility given to the agent by the principal also comes with some authority. When both parties are utility maximizers there is a high possibility that their interests and incentives are in conflict. The agent has an advantage in this relationship as the agent has access to both private and public information whereas the principal cannot fully verify whether the agent’s interests are in line with theirs leading to agency problems. This advantage from the agent can lead to engagement in opportunistic behavior. Bris and Cantale (2004) classify two views of the agency problem, an internal and external view. The external view is between

shareholders and regulators, while the internal view of the agency problem is between the management and shareholders. In this case, I focus on the internal view of the agency problem where the management is the agent and the debtholders and shareholders are the principals.

Prior literature (Guesnerie et al., 1989; Pauly, 1978; Zeckhauser, 1970) identifies two well-known frictions regarding this agency problem; moral hazard and adverse selection. Moral hazard results when the principal does not have full access to the organization, in particular the management and their decision making (motivations). Therefore, the principal cannot completely verify the actions of the agent, leading to opportunistic behavior and hidden actions which conflict with the interests of the principal. It can also be the case that

(16)

12 the agent does not put as much effort in their tasks as is expected from the principal; this is known as shirking. Adverse selection on the other hand concerns the information that the agent hides from the principal. This can occur during the hiring process of the agent where it was claimed to have certain capabilities and expertise, while in fact this may not be the case. Just like moral hazard, this is hard for the principal to verify.

Jensen and Meckling (1976) explain how the principal can reduce the agency problem by monitoring the agent and by having covenants to assure that the agent will not engage in damaging activities for the principal. Consequently, this agency-principal problem leads to a greater information asymmetry between these two parties. To reduce the information

asymmetry between the agent and principal, the principal could monitor the agent and collect information. However, if each shareholder and potential investor would directly gather information and directly monitor the agent, this would be very inefficient. Besides, not every shareholder has the skills to gather the relevant information. Financial analysts, therefore, act as intermediaries between the shareholders/debtholders (principal) and the agent

(management) to reduce this inefficiency (Bushee et al., 2010). Ahmed et al. (2002) found that accounting conservatism is a solution to mitigate agency problems as it helps to reduce moral hazard problems among a firm’s management, debtholders and shareholders.

Consequently, they found evidence that there is demand from a firm’s shareholders to use conservative accounting. Additionally, other studies (LaFond and Watts, 2008; Amiram et al., 2016) found that accounting conservatism also reduces information asymmetry. They explain that conservatism reduces the incentives and chances to manipulate earnings and thus reduce information asymmetry.

(17)

13

3. Hypothesis development

3.1 Conservatism and financial analysts’ forecast accuracy

There have been prior studies which investigated the effect of accounting conservatism on analysts’ forecast accuracy. As previously mentioned, managers are motivated by moral hazard incentives to seize their opportunities in the presence of

information asymmetry by withholding bad news and aggressively reporting earnings (Watts, 2003). Kothari et al. (2009) argue that this information asymmetry stems from hiding bad news and, therefore, accounting conservatism has an informative role. Supporting Kothari et al.’s (2009) conclusion, García Lara et al. (2013) suggest that conservatism leads to a

decrease in information asymmetry which leads to a better information environment for financial analysts, and therefore more accurate and less diffused forecasts.

Heflin et al. (2015) investigated the effect of conditional conservatism on street earnings (pro forma earnings) and GAAP earnings. They found that conditional conservatism leads to lower persistence and a decrease in the informativeness of GAAP earnings, and that this also makes it more difficult for analysts to forecast GAAP earnings. Mensah et al. (2004) believed and concluded that unconditional conservatism decreases the earnings predictability and accuracy because of the unpredictable characteristics of unconditional conservatism regarding the requirement of immediate expenditure expenses and the conditional write-downs. Both findings by Heflin et al. (2015) and Mensah et al. (2004) comply with Basu’s (1997) findings, who argues that with conservatism, bad news is less persistent because of its immediate expensing instead of taking it into account gradually. Nonetheless, Mensah et al. (2004) mention that unconditional conservatism has an indirect effect on the accuracy of the forecasts as special items and one-time recurring items are not taken into consideration when analysts produce their forecasts. Unconditional conservatism only affects the degree of

conditional conservatism applied by the firms. Therefore, Mensah et al.’s (2004) research and Sohn’s (2012) research become less relevant as they mainly focus on unconditional

conservatism.

Helbok and Walker (2004) and Sohn (2012) concluded that, on average, analysts are conscious about conservatism and that this is taken into account when they update their forecasts. However, Pae and Thorton (2010) concluded differently as they argue that analysts do not fully take the implications of conservatism into consideration, which thus results in inefficient forecasts. Nonetheless, there is some criticism regarding these studies as Helbok

(18)

14 and Walker (2004) did not control for opportunism (an upward bias to maintain a relationship with management). This is important because this would mean that even without

conservatism, there is still a potential difference in the relationship between forecast revisions and positive returns relative to the association between forecast revisions and negative

revisions. Another limitation to Helbok and Walker’s (2004) study is that its research method requires that analysts are able to anticipate future events which are not expected by the broader public. (Louis et al., 2008)

Furthermore, Pae and Thorton (2006) concluded that there is a negative association between forecast errors and Book-to-Market firms. However, according to Louis et al. (2008) these measures are not sufficient to make such a conclusion as the effect of conservatism is mainly related to the nature of accounting and the sum of assets that was recognized

beforehand. Nonetheless, as previously mentioned in paragraph 2.1.1, Pope and Walker (2003) explained that conditional and unconditional conservatism are not mutually exclusive because unconditional conservatism provides slack or prevents conditional conservatism. In this study, unconditional conservatism will not be taken into consideration. The reason for this is because in earnings forecasting, as previously explained, financial analysts do not take special items and one-time recurring charges into consideration when preparing their

forecasts (Mensah et al., 2004).

At last, Jung et al. (2017) mentions that no other study matched the level of

conservatism of an analyst’s forecasts with the asymmetric timeliness of the firms’ reported earnings. When an analyst is able to match their asymmetric forecast timeliness with the firm’s asymmetric earnings timeliness, Jung et al. (2017) perceives this as that the analyst has a good understanding of accounting conservatism. In other words, the better the

understanding an analyst has about accounting conservatism, the greater the analysts’ forecast accuracy. A good understanding of accounting conservatism means that an analyst recognizes conservative firms and is able to process and acts upon this. Prior research (Mensah et al., 2004; Sohn, 2012; Pae and Thornton, 2010) either studied the level of conservatism of analysts on average or to what extent the firm uses accounting conservatism. This study will apply Jung et al.’s (2017) research methodology. However, because this study is conducted over a relatively small period of five years, a few adjustments have been made to Jung et al.’s (2017) research methodology to make it more suitable for this study’s sample size. Jung et al. (2017) measured the analyst’s asymmetric forecast timeliness at analyst-firm-year level. Here, I measure an analyst’s asymmetric forecast timeliness in general at analyst level, thus

(19)

15 not taking each specific firm and fiscal year into account. Combining the conclusions

concerning the negative association between accounting conservatism and information asymmetry, and in response to the criticism regarding prior studies and its contradicting conclusions, the following hypothesis is developed:

Hypothesis 1: Financial analysts who have a better understanding of accounting conservatism in general, deliver more accurate forecasts.

The above hypothesis is formulated in an alternative form. It can also be reformulated in the null hypothesis that financial analysts who have a better understanding of accounting conservatism in general do not issue more accurate forecasts.

3.2 Analysts’ forecast accuracy, conservatism and gender

Additionally, I will also investigate whether the gender of an analyst moderates the relationship between an analyst’s understanding of accounting conservatism in general and the forecast accuracy.

According to the Bureau of Labor Statistics (2017), Thirty-six-point-five percent of the financial analysts in 2017 in the United States were women. Prior research examined whether the gender of analysts affects the forecast accuracy. Kumar (2010) found that women release bolder (bold vs. herding forecasting style) forecasts and, regardless of their

forecasting style, also more accurate forecasts. All-star analysts are the best in-field analysts of the United States, according to a survey which was completed by buy-side analysts (like institutional investors) and is conducted annually by the Wall Street Journal (Bagnoli et al., 2008). As expected, all-star analysts are the most accurate in the field, but the accuracy for female all-star analysts is even higher. Furthermore, Kumar (2010) investigated whether female analysts’ forecast abilities are significantly different from male analysts. He argued that women are even more accurate in market segments where the concentration of women is lower. This research also concluded that the stock market participants are partially aware of the difference in the forecasting skills between women and men since the market’s reaction is stronger to the forecasts revisions of female analysts in the long run, even though these female analysts have less media attention. However, the market processes the received information of all-star analysts faster when these analysts revise their forecasts. The evidence that women release more accurate forecasts supports Kumar’s (2010) hypothesis which says

(20)

16 that due to the self-selection process, only women who have above-average skills would choose the analyst profession as a result of discrimination in this particular labor market. Therefore, are female analysts on average more likely to have better abilities than men.

Unlike Kumar (2010), Green et al. (2009) found that women’s forecasts are slightly less accurate than men’s. They argue that the accuracy difference between men and women is not that big because female analysts are more likely to cover fewer stocks and thus, have more time to analyze each stock. However, another way to explain why women cover fewer stocks than men may also be explained by the fact that women carry a larger part of the family responsibilities. Nevertheless, I may also explain this finding of Green et al. (2009) by linking this with Kumar’s (2010) finding as he, on the other hand, explained that women cover relatively larger stocks with bigger institutional ownership. So women may cover fewer stocks, but if these are bigger stocks, this may weight against that men cover more but

smaller stocks. Another reason for the forecasts’ accuracy difference between men and women can be explained by the women’s employer, as women tend to work at top brokerage firms (Green et al., 2009). Supporting this conclusion, Kumar (2010) mentioned that female analysts’ careers are more likely to be promoted to high-status brokerages than male analysts. Additionally, the researchers (Green et al., 2009; Kumar, 2010) found that both genders have an optimism bias; however, this is stronger for men. A consequence of this is that women release less optimistic forecasts. They indicate that women are more likely to be chosen as an all-star analyst because of other non-quantifiable aspects like client service.

Another study conducted by Li et al. (2013) examined whether there is a performance difference between male and female sell-side analysts, and if this influences their career outcomes. They analyzed the performance difference by measuring the surplus of the return of investment recommendations and the level of risk taken by the analysts, which is measured by the residual risk of their portfolio implied by investment recommendations. Additionally, this research also analyzed the genders regarding their bias and career outcomes. This was measured by the percentage of sell recommendations, and by the turnover rate in between brokerage firms and the likelihood of being an all-star analyst. Li et al. (2013) concluded that the performance, translated in abnormal returns, of the recommendations between the female and male analysts were similar but differ in risk. They found that female analysts’

recommendations come with slightly lower unsystematic risks and they, therefore, take fewer risks in their recommendations.

(21)

17 On a more specific note, Francis et al. (2014) examined the association between CFO gender and corporate financial reporting decision making. Their research focused on firms who have changed from a male CFO to a female CFO. They found that female CFOs use more accounting conservatism in their financial reporting. According to Francis et al. (2014), this is explained by women’s stronger risk-averse nature, which is reflected in this study’s results as the association is more pronounced when the firm has a higher litigation risk, default risk, and systematic risk. Thus, this study provides evidence that gender affects the use of accounting conservatism. Moreover, Ho et al. (2015) studied the relationship between CEO gender and accounting conservatism. This study complements Francis et al. (2014) with similar results as they found significant evidence that female CEOs are more conservative in their accounting than male CEOs. More specifically, they found that female CEOs use more accounting conservatism than male CEOs because women are more risk-averse and ethical. Therefore, female CEOs are more likely to recognize bad news in a timelier manner. As mentioned in paragraph 2.3, women are found to be more risk-averse, less aggressive, less overconfident, more anxious, and more ethical. All these characteristics indicate a

conservative mindset and a lower likelihood to commit fraud (Powell and Ansic, 1997). While past studies investigated the effect of the gender on the forecast performance, and the effect of the gender of CFOs and CEOs on accounting conservatism, less is known about the effect of the gender of financial analysts on the understanding of accounting conservatism. More specifically, to date no study has investigated whether the gender of an analyst has an effect on to what extent the analyst understands accounting conservatism and, therefore issue more accurate forecasts. Concluding from related prior literature, I expect that female analysts are more likely to have a better understanding of accounting conservatism and thus provide more accurate forecasts than male analysts.

Hypothesis 2: The association of hypothesis 1 is more pronounced when the analyst is a woman.

(22)

18

4. Research method and design

For this research I will be applying Jung et al.’s (2017) research method. Despite this being a less popular research method because of its recency, this research method

differentiates itself from previous conditional conservatism research methods. Jung et al.’s (2017) research methodology distinguishes itself by taking the differences between individual analysts into consideration, whereas previous studies measured this on average. The research method exists in four parts. Firstly, measure the asymmetric earnings timeliness of the firm. Secondly, measure the asymmetric forecast timeliness of each individual analyst. Thirdly, match the analyst’s asymmetric forecast timeliness with the firm’s asymmetric earnings timeliness. Finally, when the independent variable (match) is estimated, I examine whether there is a significant correlation between the independent variable and the dependent variable. The expectation they are positively associated with each other. Additionally, I will analyze whether the gender of the analyst moderates the association between the independent and dependent variable. It is expected that female analysts have a better understanding of accounting conservatism than male analysts, and therefore issue more accurate forecasts. 4.1 Sample

To conduct this research, I will use data from 2012 to 2016 from the S&P 500

constituents lists. The reason for taking a smaller period than similar prior research is because of hypothesis 2, for which I have had to hand-collect information in order to determine the gender of the analysts. Furthermore, I decided not to include data from the 90’s as the market has developed a lot in the past 20 years; not only has there been a lot of changes in the laws and regulations, stakeholders have stricter requirements and higher expectations, and firms have changed a lot over the years as they have become more complex. For this research I derived data from IBES US detail database, CRSP, and Compustat, which are available through Wharton Research Data Services (WRDS). Moreover, to identify the gender of the analysts, information will be hand collected by retrieving the analyst codes, initials, and the last names of the analysts. Using the initials and the last name of each analyst, the first name of an analyst will be identified through investorwand.com, tipranks.com, or finance websites such as Bloomberg and The Wall Street Journal. If the first name of an analyst is unisex, I will search for finance articles who mention whether the analyst is a man or woman (by referring to he or she). Additionally, with the first and last name of an analyst I am also able to look up pictures of this analyst. To assure reliability of the collected data, a requirement

(23)

19 from Hugon and Muslu’s (2010) will be taken into account which mentioned that a stock price under $1 will not be used in the sample to avoid any consequences resulting from small denominators and small stocks.

4.2 Independent variable

Measuring the independent variable consists of three parts. The first part is to measure the firm’s asymmetric earnings timeliness. Secondly, the asymmetric timeliness of the

analyst’s forecast, and lastly, to determine to what degree these variables are aligned with each other.

4.2.1 Asymmetric earnings timeliness of the firm

In order to measure the firm-year asymmetric earnings timeliness of the firm, I will be using the following piecewise linear regression of earnings response on stock returns (Basu, 1997) to estimate the asymmetric earnings timeliness per firm:

𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆

𝐿𝑎𝑔𝑃𝑟𝑖𝑐𝑒 = β0 + β1RET

Annual_{+ β}

2D + β3RETAnnual*D + ε

(1A)

▪ Actual EPS represents the actual earnings per share retrieved from I/B/E/S.

▪ Lag of Actual EPS represents the actual EPS preceding the fiscal year.

▪ LagPrice is the beginning-of-fiscal-period stock price of the firm.

▪ RETAnnual_{stands for the buy-and-hold return over the fiscal year.}

▪ D represents a dummy variable which is 1 if RETAnnual_{is negative and 0 if not.}

β3 is the asymmetric timeliness coefficient, also referred as the incremental coefficient

on negative returns (FirmATCL_{). This conditional conservatism measure has been widely}

implemented by several studies, including Pae and Thornton (2003), Pope and Walker (1999) and Heflin et al. (2015). Additionally, the following regression model (equation 1B) is to measure the change in the actual EPS per firm:

(𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆−𝐿𝑎𝑔 𝑜𝑓 𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆)

𝐿𝑎𝑔𝑃𝑟𝑖𝑐𝑒 = β0 + β1RET

Annual_{+ β}

2D + β3RETAnnual*D + ε

(1B) ▪ Actual EPS represents the actual earnings per share retrieved from I/B/E/S.

▪ Lag of Actual EPS represents the actual EPS preceding the fiscal year.

▪ LagPrice is the beginning-of-fiscal-period stock price of the firm.

(24)

20

▪ D represents a dummy variable which is 1 if RETAnnual_{is negative and 0 if otherwise.}

β3 is the asymmetric earnings timeliness coefficient of a firm (FirmATCC), also

referred as the incremental coefficient on negative returns. In pursuance of the required data for Lag of Actual EPS, data will also be derived from 2011. To counteract the criticism of Basu’s (1997) measure that it does not take firm-specific measurements in its regression, I will apply control variables based on Mensah et al.’s (2004) research. The reason for not using Khan and Watts’ (2009) C-Score measure, which is based on Basu’s (1997) but further developed, is because the asymmetric earnings timeliness of the firm measure should be the same as the analysts’ asymmetric forecast timeliness measure. The reason why equation 1B is added to the research method is because a similar measure will be used when approximating the analysts’ forecast asymmetric timeliness (equation 2). Since these two measured will be matched with each other, it is important that asymmetric earnings timeliness and asymmetric forecast timeliness are measured analogously to mitigate potential errors.

Even though equation 1B is not as commonly used as equation 1A, it has been used in prior literature like Ball et al. (2013). It is expected that these two variations in the regression model to measure the firm’s asymmetric earnings timeliness coefficient will complement each other as the first regression model is time-series based while the second regression model looks at the unexpected earnings. For this measure (regression) I will be using I/B/E/S actual earnings and not GAAP earnings, since a clear majority of the analysts produce their forecasts based on earnings that exclude special items and one-time recurring costs like restructuring expenses (Bradshaw and Sloan, 2002).

4.2.2 Analysts’ asymmetric forecast timeliness

In order to measure the analysts’ asymmetric forecast timeliness, I will apply a measure which has been introduced by past studies, like Givoly and Lakonishik (1979) and Lys and Sohn (1990), that studied the information content of the earnings forecast revisions by analysts. Lys and Sohn (1990) concluded that analysts include public information that is reflected in the stock return to a certain extent. The measure is driven by the association between analyst forecast revisions and stock returns. Jung et al. (2017) took their model for further development and extended this by incorporating an indicator variable for negative stock returns and its interaction term with stock. However, this measure has been modified to make it more suitable for this study since I conduct the analysis over a period of five years. Instead of measuring the analysts’ asymmetric forecast timeliness on an analyst-firm-year

(25)

21 level, this will be measured per unique analyst without taking each firm he or she follows into account. This led to the following regression model used to measure the asymmetric forecast timeliness per unique analyst:

REV

{

= (𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝐸𝑃𝑆 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡−𝑃𝑟𝑒𝑐𝑒𝑑𝑖𝑛𝑔 𝐸𝑃𝑆 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡)

𝐿𝑎𝑔𝑃𝑟𝑖𝑐𝑒

}

=β0 + β1RET + β2D + β3RET*D + ε

(2) REV is the change in an analyst’s two sequent EPS forecasts for a firm. In other words, its

forecast revision.

▪ Current EPS forecast is an analyst’s EPS forecast of a firm’s earnings of the current fiscal year. In other words, this is an analyst’s most recent EPS forecast issued on a firm-year level.

▪ Preceding EPS forecast is an analyst’s EPS forecast of the same firm and fiscal year but precedes the Current EPS forecast. In other words, preceding EPS forecast is the second most recent EPS forecast issued by an analyst for the same firm and fiscal year. Whereas Current EPS forecast is the revision forecast for the same firm and the fiscal year that comes after the preceding EPS forecast.

▪ Lagprice is equal to the stock price of the firm derived from the end of the month when the analyst’s preceding EPS forecast was produced.

▪ RET is the buy-and-hold return of the stock over the period beginning on the preceding EPS forecast till the date of the current EPS forecast of the analyst (i.e., revision period).

▪ Additionally, D is a dummy variable that is equal to 1 if RET is negative and 0 if not. Jung et al. (2017).

β3 comprises the asymmetric forecast timeliness, as in the degree of recognizing good

and bad news, in an analyst’s forecast revision. This complies with the asymmetric timeliness coefficient by Basu (1997), which was also processed in equation 1B. Here, β3 is an analyst’s asymmetric forecast timeliness coefficient in general (AnalystATC). To gather the available

EPS forecasts revision that have been released by analysts, data from 2012 to 2016 will be used, and the regression will be examined at the analyst level.

4.2.3 Measure of timeliness match

When the level of a firm’s asymmetric earnings timeliness (FirmATCC(L))and an analyst’s asymmetric forecast timeliness is estimated (AnalystATC), I will be measuring to

(26)

22 what extent an analyst’s overall asymmetric forecast timeliness matches with each firm (FirmATCC(L)_{) that he or she followed in the sample, and therefore measure to what extent the}

analyst understands accounting conservatism. This will be measured as follows: MATCH(C)L(i,j)= -1 * (AnalystATC(i) – FirmATCC(L) (j))

(3)

In this equation, i and j stands for analyst and firm respectively. The L and C indicate whether I am using the firm’s asymmetric earnings timeliness based on the level of the actual EPS (Firm.ATCL_{) or the change in the actual EPS (Firm.ATC}C_{) when estimating the}

timeliness match. MATCH(i, j) computes the absolute difference between an analyst’s asymmetric forecast timeliness coefficient minus the firm’s asymmetric earnings timeliness.

MATCH (i, j) will be measured by the firm’s latest earnings following its announcement date (j). Jung et al. (2017)

4.3 Dependent variable; analysts’ forecast accuracy

To measure the effect of the analyst’s accounting conservatism understanding on the accuracy of the analyst’ forecast, the next OLS regression is estimated:

ACCURACY = a0 +a1MATCHC(L)+a2Firm.ATC C(L) + a3FSIZE + a4Horizon + a5AFollowing

+a6EarningsVol + a7FEMALE

(4A) Where:

ACCURACY is measured as -1 * (Actual EPS – Forecasted EPS) deflated by the stock price at the last trading day of the month when the analyst’s forecast was made.

MATCHL(C) is estimated by the -1 * the absolute difference between the firm’s asymmetric earnings timeliness (Analyst.ATC) and the analyst’s asymmetric forecast timeliness

(Firm.ATC C(L)).

Firm.ATCC(L)= (1A)

(1B) ▪ FSIZE = Total assets.

▪ Horizon = Actual EPS announcement date – Forecasted EPS announcement date (in absolute days).

(27)

23 ▪ EarningsVol= (𝐶𝑉_𝑋𝑖,(𝑡−1)−(𝑡−5)) standard deviation of the actual EPS before special

items over the preceding 5 years.

▪ FEMALE (dummy variable)= 1 if the analyst is female, 0 if otherwise.

To measure whether female analysts issue more accurate forecasts than male analysts by having a better understanding of accounting conservatism, the following OLS regression is estimated:

ACCURACY = a0 +a1MATCHC(L)*FEMALE+a2MATCHC(L)+ a3FEMALE+a4Firm.ATCC(L ) +

a5FSIZE + a6Horizon + a7AFollowing +a8EarningsVol

(4B) Equation 4B is almost identical to equation 4A. The differences between these two OLS regression is the independent variable. In equation 4B, MATCHC(L) and FEMALE are the main effects, but these two variables are also interacted with each other as an interaction independent variable. Consequently, by interacting FEMALE with MATCHC(L)_{, dummy}

variable FEMALE becomes a moderator in the regression.

The control variables used in equation 4A and 4B include analysts and firm characteristics to control for the bias in the independent variable.

FSIZE: Firm size is a commonly used control variable. Prior research like Das et al. (1998) mentioned that this is often used as a proxy for publicly available information. Bigger firms are usually, compared to relatively small firms, stricter monitored by the market and therefore have more information available about the firm. More publicly available

information indicates that it is easier to predict the future performance of the firm and a decrease in bias. Additionally, Bhushan (1989) concluded that firm size and analyst following are related to each other, as large firms have on average more analysts following than the relatively small firms.

Horizon: Forecast horizon, is the days between the actual EPS announced by the firm and when an analyst issued his or her forecast on the EPS. Prior research like Eames and Kim (2012) and Lys and Soo (1995), found that the greater the distance between the actual EPS and the forecasted EPS, the more likely forecast errors occur since the availability to relevant information islower. This means that forecasts that are issued closer to the actual EPS announcement date have a higher accuracy probability compared to forecasts that are issued in an earlier stage.

(28)

24 AFollowing: Lys and Soo (1995) concluded that forecast accuracy improves with the number of analyst following the firm. They argued that the number of analysts following is a proxy for the information environment of a firm. As a large number of analysts following a firm, suggests great information availability. Additionally, there is higher competition when the number of analysts following a firm increases. This results in analysts making greater efforts to issue more accurate forecasts and therefore, a positive association is expected between the number of analysts following a firm and its forecast accuracy.

EarningsVol: Numerous prior literature (Kross et al. 1990; Lim, 2001) argued that forecasts tend to be less accurate as analysts tend to be more optimistically biased when the earnings volatility of that firm is greater in the long term (here five years). The measure for earnings volatility is based on Mensah et al.’s (2004) research where it was measured by the coefficient of standard deviation of a firm’s earnings before special items and deflating this by its stock price over a period of five years. Thus, the expectation is that the more volatile the earnings of the firm were over the last five years, the less accurate the forecasts are.

FEMALE: As previously explained in the hypothesis development, the expectation is that women have a better understanding of conservatism since they are more risk-averse (Niederle and Vesterlund, 2007). Therefore, the expectation is that female analysts issue more accurate forecasts compared to male analysts.

(29)

25

5. Empirical results

5.1 Final sample

The sample consists of S&P 500 firms that were present in the S&P 500 constituents list between 2012 and 2016. This consisted of 393 firms, however, for calculation complexity purposes, only firms that had the 31st of December as their fiscal year end were taken into consideration. This led to a deduction to 281 firms who have the 31st of December as their fiscal year end date. For equation 1A, 1B and a few control variables, it is also required to retrieve information from the preceding fiscal year. This means that data from 2011 to 2016 has been retrieved from CRSP, Compustat and I/B/E/S. While collecting the data a few requirements have been met which led to a reduction in the final sample size.

The following requirements were met when collecting the data:

▪ The firm should at least have one negative annual return out of five years. In Jung et al. (2017) the requirement regarding the annual return of the firm is that out of the eight observations per firm, it should include at least two negative returns (and two positive returns). This requirement has been adjusted conform to this study’s analysis period of five years.

▪ The difference between the current and preceding forecast of an analyst should not be equal to zero. If these differences would be zero in the sample, then this would lead to an error when examining the asymmetric forecasts timeliness regression model as its equation requires to deduct the Preceding EPS forecast from the Current EPS forecast and scale this by its LagPrice.

▪ An analyst should at least have issued ten forecasts during 2012 -2016, and it is not necessary for these ten forecasts to be firm related. This minimum of ten forecasts per analyst means that they have issued two forecasts per firm per fiscal year. Originated from Hugon and Muslu (2010), Jung et al. (2017) required that each analyst should issue at least eight forecasts per year per firm of which two negative and two positive revision period stock return. However, this requirement has been adjusted to fit the analysis’ period in this research.

▪ The firm needs to have forecasts issued by both female and male analysts. ▪ The genders of the analysts are known.

(30)

26 The overview below gives clear insight into the data collection process, which leads to a final sample size of 24,598 observations.

TABLE 1 Sample

Observations Collected observations (only firms with 31st of December as fiscal year end): 138,232 After deduction of firms without one or more negative annual return: 40,715 After deduction of difference between preceding and current forecast equal to 0: 40,071 After deduction of analysts with less than 10 forecasts issued in 2012-2016: 31,290 After deduction of firms with forecasts issued by only male analysts: 24,898 (After deduction of which the gender of the analyst could not be identified: ) (21,290)

Final observations size: 24,898

(21,290) The final sample consists of 700 firm-years and 875 analysts of which the gender is known (in total 1,068 analysts).

(31)

27 5.2 Descriptive statistics

TABLE 2 Descriptive Statistics

Variable Mean Std. Dev. Lower Quartile Upper Quartile Median Minimum Maximum

Accuracy -0.00144 0.0348 -0.0014 0.0008 -0.0002 -1.7929 0.3772 MATCHC _0.0251 _0.6725 _-0.0299 _0.0443 ₀ _-3.4135 _3.5924 MATCHL _-0.0078 _1.1788 _-0.0283 _0.0481 ₀ _-8.7821 _4.0632 MATCHC_*FEMALE _0.0213 _0.0512 ₀ ₀ ₀ _-0.0677 _0.1384 MATCHL_*FEMALE _0.0281 _0.0664 ₀ ₀ ₀ _-0.0653 _0.1811 AnalystATC 0.0054 0.1143 -0.0130 0.0246 0.0009 -0.5626 0.6034 FirmATCC _0.0282 _0.6161 ₀ ₀ ₀ _-3.4222 _3.5886 FirmATCL _0.0031 _1.1479 ₀ ₀ ₀ _-8.7959 _4.0268 REV -0.0002 0.0023 -0.0002 0 0 -0.0126 0.0080 FEMALE(Dummy) 0.0969 0.2958 0 0 0 0 1 EarningsVol 1.4558 1.3607 0.5788 1.8348 0.9482 0.1322 7.8246 FSIZE 124023.20 304409.9 10178.2 86174 31782 2075.784 1880382 AFollowing 30.3849 8.5518 24 36 30 9 51 HORIZON 113.6799 80.8251 56 175 97 3 363

Notes: table 2 shows the descriptive statistics of the dependent, independent and control variables. Furthermore, it also shows AnalystATC and FirmATCC(L) _{which were used to calculate the}

four variants of the dependent variable MATCH (MATCHC(L)_{and MATCH}C(L)_{*FEMALE. The given statistics are based on 24,898 observations over a period from 2012 to 2016. Except}

(32)

28 The descriptive statistics of the dependent, independent and control variables are given in table 2. In order to reduce the influence of extreme observations on the regression, the above-mentioned variables are winsorized. This is done by setting the top 1% of the smallest and largest values equal to the less extreme observations. Additionally, log

transformation has been applied to FSIZE, which is measured by the total assets, to make the distribution of the values less skewed.

The mean of MATCHL, the absolute difference between the asymmetric forecast timeliness coefficient (AnalystATC) and the asymmetric earnings timeliness coefficient (FirmATCL)), is -0.0078. Furthermore, the mean of AnalystATC, and FirmATCL are respectively 0.0054 and 0.0031. This is relatively close to zero, and these numbers can be explained by the sample of this research resulting from adjustments in two requirements originating from Jung et al. (2017). As explained in paragraph 5.1, due to this study’s relatively small analysis period of five years, the requirement of having at least eight annual return observations per firm that consists at least two positive and two negative annual returns, has been adjusted. This requirement has been adjusted to at least one negative annual return out of five fiscal years of each firm. Additionally, the requirement of Jung et al. (2017) that the analyst has at least issued eight earnings forecasts with a minimum of two negative and two positive revision period stock return for a firm has been adjusted as well. The

adjusted requirement is that an analyst issued at least ten forecasts during this study’s analysis period. From the minimum of ten forecasts issued by an analyst, every two forecasts are regarding the same firm and fiscal year, so these observations can be applied when

conducting equation 2, and at least one of the revision period stock returns must be negative. In short, the annual return requirement for both variables have been mitigated to requiring at least one negative period holding stock return (both annual and revision period return) instead of two.

However, the adjustment of these requirements to better match this study’s analysis period affects MATCHL, AnalystATC, and FirmATCL. The regression model for both AnalystATC and FirmATCC(L) involves a dummy variable. This dummy variable in the regression model of FirmATCC(L) is equal to one when the annual buy-and-hold return of the firm is negative, and zero if it is positive. In the regression model that gives the output of the AnalystATC’s coefficient, the dummy variable is equal to one when the buy-and-hold return between the current and preceding EPS forecast (revision period) is negative, and zero if otherwise. Consequently, the output of both AnalystATC and FirmATCL contain more zero

Gender, accounting conservatism and financial analysts’ forecast accuracy