Analysts as Intermediary Between Managers and
Investors – A Closer Look at the Financial Sector
University of Groningen
Name: J.A. Jansen Student number: s2173190 Study program: MSc Accountancy Supervisor: prof. Dr. D.A. de Waard
Co-assessor: Drs. L.M. Wielens
Abstract: This paper shows an attempt to measure the relationship between complexity based
information and the behaviour of analysts and investors in the financial sector. Where previous research mainly focused on the readability of disclosures when addressing the understandable knowledge of investors, this paper provides evidence that investors do not use accounting information because they misunderstand the complexity of important processes within firms. Results on three different complexity measures partly confirm that analysts are more interested in firms with a higher complexity. Moreover, investors seem to use information published by analysts to control for the informational gap when analysing complex firms. Therefore, analysts can be seen as intermediary party that provides more investment based information to investors.
Field key words: Analysts, Complexity, Financial Sector, Investors, Informational Content,
Recommendation.
Date: 22 January 2017 Word count: 12.373
1. Introduction
Financial markets can be conceived as complex systems. As a result of financial innovation over the past years, a rise of complex securities has occurred (Brunnermeier and Oehmke, 2009). While the main use of financial information should be helping users when taking decisions (FASB, 1978), the increasing complexity of the market makes it more difficult for investors to analyse firms in the financial sector and investors should put more effort in investing in the financial sector. Moreover, within financial markets there are differences as well. Bank, insurers and investment funds all have their own complex systems and regulations. For investors an information problem occurs, by cause of an increasing amount and increasing complexity of data, it becomes more difficult for investors to analyse the information and draw correct conclusions (Brunnermeier and Oehmke, 2009).
This expansion in complexity and the amount of information that is disclosed by firms, resulted in different concerns about the effectiveness of disclosures. Consequently, the Securities and Exchange Commission (SEC) and the Wall Street Journal issued concerns about the readability of annual reports (Schroeder, 2002; Cox, 2007). The SEC has a serious doubt whether investors still understand the information disclosed in the annual report of a firm. Accordingly, the SEC issued a report to help firms with writing their annual report because the main issue, according to the SEC, is that the disclosures are too difficult to read (SEC, 1998).
By cause of the gap between disclosed information and information actually used by investors for making investment decisions, different questions arise about the role of analysts in providing information that can be used to fill this gap. Are analyst basically providers of this kind of information and can they be seen as intermediary for this kind of information to investors? Especially, due to the more complex nature of the financial market, are analyst recommendations used more for firms facing higher complexity?
Previous researchers considered the relationship between analyst reports and the market reaction on their recommendation (Abarbanell, 1991; Bradshaw, Drake and Myers, 2012; Souček and Wasserek, 2014). Overall, these studies conclude that there is a positive market reaction following a change in the analyst recommendation. Therefore, investors can use the recommendations to gain abnormal returns. A main finding for this study is that when the disclosed information is less readable, the analysts that follow a firm is higher (Lehavy, Li and Merkley, 2011). This indicates that when investors have to put more effort in analysing disclosed information, analysts respond by giving more information to the investors. This
contributes to the discussion of the SEC by providing evidence that investors cannot use annual reports of low readability since investors do not understand the disclosed information.
While previous researchers (Lehavy et al., 2011) and the SEC mainly focused on the readability of the disclosures of firms, in this paper I will take another sight. Is it just this readability of the disclosed information or do investors not use the information because they simply do not understand the complexity of important processes of these firms? This is an important difference when drawing conclusions about the usage of firm disclosures. While when it is just the readability of the filings this will mean that firms can make a difference by writing their disclosures in better plain English. However, when the problem in understanding is due to the complex operations financial firms face today, this will mean that the problem has to do with the capability and knowledge of investors. Moreover, where other researchers mainly excluded financial firms because of their different disclosure practices (Chen, Miao and Shevlin, 2015), I will focus solely on the financial sector to get an in depth understanding of the specific sector. Resulting, the research question of this paper is: Do investors use analyst reports to solve for their misunderstanding of the complex businesses financial companies face? In this study I argue that, if a firm is more difficult to analyse for an investor, because of the complexity of the firm, this will increase the demand for more information. Therefore, there is a higher demand for analysts. My results are consistent with my predictions. Indeed, a more complex firm will have more analysts following the firm.
Next, I take a look at the relationship between the informativeness of an analyst report and the complexity of the firm. In correspondence with the higher demand for the work of analysts, I am predicting that reports on more complex firms are more valuable for investors since they give more information. Once more, the results confirm my predictions and suggest that investors indeed find analyst reports more valuable when they are about more complex firms.
The conclusions of this paper are interesting for different parties involved in the financial market. For investors the findings are interesting because they tell them more about the time they have to spend on analysing information. When investors got a limited amount of time it could be more profitable to use the report of an analyst because the information given by the firm is too complex. Moreover, if investors do not understand the industry they would like to invest in, it could be beneficial to not even try to understand the information, but use a report of an analyst instead. For the accounting industry this research is interesting because it tells more about the usage of analyst forecasts and vice versa the usage of accounting information. The results of this paper suggest that analysts can be used as intermediaries to give
the investors suitable information for their investments. For managers the results of this study will give a better view on the influence of the quality of the annual report. In essence, it will help managers in making decisions about the way they disclose their information. Should managers make their reports more readable, or is this not the way to solve the informational problem because the problem lies deeper, namely in the complex operations of these financial firms?
This paper continues as follows. Section two will focus on previous literature about analysts’ forecasts and the financial market. Additionally, it will describe the previous focus of academics on readability and the focus on complexity of this study. Chapter 3 will describe the data and will further explain the sample and the methodology used. Section 4 will give the results of the tests, where chapter 5 will provide some concluding remarks. At last, section 6 will describe the limitations of this study and will provide opportunities for future research.
2.
Literature
In this section I will describe results of other academic papers that are important for the research question of this paper. Previous research is used to construct a theoretical framework and will result in different hypotheses that will be tested in the following chapters.
2.1 Firm information and complexity
Information disclosed in annual reports is largely firm specific and to properly understand the information a user should have a basic level of knowledge about the industry or the user should have a certain level of knowledge about accounting. Because each user will have different levels of knowledge, investors will have different capabilities in analysing the disclosed information (Indjejikian 1991; Ball 1992). Previous academic papers already researched the value relevance of annual reports. First, literature confirms that annual reports are disclosures of relevant information since there is a post announcement drift in the stock price (Ball and Brown, 1968; Healy, Huttom and Palepu, 1999). However, a more recent paper argued that the value relevance of accounting information has decreased over the past few decades in the U.S. Notwithstanding, this decline is not measured in non-U.S. studies. (Beisland, 2009).
Moreover, the more involved one is with a company, the more private information one will have. Due to the separation of ownership and control, the management of companies are more involved in the firm's operations than the shareholders of the firm (Seyhun, 1986). This mostly applies for information about the performance of a firm because managers are involved with the operations and shareholders are not. Of course, managers have to provide information
about the performance of the firm to shareholders. This information, however, is found to be very positive and incomplete (Desai, Hogan and Wilkins, 2006). Therefore, how closer one is to the activities of an organization, the smaller the informational gap for this person shall be (Ndofor, Wesley and Priem, 2015).
Furthermore, when a firm is complex, it will have more operations and divisions that will lead to profit (Heeley, Matusik and Jain, 2007). Therefore, it is more difficult for investors to understand the business and make connections between the different segments and different kinds of information about it. More specific, because it costs more time for investors to analyse firms in complex industries, the information asymmetry between the shareholders and managers is greater for complex firms (Heeley et al., 2007).
Financial markets, in particular, have become a lot more complicated over the past years. As a result of innovation in financial products, risk is better shared. However concurrently, a lot of new complex financial products arose (Brunnermeier and Oehmke, 2009). Considering this increase of complex products, users of financial disclosures have to deal with complex information progressively. According to Brunnermeier and Oehmke (2009) simply disclosing more information to investors does not solve the complexity issue. Even worse, this will lead to an information overload where an overabundance of information is given but this information is not understood. Standardization or regulation by a third party (for example the IFRS) can reduce complexity by producing more homogeneous information. Therefore, different additional regulations are set for financial companies.
Concluding, accounting information seems to be valuable. However, in the U.S., the value relevance of accounting information has decreased over time. Considering the rise of complex systems in the financial sector over time, it has become harder for investors to analyse firms in this sector and the information asymmetry has become greater. By cause of missing useful information, an informational gap for investors has occurred where a lot of information is given, however, the question is whether this information is understood.
2.2 Financial analysts
The informational gap that occurs for investors creates an opportunity for others. Financial analysts responded to this gap by making information useful to investors through specific advice and forecasts about a firm. Important when analysing the intermediary role of analysts, is the relationship between annual reports and analyst forecasts. Although, some say that analysts base their forecasts on more information than just the annual report, analysts still use company information from the annual report to support their forecasts (Fogarty and Rogers,
2005; Flöstrand and Ström, 2006). Flöstrand and Ström (2006) conclude that analysts use both the financial as the non-financial information disclosed by firms and consider it to be valuation relevant.
Researchers about the content of the annual report conclude that more informative information leads to a higher analyst following and less diffusion among the different analysts (Lang and Lundholm 1996; Botosan and Harris, 2000; Barth, Kasznik and McNichols, 2001; Barth et al. 2001). Others find an immediate stock price reaction following on buy and sell recommendations, which indicates that analyst recommendations are informative (Brav and Lehavy, 2003; Michaely and Womack, 2005). Confirming previous literature, if the recommendations that analysts publish are informative, a price change after the announcement date is expected. When the semi-strong form of market efficiency holds, this indicates that when the information is published publically the market should respond in a short time period (Malkiel and Fama, 1970; MacKinlay, 1997).
Lehavy, Li and Merkley (2011) combine previous studies and find that firms with disclosures that have less informational content will have a higher analyst following. Moreover, they argue that the proportion of returns that connects with the reports of analysts is higher for firms with a lower readability of disclosed items. Lehavy et al. (2011) succeed to extract conclusions on the impact of the overall readability of the disclosures of firms on the way analysts behave and the impact on the behaviour of investors. Lehavy et al. (2011) argue that because less readable information will cost more for investors to interpret, they value analyst following higher and therefore there are more analyst reports on firms with less readable communications. The readability of disclosures is measured by the Fog Index, were 10-K reports are used as communication reports. The findings of the paper are consistent with the expectations, the analyst following for firms with less readable disclosures is higher. Moreover, the reports that analyst issue on companies with less readable information are more informative to investors. Therefore, Lehavy et al. (2011) conclude that when a 10-K filing has lower readability, analysts reports are used more because users do not understand the information that is disclosed by the firm.
In the financial sector, risk disclosures are obligated by law. In this way, financial firms give more information about the risks they occur. Following Brunnermeier and Oehmke (2009) this would decrease the complexity of information. However, these risk disclosures are viewed upon as difficult or very difficult to understand (Linsley and Lawrence, 2007). Because of the additional regulation on risks in the financial sector (for example the Solvency Modernization
Initiative1), the findings of Lehavy et al. (2011) and Linsley and Lawrence (2007) are important
for companies in the financial sector because this additional reporting makes the annual report harder to read.
Concluding, previous literature agrees on the influencing role of analysts. A major step is made by the findings of Lehavy et al. (2011) who conclude that the readability of firm disclosures influences the way analysts behave and influences the usefulness of analyst reports. This study will be in accordance with the study of Lehavy et al. (2011). I argue in the same way, however, in this research the behaviour of analysts and investors is not connected to the readability of disclosures but to the complexity of the operations of the firm were the analyst report documents about2.
2.3 Criticism on readability scores
Lehavy et al. (2011) use the Fog Index to measure the readability of 10-K filings. However, readability indexes are under criticism (Klare, 2000; Anderson and Davison, 1988; Jansen and Lentz, 2008; Kraf and Pander Maat, 2009; Feenstra, 2012). There is argued that, the readability formulas do not take the interaction between the reader and the text into account. The knowledge, the interest and the skills of the reader are neglected. Moreover, these formulas are designed to test knowledge of the primary school and it is not demonstrated that the function also works for specialized texts, like annual reports.
Accordingly, in this study I will not focus on the written complexity of a firm’s disclosures. Instead, I will focus on the complexity of the firm and industry where it operates in. Feenstra (2012) suggest different other methods that should work when testing the complexity in texts. Two of them (content analysis and genre analysis) are approved as valid methods. Still, these methods only focus on the complexity of texts and do not take the complexity of the company into account. Therefore, these measures cannot be used in this study. A new measure of the quality of the disclosure of firms is given by Chen, Miao and Shevlin (2015). Disclosure quality is measured as the ‘fineness’ of accounting data and is based on the total non-missing line items in annual reports. However, after examining the measure for what it captures and what not, Chen et al. (2015) argue that the disclosure quality measure should only be used when measuring the disclosure quality and not when measuring the complexity of a firm's operations. Because in this study I will specifically look to the operational
1 The Solvency Modernization Initiative (SMI) is the comparable regulation in the United States for Solvency in
Europe.
complexity of the firm itself and not the disclosure, this measure cannot be used in this study as well.
Hence, to measure complexity I will use three different methods. First, industry complexity is measured as constructed by Ndofor et al. (2015). Moreover, two other methods are used in measuring complexity, in this way I can ensure that different aspects of complexity are captured. Further details will be given in the methodology section.
2.4 Hypothesis development
The first hypothesis that is tested is about the number of analysts that follow a firm. I assume that analysts know about the information gap mentioned in previous literature. When a firm is harder to analyse, there will be a higher demand for the work of analysts. Therefore, analysts will follow more firms that give less informative information or more complex information to their users. In this way, the information that analysts publish is more valuable to investors and therefore analysts can earn more. In the financial sector this will be even more the case than in other sectors because of the complex nature of financial companies (Brunnermeier and Oehmke, 2009). More specific, this will result in a higher profitability for analysts when they analyse firms where the informational gap is the biggest. Therefore, the analysts that follow a firm is predicted to be higher for firms with a higher complexity. This results in the following hypothesis:
𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠 1 (𝐻1): 𝐴 ℎ𝑖𝑔ℎ𝑒𝑟 𝑙𝑒𝑣𝑒𝑙 𝑜𝑓 𝑐𝑜𝑚𝑝𝑙𝑒𝑥𝑖𝑡𝑦 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑠 𝑡ℎ𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑛𝑎𝑙𝑦𝑠𝑡𝑠 𝑡ℎ𝑎𝑡 𝑓𝑜𝑙𝑙𝑜𝑤 𝑎 𝑓𝑖𝑟𝑚.
Consistent with the theory above investors should value a report of an analyst higher when the report gives more new information. Because the financial industry is complex and additional disclosures are made about the risks the company faces, it is harder for investors to analyse the information given by the firm. Therefore, when there is new information given, in the form of an analyst report, this should be valued higher by investors when the analyst report is about a company that is harder to analyse. Following the semi-strong form of market efficiency (Malkiel and Fama, 1970) the market should respond immediately when new information is given. Because investors value analyst reports on complex firms higher than reports on less complex firms, a larger market reaction is expected on the publication of analyst recommendations on firms with a higher complexity. This results in the following hypothesis:
𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠 2 (𝐻2): 𝐴𝑛𝑎𝑙𝑦𝑠𝑡 𝑟𝑒𝑝𝑜𝑟𝑡𝑠 𝑎𝑟𝑒 𝑣𝑎𝑙𝑢𝑒𝑑 ℎ𝑖𝑔ℎ𝑒𝑟 𝑤ℎ𝑒𝑛 𝑎 𝑓𝑖𝑟𝑚 𝑖𝑠 𝑚𝑜𝑟𝑒 𝑐𝑜𝑚𝑝𝑙𝑒𝑥.
3. Sample and Variables
The following section first describes the sample used in this study. Thereafter, the different variables used to estimate the predictions are described. At last, the control variables added to the estimations are described.
3.1 Sample
The sample used in this study consists of recommendations on U.S. firms for the years 2005-2015. U.S. firms are chosen because of the large availability of data about analyst recommendation in the U.S. To separate the different industries, first only financial firms are chosen and these firms are separated in different industries using an industry code3. This results in 7 different industries within the financial sector namely: Banking, Finance & Loan, Financial services, Insurance, Investments, Multi-Industry Finance and Savings and Loans4. By combining data from Datastream, Compustat and the I/B/E/S database, the data collection resulted in a sample with 5,943 observations about 194 different financial firms. Companies that did not have a match are excluded from the sample.
3.2 Variable description
It is hard to find a workable definition for complexity (Brunnermeier and Oehmke, 2009). Hence, it is hard to measure complexity with just one variable. Therefore, in this study I use three different measures of complexity that are argued before to measure complexity. In this way I can compare the different proxies and draw better conclusions over complexity as a whole. Below the different measures for complexity are described. Following Lehavy et al. (2011) I use complexity as a proxy for the costs to analyse a specific firm.
Complexity based on heterogeneity. I follow Ndofor et al. (2015) in measuring
industry complexity. Ndofor et al. (2015) determine industry complexity based on the level of heterogeneity in a specific industry. When a firm has a lower number of competitors, the firm is less concerned about sharing information to competitors and other external parties. Therefore, information will more include the same content and is better to analyse for users. Hence, when
3 Sector/Industry/Group codes are used. Because these codes are available in I/B/E/S, each firm is already
assigned to a different code in the database.
4 The SIG codes computed by I/B/E/S are constructed using the S&P classification system. A Description of
an industry consists of few homogenous firms the industry will be less complex to analyse. Contrariwise, industries that are less concentrated will have more differences between firms, which will lead to more competition and more unpredictability in the industry. Therefore, less concentrated industries are harder to analyse and are therefore seen as more complex.
Following Ndofor et al. (2015), industry complexity is calculated as the inverse of the 4 firm ratio. To get this ratio, first the total sales of each industry is calculated by adding up the sales for all firms in the industry in Compustat. The four largest firms, based on sales, are selected and the 4 firm ratio is calculated by dividing the sales of the four largest companies by the total sales for the industry. Therefore, the 4 firm ratio for each industry is calculated as follows:
4 Firm ratio = Sales 4 largest firms
Sales total industry (1)
Formula (1) results in a different ratio for each industry for the years 2005-2015. The ratio works as follows, if the score of the ratio is 1 this means that the industry consists of 4 firms or less and therefore the industry is highly concentrated. While when the ratio gives a value of 0.2 this means that the four biggest firms in the industry possess one fifth of the total industry. Therefore, the lower the four-firm ratio will be, the more complexity there is in a specific industry.
Complexity based on segments. Following Lehavy et al. (2011) and Bradshaw, Miller
and Serafeim (2008) I use the natural logarithm of the number of segments a firms operates in as a proxy for the operational complexity. To construct this measure the number of different SIC codes assigned to a firm are collected using Datastream. A SIC code is assigned to a firm if the firm has business activities in the area the SIC codes stands for. Therefore, each firm has at least one SIC code, but the number of SIC codes can increase to a maximum of 8 when a firm is engaged in multiple business activities. The more business activities a firm is engaged in, the more complex a firm is (Heeley et al., 2007).
Market complexity. Lastly, I use the information uncertainty for each specific firm to
measure the complexity of the specific stock. Therefore, the standard deviation of each firm’s monthly stock returns are included. Bhushan (1989) argues that higher volatility increases the value of private information. Moreover, Lehavy et al. (2011) use the volatility of firm monthly stock returns as a control variable for business complexity because they argue that a higher volatility increases the cost to analyse the specific firm. Therefore, a third variable is
constructed to measure the complexity of a specific firm’s stock, namely the standard deviation of the prior month stock returns5.
Analyst Following. Consistent with previous literature (Lehavy et al, 2011; O’Brien
and Bhushan, 1990), analyst following is measured as the number of analysts that are included in the I/B/E/S consensus forecasts in a specific month.
Analyst recommendation. Each analyst publishes a recommendation, which are
collected in the I/B/E/S database. However, analysts all have different estimator texts. Thomson Reuters transferred the original estimator texts into IBES texts which results in an ordinal variable ranging from 1 to 5. Each number refers to “strong buy”, “buy”, “hold”, “underperform” and “sell” respectively.
Informativeness of analyst reports. The informativeness of analyst reports is
measured in different ways in previous literature. The market reaction after a recommendation is published can be measured using an event study. Because of the rationality of the market, effects of a specific event should be captured in the market price immediately (MacKinlay, 1997). Therefore, when an analyst report is published this should be reflected in the market price of a security. However, because market reactions on analyst reports are mostly visible on the announcement day itself (Brav and Lehavy, 2003; Michaely and Womack, 2005), other measures are constructed to test the informativeness of analyst reports.
The measure used in this study is constructed by Frankel, Kothari and Weber (2006). This method results in a single measure where only the announcement day is included as reaction on the analyst report. Frankel et al. (2006) argue that the measure includes the surprise effect which results in large stock price reactions. Because the method results in one Analyst Informativeness variable, it is of a higher value than an event study because in this way I can control for other influencing factors. Following Frankel et al. (2006) Analyst Informativeness (AI) is a measure of the informativeness of an analyst report date. Therefore, the stock price of the announcement date is divided by the sum of the stock returns on all trading days in the calendar year6 in which the analyst recommendation is published using the following formula:
𝐴𝐼 = ∑∑𝑡=1𝑅𝑡,𝑠
𝑅𝑡,𝑠
𝑡=1𝑡𝑜250 (2)
Following Lehavy et al. (2011) I omit the final step that is done by Frankel et al. (2006). Frankel et al. (2006) divide the formula above by the number of revision dates. This results in
5 For each firm the number of working days is set on 21 days to assure that the measure constructed is the same
for each firm.
6 Following Frankel et al. (2006) a calendar year is defined as 250 trading days to make sure each measure
a measure of the average informativeness of an analyst report. I do not include this step because in this research the focus is not on the average informativeness but each recommendation is studied separately.
The measure for the informativeness of analyst recommendations gives a value of 0.004 if analysts would supply no information because in this situation the average return would be expected and therefore, the value of AI is 1/250th of the sum of the returns.
3.3 Control variables
Previous academic papers researched subjects associated with the informational content of disclosures, the complexity of industries and analyst recommendations before. This study uses several control variables that, in the past, have been shown to relate with analyst forecasts. All control variables mentioned below are used in both hypothesis because the same influence is expected.
Various studies find that the size of a company is one of the most influencing factors on the proportion of analysts that follow a firm (O’Brien and Bhushan, 1990; Lang and Lundholm, 1996; Barth et al. 2001). Following Lehavy et al. (2011), a larger firm size should result in a higher analyst following and should indicate a better information environment, more complexity, and greater demand for analyst forecasts. The natural logarithm of the market value at the end of the year prior to the recommendation publishing is used as a proxy for size.
Following Barth et al. (2001) and Lehavy et al. (2011), I control for the growth of a firm. When a firm has a higher growth rate the analysts that follow the firm is higher because of more potential value for investing. However, it is harder for analysts to make a forecast about firms with a higher growth level because of their unpredictability (Lehavy et al., 2011). Lehavy et al. (2011) and Barth et al. (2001) find a positive association between growth and analyst following, however a negative association is found by others. The growth of a firms is calculated as the average growth rate in sales over the prior three years before the recommendation is made (Lehavy et al., 2011).
Moreover, Barth et al. (2001) and Lehavy et al. (2011) control for the firms’ information environment. Barth et al. (2001) use intangible assets to draw conclusions on the information environment and argue that it is harder for investors to analyse a firm with more intangible assets and therefore, the analyst following on these firms is higher. Accordingly, they find that more analysts follow a firm with higher research and development expenses and higher advertising expenses. I follow both Barth et al. (2001) and Lehavy et al. (2011) and compose an Advertising variable to control for the information environment. Advertising is defined as
the ratio of advertising expense to operating expense. No R&D variable is added because of limited data availability.
4. Results
In this section the hypotheses constructed above will be tested. First, the descriptive statistics of the variables are given to give a better feeling of the data. Thereafter, a correlation analysis is performed to make sure that the data does not involve multicollinearity. The section continues with the testing of hypothesis 1 and finishes with testing hypothesis 2.
4.1 Descriptive statistics
The descriptive statistics are separated in three parts. First, the descriptive statistics of the number of observations are given to give a better indication of the distribution of the observations over the different years. Second, descriptive statistics about the three complexity measures are given. Third, the descriptive statistics of the overall sample are given.
Table 1 presents the number of observations from 2005 to 2015. Notable is the decrease in the number of observations in the sample years. Because I/B/E/S adds the data in the database after the recommendations are made and there should be at least three recommendations on a firm before a firm is included in the database, it is logical that the number of recommendations decline over time. However, in this study this is not a problem because different years are not compared with each other. The number of different firms included in the sample are at a maximum in 2006 with 109 and at a minimum of 73 in 2015. However, the mean number of observations per firm do not decrease at the same level as the number of observations do. First, an increase is noticed but after 2009 the mean number of observations decrease to the lowest mean in 2013 and 2014.
Table 2 presents the descriptive statistics of the three different measures of complexity. The descriptive statistics are divided into the different industries to make it possible to compare each industry based on complexity. The mean and median of the 4 firm ratio for the total sample are 0.23 and 0.16, respectively. This means that in my sample, overall, the four largest firms possess approximately one fifth of the total industry. Overall, the scores on the 4 firm ratio are close to one-another, with a standard deviation of 0.12. The number of segments are more divided across the sample, with a standard deviation of 2.20 and a mean and median of 3.70
Table 1: Descriptive statistics for number of observations Year Number of observations Number of firms Mean number of observations per firm 2005 644 108 6.0 2006 685 109 6.3 2007 641 106 6.0 2008 660 94 7.0 2009 670 87 7.7 2010 560 83 6.7 2011 492 82 6.0 2012 491 83 5.9 2013 392 78 4.6 2014 367 80 4.6 2015 341 73 4.7 Mean 495.3 82 6.0
This table reports the number of observations and the number of firms for each year of the sample period.
Mean number of observations per firm is calculated as: Number of observations / Number of firms.
and 3 respectively. Moreover, the standard deviation of returns has a mean of 1.10, a median of 0.72 and a standard deviation of 1.23.
Notable, are the differences of complexity between the different industries and the differences in the order in each complexity measure. While the 4 firm ratio and the standard deviation of returns both indicate that Financial Services and Investments are the most complex, the number of segments measure indicates that Finance and Loan and the Banking industry are the most complex industries. The low standard deviation for the 4 firm ratio for each industry is logical because large differences in the complexity of an industry between years are not expected. However, the Banking and Savings and Loan industries have an interquartile of 0.12 and 0.34, respectively. This indicates that the complexity of these industry vary over the sample period. Specifically in the Banking industry, the 4 firm ratio has the highest values in 2008 and 2009 and thereafter the 4 firm ratio decreases again. The high values can be explained due the financial crisis that started between July 2007 and September 2008 (Eichengreen, Mody, Nedeljkovic and Sarno, 2012). In the crisis a lot of banks went bankrupt which could explain the high score of the ratio.
Table 2: Descriptive statistics for firm complexity
Industry (SIG) Variable Count Mean Std. Dev Q1 Median Q3
Banking 4FirmRatio 1663 0.37 0.06 0.31 0.38 0.43
Number of Segments 1663 5.06 2.16 3 5 8
STD of Returns 1663 0.96 0.91 0.40 0.69 1.18 Finance and Loans 4FirmRatio 382 0.23 0.01 0.22 0.23 0.24
Number of Segments 382 5.53 2.79 3 8 8
STD of Returns 382 0.59 0.55 0.23 0.40 0.74 Financial services 4FirmRatio 687 0.15 0.02 0.13 0.15 0.17
Number of Segments 687 3.37 2.25 1 3 4 STD of Returns 687 0.91 1.74 0.24 0.47 1.00 Insurance 4FirmRatio 252 0.33 0.01 0.31 0.32 0.34 Number of Segments 252 3.81 2.05 2 4 5 STD of Returns 252 0.93 0.79 0.41 0.69 1.17 Investments 4FirmRatio 2456 0.13 0.01 0.12 0.13 0.14 Number of Segments 2456 2.44 1.13 2 2 2 STD of Returns 2456 1.44 1.35 0.52 1.04 1.88 Multi-Industry Finance 4FirmRatio 47 0.18 0.07 0.15 0.16 0.17
Number of Segments 47 2.74 0.79 2 3 3
STD of Returns 47 0.42 0.34 0.19 0.33 0.52
Savings and Loans 4FirmRatio 456 0.33 0.16 0.19 0.22 0.53
Number of Segments 456 4.49 1.92 2.5 4 6
STD of Returns 456 0.69 0.79 0.24 0.45 0.82
Total 4FirmRatio 5943 0.23 0.12 0.13 0.16 0.31
Number of Segments 5943 3.70 2.20 2 3 5
STD of Returns 5943 1.10 1.23 0.37 0.72 1.37 This table reports descriptive statistics on the three different complexity measures: 4 Firm Ratio, Number of Segments and Standard Deviation of Returns of the prior month. The 4 Firm Ratio is computed as follows: Sales 4 largest firms / Sales total industry. The seven industries are based on the S/I/G industry classification.
Table 3 presents the descriptive statistics of the total sample of 5,943 observations. The mean number of analysts that follow a firm (NUMANALYST) is 12.81 with a median of 12. The mean of the measure for analyst informativeness (AI) is 0.004. This indicates that overall the trading days have the same informative value. However, there are small differences between the trading days (standard deviation of 0.001). Consistent with table 2 the mean and median of the four-firm ratio (4FIRMRATIO) are 0.23 and 0.16, respectively. The mean (median) for the number of segments (SEGMENTS) and the standard deviation of returns of the prior month (STDRETURNS) are conform table 2 as well and are 3.70 (3) and 1.10 (0.72), respectively.
The descriptive statistics for the control variables are also provided in Table 3. For firm size (SIZE), the mean and median are 12,700 and 3,455. Large outliers are excluded from the sample. However, the interquartile for firm size is 9,606, which indicates that there are still large differences in firm size. The mean and median of the sales growth in the prior 3 years before the announcement (GROWTH) are 10.82 and 6.37, respectively. This indicates that there are more firms that have a lower growth over the past 3 years, which is logical because there are also firms with negative growth rates included in the sample. Advertising expense (ADVERTISING) does not obtain the whole sample because of a lack of data. Advertising expense contains 2,713 observations with a mean and median of both 0.03. The analyst recommendation (RECOM) varies between 1 and 5, the mean and median for the recommendation are 2.61 and 3, respectively. This indicates that analysts in the sample, mostly give a hold (3) recommendation.
4.2 Correlation Analysis
The correlations between the dependent and independent variables are presented in Table 4. Notable is that the correlation between the size of a firm and the number of analyst that follow a firm is very high and significant (0.8083). However, this is no problem for the regression because NUMANALYST is my dependent variable in hypothesis 1 while LOGSIZE is a control variable. This finding indicates that larger firms have more analyst following which confirms previous literature (Lehavy et al., 2011).
Moreover, these results show positive correlations between most of the variables and the analyst following. This is consistent with the prediction of hypothesis 1 that the complexity of a firm results in a higher following by analysts. Interesting is the negative correlation between NUMANALYST and GROWTH, which is not consistent with previous literature. This suggests that higher growing firms will have less attention from analysts, which is contrariwise with
Table 3: Descriptive statistics of variables
Label Variable n Mean Std. Dev Q1 Median Q3
Analyst following NUMANALYST 5943 12.81 7.77 6 12 18
Analyst Informativeness AI 5943 0.004 0.001 0.004 0.004 0.005
4 firm ratio 4FIRMRATIO 5943 0.23 0.12 0.13 0.16 0.31
Number of Segments SEGMENTS 5943 3.70 2.20 2 3 5
Standard Deviation
Returns STDRETURNS 5943 1.10 1.23 0.37 0.72 1.37
Market Value Firm SIZE 5943 12700 28202 791 3455 10397 Sales growth prior 3 years GROWTH 5943 10.82 44.57 0.26 6.37 16.39 Advertising expense to
total operating expense ADVERTISING 2713 0.03 0.03 0.01 0.03 0.04
Recommendation RECOM 5943 2.60 0.96 2 3 3
This table reports descriptive statistics for all variables used in this study. The variables are combined over the years 2005-2015. Variable definitions: NUMANALYST= AI= 4FIRMRATIO= SEGMENTS= STDRETURNS= SIZE= GROWTH= ADVERTISING= RECOM=
Number of analysts that follow a firm in the month that the recommendation is made. Market return of the announcement day divided by the total market return over the entire year. First measure of complexity. Calculated as the sum of the sales of the four largest firms in an industry divided by the total sales in that industry.
Second measure of complexity. Number of reported business segments in Datastream. Business segments are based on SIC codes.
Third measure of complexity. Standard deviation of the firm’s stock return from the month prior to the recommendation.
Total market capital of the firm (in millions) in the fiscal year of the recommendation Average growth rate of firm sales over the prior 3 years.
Advertising expense as a percentage of operating expense in the year prior to the recommendation. Recommendation code in I/B/E/S. The code ranges from 1 to 5, where each number refers to “Strong buy”, “Buy”, “Hold”, “Underperform” and “Sell” respectively.
Table 4: Correlation between dependent and independent variables
Variable NUMANALYST AI INV4FIRMRATIO LOGSEGMENTS STDRETURNS LOGSIZE GROWHT ADVERTISING
NUMANALYST 1 AI 0.048*** 1 INV4FIRMRATIO 0.101*** 0.062*** 1 LOGSEGMENTS 0.239*** 0.024* -0.480*** 1 STDRETURNS 0.171*** -0.101*** 0.147*** -0.087*** 1 LOGSIZE 0.808*** -0.057*** 0.087*** 0.294*** 0.246*** 1 GROWTH -0.104*** 0.000 -0.005 -0.034*** -0.011 -0.098*** 1 ADVERTISING 0.152*** 0.096*** 0.345*** -0.094*** 0.090*** 0.155*** -0.045*** 1 *, **, *** p<0.1; p<0.05; p<0.01, respectively.
results of Barth et al. (2001) and Lehavy et al. (2011) but consistent with their alternative explanation7.
When looking at the correlations between Analyst Informativeness and the independent variables more negative correlations appear. Both STDRETURNS and LOGSIZE show to have a negative correlation with Analyst Informativeness, were a positive relationship was predicted. Other correlations are not outstanding and are consistent with the predictions. Moreover, VIF scores are all below 1.5 which indicates that there are no multicollinearity issues.
Additionally, to check if the correlations between the dependent and independent variables are influenced by other independent variables the partial correlations are shown in Table 5. Panel A shows the partial correlations of the number of analysts with the independent variables while panel B shows the partial correlations of the Analyst Informativeness with the independent variables. Notable, is the significant switch of the positive correlation between NUMANALYST and STDRETURNS to a negative partial correlation. This negative partial correlation indicates that when a firm has a higher standard deviation of returns it will have less analysts that follow the firm. This is not consistent with the predictions and results of Table 4. Hence, this change in correlation indicates that the coefficient between the number of analysts and the standard deviation of returns is influenced by the other variables. Other significant changes in partial correlations do not switch from positive to negative of from negative to positive. Therefore, they are not considered to influence the results.
Table 5: Partial correlation between dependent and independent variables
Panel A Panel B
Variable Partial correlation Partial correlation
INV4FIRMRATIO 0.034 0.041 LOGSEGMENTS -0.007 0.034 STDRETURNS -0.131*** -0.151*** LOGSIZE 0.841*** -0.135*** GROWTH -0.147*** -0.015 ADVERTISING 0.013 0.095*** *, **, *** p<0.1; p<0.05; p<0.01, respectively.
This table reports partial correlations for all variables used in this study. Panel A shows the partial correlations for the number of analyst. Panel B shows the partial correlations for Analyst Informativeness.
7 Barth et al. (2001) and Lehavy et al. (2011) argue that analyst find it more difficult to make predictions about
highly growing firms. Therefore, costs to analyze these firms increase. If these costs become too high, analysts will not follow these highly growing firms anymore.
4.3 Analyst following and the complexity of industries
The prediction that the number of analyst that follow a firm is influenced by the complexity of a firm is estimated using the following regression:
𝑁𝑈𝑀𝐴𝑁𝐴𝐿𝑌𝑆𝑇, = 𝛽0+ 𝛽1𝐼𝑁𝑉4𝐹𝐼𝑅𝑀𝑅𝐴𝑇𝐼𝑂,+ 𝛽2𝐿𝑂𝐺𝑆𝐸𝐺𝑀𝐸𝑁𝑇𝑆,+
𝛽3𝑆𝑇𝐷𝑅𝐸𝑇𝑈𝑅𝑁𝑆, + 𝛽3𝐿𝑂𝐺𝑆𝐼𝑍𝐸,+ 𝛽4𝐺𝑅𝑂𝑊𝑇𝐻,+ 𝛽5𝐴𝐷𝑉𝐸𝑅𝑇𝐼𝑆𝐼𝑁𝐺,+ 𝜀, (3)
The regression is estimated using an ordinary least-squares regression with fixed or random8 effects.
Table 6 present the results of the estimation, with the t-statistics or z-statistics9 presented in brackets. The standard errors are clustered at the firm level. Model 1, Model 2 and Model 3 report the results of each separate complexity measure. Model 1 shows, as expected, that the coefficient for the inverse of the 4 firm ratio is statistically significant and positive. Which indicates that when a firm becomes more complex, the firm attracts more analysts. Therefore, analysts react on the additional demand of investors for information when firms are harder to analyse. Consistent with prior literature about firm size (O’Brien and Bhushan, 1990; Lang and Lundholm, 1996; Barth et al. 2001; Lehavy et al., 2012) I find a positive significant coefficient for the natural logarithm of the firm size. Accordingly, larger firms have more analysts that follow them. However, the coefficient that controls for the growth of a firm does not confirm prior research. As can be seen from table 6 the coefficient for sales growth is significant and slightly negative. This indicates that when a firm has a higher growth rate over the past 3 years, less analysts will follow the firm. Model 6 reports results after controlling for the information environment (ADVERTISING)10. Consistent with Barth et al. (2001) a significant and positive coefficient is estimated for ADVERTISING. This indicates that analyst following is higher for firms with higher advertising expenses. Other results do not change after controlling for the information environment.
Model 2 of Table 6 presents the results on the regression with only the number of segments included as a measure for complexity. The coefficient on the number of segments a firm operates in (LOGSEGMENTS) is positive and significant. This is not consistent with prior literature (Bhushan, 1989; Lehavy et al, 2011) but does confirm the expectation that a firm that
8 The random effects model is used when LOGSEGMENTS is included in the regression. This variable is time
invariant and therefore, stays the same over time. Time invariant variables cannot been estimated in a fixed effects model. Therefore, a random effects model is used.
9 T-statistic is given when the model performed is a fixed effects model. Z-statistic is given when the model
performed is a random effects model.
10 A new regression is estimated due to the low amount of observations with data about advertising expenses
(2,713). Each different regression is re-estimated with the control variable ADVERTISING included (Model 6,7,8,9 and 10).
Table 6: Regression model on the relationship between firm complexity and analyst following (Hypothesis 1)
Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10
INTERCEPT -0.731 (-0.94) -4.492*** (-6.51) 2.556*** (3.23) -1.336* (-1.72) -9.527*** (-13.15) -7.460*** (-5.03) -9.649*** (-9.87) -5.512*** (-3.69) -8.464*** (-5.75) -12.671*** (-12.59) INV4FIRMRATIO 1.439*** (24.61) 1.353*** (22.99) 1.077*** (3.97) 1.298*** (13.66) 1.177*** (12.36) 0.867*** (10.03) LOGSEGMENTS 0.452*** (1.02) 1.791*** (3.97) 0.476 (0.76) 0.619 (0.98) STDRETURNS -0.506*** (-12.41) -0.353*** (-9.09) -4.067*** (-10.24) -0.738*** (-9.66) -0.586*** (-7.79) -0.672*** (-8.84) LOGFIRMSIZE 0.714*** (7.34) 1.692*** (20.47) 1.370*** (13.57) 0.900*** (9.12) 1.460*** (17.46) 1.591*** (8.89) 2.236*** (17.28) 2.086*** (11.32) 1.857*** (10.30) 2.283*** (17.55) SALESGROWTH -0.002** (-2.53) -0.002** (-2.13) -0.002* (-1.83) -0.002** (-2.35) -0.002** (-2.36) -0.058*** (-10.38) -0.054*** (-9.54) -0.050*** (-8.82) -0.057*** (-10.31) -0.057*** (-10.53) ADVERTISING 110.401*** (19.12) 96.84*** (-9.87) 112.907*** (19.24) 107.865*** (18.86) 88.119*** (16.85)
Fixed Effects Yes No Yes Yes No Yes No Yes Yes No
Random Effects No Yes No No Yes No Yes No No Yes
Observations 5,943 5,943 5,943 5,943 5,943 2,713 2,713 2,713 2,713 2,713 R-squared 0.12 0.02 0.05 0.13 0.12 0.22 0.16 0.19 0.24 0.23 AIC 28,482.9 29,938.3 28,921.2 28,400.1 29,350.8 13,081.8 13,743.3 13,203.4 13,021.5 13,484.9 F-statistic 251.03 97.01 211.59 182.21 154.38 161.23 Chi^2 441.09 1018.47 728.08 972.61 Prob(F-statistic/Chi^2) 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
*, **, *** p<0.1; p<0.05; p<0.01, respectively.
This table presents coefficient estimates and t-statistics from the regression of number of analyst on complexity variables and control variables. In model 6, 7, 8, 9 and 10 the control variable ADVERTISING is included. Therefore, there are less observations than in the first 5 models.
is active in more segments attracts more analyst following. Results on control variables are consistent with model 1 after the 4 firm ratio is omitted and the number of segments are included. However, after controlling for the information environment (model 7), the coefficient for the number of segments is not significant anymore.
Besides, model 3 of table 6 shows the results of the regression of the standard deviation of returns of the prior month (STDRETURNS) on NUMANALYST. The coefficient of the standard deviation of the returns of the prior month before the recommendation is significant and negative. This does confirm the results of the partial correlation given in Table 5. Prior research (Bhushan, 1989; Lehavy et al., 2011) find a positive association between stock return volatility and analyst following which indicates that more analyst follow firms when private information is valuable. However, I find the opposite. This result is not conform results of Bhushan (1989), however it does confirm his alternative explanation that analysts bear more costs when analysing firms with higher return volatility.
Because model 7 indicates that the number of segments a firm operates in does not affect analyst following, a separate regression is estimated where both the 4 firm ratio and the standard deviation of returns of the prior month are included (model 4, model 9). Both measures stay significant.
At last, all the three measures for complexity are included in a regression (model 5, model 10). The results show that all three measures of complexity are significant when they are all included. However, when I control for the information environment, LOGSEGEMENTS is not significant anymore.
To compare the different models the Akaike information criterion (AIC) is used. The lower the AIC, the better is the fit of the model. When comparing model 1 to 5 with each other the lowest AIC is found for model 4. This indicates that the 4 firm ratio and the standard deviation of returns together give a model that explains the most. In addition, AIC is particularly low when the advertising variable is included. This indicates that the model is better when I control for the information environment. Accordingly, model 9 is the model with the best fit.
Next, I examine differences between the industries to get a proper view of the complexity measures. A stronger relationship is expected for the more complex industries because when an industry becomes more complex, the results of table 6 indicate that more analysts should follow them. From Table 2 we know the different rankings for each complexity measure. Table 7 presents the results of the different regressions for each industry, with t-statistics presented in brackets. From the results of Table 6 we know that model 9 has the best
fit. Hence, this model will be tested for each industry separately. Testing the null-hypothesis that all industries within the financial sector react the same results in a rejection. Therefore, the different coefficients in Table 7 can be compared between the industries.
The results on both complexity measures indicate that results of Table 6 are influenced by separate industries. More specific, only three industries appear significant on the 4 firm ratio, while Table 6 shows a positive significant coefficient. Accordingly, the Insurance, Investment and the Savings and Loans industries influence this result. Table 7 shows that the 4 firm ratio has the strongest relation with the analyst following in the Insurance industry.
Furthermore, results on the coefficient on STDRETURNS confirm the results presented in Table 6. Each coefficient appears significant and negative, where the Finance and Loan industry has the strongest predictor for analyst following.
When comparing these results with the means of the complexity measures presented in Table 2, it suggest that in each industry there are different reactions on the level of complexity. From Table 6 we know that a more complex firm has more analysts that follow the firm. Despite, Table 7 indicates that a higher mean of complexity in a specific industry does not involve a higher relationship between the measures for complexity and analyst following. Therefore, I can conclude that in all industries except the Finance and Loan industry, a higher complexity (lower 4 firm ratio) results in a higher analyst following. However, there are differences between each sector. A high overall complexity in an industry does not mean that the relationship between complexity and analyst following is strongest in that sector. For each industry the tested relationship seems to be different, in the Insurance industry analyst analysts seem to react most heavily on differences in 4 firm ratio. Moreover, confirming table 6, when returns become more volatile analyst will follow a firm less. Again according to Table 2, it does not indicate that a higher overall mean results in a stronger relationship in an industry.
4.4 Market reaction following on the announcement of an analyst recommendation
The second hypothesis predicts that analyst reports are more valuable for investors when a firm is more complex.
Table 8 provides the means of the abnormal returns around the announcement of an analyst recommendation. In Panel A, the abnormal returns are divided by the I/B/E/S recommendation code, while in Panel B, they are divided by industry and recommendation code. Overall, in my sample analyst seem to give more positive than negative recommendations. Mostly, analyst give a hold recommendation, which is consistent with the descriptive statistics in Table 3. Confirming previous literature (Abarbanell, 1991; Bradshaw,
Table 7: Regression model on the relationship between firm complexity and analyst following separated by industry
Variable Banking Finance Finsvc Insure Invest Savloan
INTERCEPT -6.400** (-2.45) 4.460 (1.02) 0.174 (0.07) -1.955 (-0.46) 16.292*** (3.83) -11.603*** (5.11) INV4FIRMRATIO 0.271 (1.11) -0.263 (-0.29) 0.074 (0.43) 5.565*** (5.26) 2.969*** (13.59) 1.092*** (8.35) STDRETURNS -0.442*** (-4.24) -0.918*** (2.65) -0.313** (-2.02) -0.270** (-2.36) -0.329** (-2.45) -0.515*** (-2.92) LOGFIRMSIZE 1.948*** (6.74) 0.991** (2.27) 0.612 (1.64) -0.019 (-0.05) -2.039*** (-4.31) 1.526*** (4.41) SALESGROWTH -0.041*** (-4.47) -0.071*** (-4.88) 0.070*** (5.38) -0.057*** (-2.63) 0.078*** (4.35) 0.093*** (7.31) ADVERTISING 167.503*** (19.44) 346.554*** (12.15) 21.591** (2.15) -57.987*** (-5.61) -75.965*** (-6.55) 221.006*** (-5.11)
Fixed Effects Yes Yes Yes Yes Yes Yes
Observations 1,391 257 219 95 416 328
R-squared 0.30 0.73 0.26 0.40 0.49 0.60
F-statistic 116.45 132.92 14.67 10.82 77.01 92.97
Prob 0.000 0.000 0.000 0.000 0.000 0.000
*, **, *** p<0.1; p<0.05; p<0.01, respectively.
This table presents coefficient estimates and t-statistics from the regression of number of analyst on complexity variables and control variables separated by industry. The Multi-Finance industry is excluded because of the limited amount of data in this industry.
Drake and Myers, 2012; Souček and Wasserek, 2014), there is a positive abnormal return when a “Strong buy” or “Buy” recommendation is given while a “Hold” , “Underperform” or “Sell” recommendation results in a negative abnormal return.
Following the results from Table 8, the largest reaction on the market can be seen on the announcement date of analyst reports. However, abnormal returns provided in Table 7 are small and insignificant11. Therefore, the analyst informativeness measure constructed by Frankel et al. (2006) is used to test the second hypothesis. To control for other influencing variables, the following regression is estimated:
𝐴𝐼, = 𝛽0+ 𝛽1𝐼𝑁𝑉4𝐹𝐼𝑅𝑀𝑅𝐴𝑇𝐼𝑂,+ 𝛽2𝐿𝑂𝐺𝑆𝐸𝐺𝑀𝐸𝑁𝑇𝑆,+ 𝛽3𝑆𝑇𝐷𝑅𝐸𝑇𝑈𝑅𝑁𝑆,+ 𝛽4𝐿𝑂𝐺𝐹𝐼𝑅𝑀𝑆𝐼𝑍𝐸, + 𝛽5𝑆𝐴𝐿𝐸𝑆𝐺𝑅𝑂𝑊𝑇𝐻,+ 𝛽6𝐴𝐷𝑉𝐸𝑅𝑇𝐼𝑆𝐼𝑁𝐺,+ 𝜀, (4)
Similar to regression (3), this regression is estimated using an ordinary least-squares regression with fixed or random effects.
Table 9 presents the findings of the regression12. t- and z-statistics presented in brackets, are subject to firm level clustered standard errors. Coefficients are mostly significant, however all coefficients are very modest. This can be explained due to the low value of AI which has a mean of 0.004. The coefficient of the inverse of the 4 firm ratio is significant and positive (model 1) which indicates that higher heterogeneity, therefore a higher complexity of the industry, will result in a higher value of the analyst recommendation because private information is more valuable. The same applies to the coefficient for the number of segments a firm operates in (model 2), suggesting that when a firm operates in more different sectors, the informativeness of an analyst recommendation is higher because the disclosed information is more complex. Both evidences are in accordance with the expectation that investors regard information of more complex firms as harder to analyse and therefore, value analyst recommendations on these firms higher.
Similar to the results for Hypothesis 1, my measure for the standard deviation of returns of the previous month before the recommendation gives a significant and slightly negative coefficient (model 3). This indicates that the informativeness of analyst reports is decreasing while a firm’s stock price fluctuates more in the prior month before the recommendation. Concluding, the complexity measures 4FIRMRATIO and SEGMENTS suggest that the
11 Significance is tested by dividing the abnormal return on the announcement day by the standard deviation of
the abnormal return (MacKinlay, 1997).
12 Means of AI separated by the recommendation code do not differ from each other. Therefore, the regression
Table 8: Descriptive statistics for Abnormal returns Panel A: Abnormal returns by recommendation
Year Count T=-1 T=0 T=1 T=2 T=3 Strong buy 1021 .0011 .0137 .0035 .0012 -.0005 Buy 1507 -.0002 .0097 .0022 .0004 .0011 Hold 3255 -.0021 -.0062 -.0020 -.0002 .0001 Underperform 555 -.0017 -.0206 -.0083 -.0026 -.0007 Sell 216 .0029 -.0316 -.0086 -.0013 .0036 Total 6554 -.0010 -.0015 -.0009 -.0001 .0003
Panel B: Abnormal returns by industry and recommendation
Industry (SIG) Recommendation Count T=-1 T=0 T=1 T=2 T=3
Banking Strong buy 302 .0056 .0148 .0016 .0024 -.0010 Buy 481 .0000 .0123 .0034 .0004 .0018 Hold 913 -.0000 -.0088 -.0015 .0005 .0001 Underperform 128 -.0028 -.0279 -.0057 -.0016 -.0005 Sell 32 .0175 -.0340 -.0064 -.0029 .0042 Savings and Loan Strong buy 81 .0026 .0135 .0006 -.0012 -.0005 Buy 94 .0005 .0121 .0048 -.0009 .0030 Hold 280 -.0058 -.0061 -.0043 .0050 -.0005 Underperform 32 -.0012 -.0201 -.0064 .0024 .0020 Sell 23 -.0024 -.0456 -.0161 .0142 .0018 Insurance Strong buy 40 .0045 .0127 .0078 -.0053 .0007 Buy 60 -.0055 .0198 -.0006 .0014 -.0039 Hold 148 -.0030 -.0067 -.0051 .0023 -.0020 Underperform 20 -.0031 -.0268 -.0058 -.0057 .0056 Sell 11 -.0086 -.0434 -.0155 .0205 -.0051 Finance and Loan Strong buy 70 -.0045 .0175 .0039 -.0005 -.0003 Buy 89 -.0028 .0056 .0040 .0013 -.0003 Hold 208 .0026 .0026 -.0031 -.0021 -.0012 Underperform 25 .0019 -.0220 -.0172 -.0148 .0002 Sell 19 -.0088 -.0085 -.0092 -.0032 -.0005
Multi-Industry Finance Strong buy 19 -.0014 .0050 -.0029 .0026 .0004 Buy 12 .0022 -.0026 .0015 -.0029 .0012 Hold 15 .0065 -.0087 .0061 .0055 -.0019 Underperform 1 -.0038 .0005 -.0025 .0058 -.0077
Sell - - - -
Financial services Strong buy 112 -.0063 .0199 .0058 .0066 .0014 Buy 186 -.0015 .0114 .0001 .0016 .0027 Hold 345 -.0044 -.0112 -.0019 -.0047 -.0010 Underperform 84 -.0011 -.0264 -.0029 -.0017 .0032 Sell 34 .0227 -.0445 -.0094 -.0106 -.0044 Investments Strong buy 397 .0002 .0110 .0048 .0001 -.0008 Buy 585 .0007 .0065 .0016 -.0000 .0003 Hold 1,346 -.0029 -.0045 -.0015 -.0005 .0009 Underperform 255 -.0018 -.0145 -.0106 -.0020 -.0031 Sell 97 -.0040 -.0261 -.0065 -.0032 .0084
*, **, *** p<0.1; p<0.05; p<0.01, respectively.
This table presents the descriptive statistics of the abnormal returns. Panel A divides the abnormal returns by recommendation code. Panel B divides the abnormal returns by both the recommendation code and by industry.
Table 9: Regression model on the relationship between firm complexity and analyst informativeness (Hypothesis 2)
Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10
INTERCEPT 8.10e-3*** (35.92) 5.19e-3*** (40.07) 8.25e-3*** (37.59) 7.88e-3*** (35.21) 4.72e-3*** (33.54) 8.15e-3*** (21.08) 4.61e-3*** (23.60) 8.04e-3*** (21.33) 7.82e-3*** (20.52) 4.35e-3*** (21.63) INV4FIRMRATIO 1.60e-4*** (9.46) 1.29e-4*** (7.60) 8.68e-5*** (6.77) 1.25e-4*** (5.04) 8.62e-5*** (3.49) 5.59e-5*** (2.75) LOGSEGMENTS 3.00e-4*** (4.35) 4.09e-4*** (5.74) 4.59e-4*** (3.88) 4.33e-4*** (3.70) STDRETURNS -1.44e-4*** (-12.74) -1.30e-4*** (-11.41) -1.41e-4*** (-12.50) -1.97e-4*** (-10.24) -1.86e-4*** (-9.56) -2.21e-4*** (-10.93) LOGFIRMSIZE -6.34e-4*** (-22.51) -2.34e-4*** (-13.31) -5.21e-4*** (-18.63) -5.66e-4*** (-19.89) -2.29e-4*** (-12.50) -6.62e-4*** (-14.18) -2.51e-4*** (-9.15) -5.60e-4*** (-14.04) -5.77e-4*** (-12.36) -2.10e-4*** (-7.59) SALESGROWTH -1.42e-7 (-0.54) -1.18e-7 (-0.45) -3.75e-8 (-0.14) -7.46e-8 (-0.29) -9.27e-8 (-0.36) 7.28e-6*** (4.97) 3.08e-6** (2.19) 8.12e-6*** (5.67) 7.60e-6*** (5.28) 3.33e-6** (2.42) ADVERTISING 8.15e-3*** (21.08) 1.36e-2*** (23.60) 1.69e-2 (11.39) 7.83e-3*** (20.52) 1.24e-3*** (9.87)
Fixed Effects Yes No Yes Yes No Yes No Yes Yes No
Random Effects No Yes No No Yes No Yes No No Yes
Observations 5,943 5,943 5,943 5,943 5,943 2,713 2,713 2,713 2,713 2,713
R-squared 0.09 0.07 0.10 0.11 0.09 0.12 0.012 0.16 0.16 0.15
AIC -68,359.3 -67,493.8 -68,432.9 -68,490.4 -67,716.8 -31,699.4 -31,239.0 -31,780.1 -31,790.8 -31,363.39
F-statistic 179.00 205.10 169.80 102.35 125.14 102.98
Prob(F-statistic/Chi^2) 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 *, **, *** p<0.1; p<0.05; p<0.01, respectively.
This table presents coefficient estimates and t-statistics from the regression of analyst informativeness on complexity variables and control variables. In model 6, 7, 8, 9 and 10 the control variable ADVERTISING is included. Therefore, there are less observations than in the first 5 models.
informativeness of analyst reports is increasing when the complexity of a firm increases. Contradicting, the significant negative coefficient for STDRETURNS indicates the reverse.
All results stay the same when the three measures of complexity are estimated in one regression (model 4 and model 5). The variable that controls for the size of the firm gives a slightly negative coefficient, which is in contradiction with the coefficient for firm size in hypothesis one. The coefficient for growth is not significant. However, when I also control for advertising expenses (model 6, 7, 8, 9 and 10) the coefficient for growth changes to a positive coefficient. This confirms the results of Lehavy et al. (2011). Other results stay the same after controlling with the advertising variable.
Furthermore, when comparing the different models with each other based on the AIC, model 1 to 5 have a better fit than model 6 to 10. Therefore, when testing this hypothesis it is better to not include the advertising control variable. In addition, model 4 has the lowest AIC which indicates that combining the 4 firm ratio and the standard deviation of returns gives the best model.
Similar to hypothesis 1, I estimate the same regression for the different industries to make claims about the differences among the industries. Again the null hypotheses that all industries react the same is rejected, therefore there are differences between the industries in the financial sector. Table 10 presents the results of the regressions for each industry, with t-statistic presented in brackets. Similar to Table 9 the coefficients are very small because of the size of AI. Following the coefficients for the 4 firm ratio, the analyst informativeness reacts strongest on complexity in the Finance and Loan industry. Additionally, all coefficients are positive and significant in accordance with Table 9. Moreover, with the last measure for complexity (STDRETURNS) I find differences in results. Financial services has a positive coefficient, however it is not significant13. The other industries react in accordance with the
results presented in Table 9, with the Insurance industry as most negative coefficient in predicting the analyst informativeness.
5. Conclusion and Discussion
As stated in the introduction, this paper connects the complexity of firms to the behaviour of analysts and the influence of their reports. Specifically, this study tried to find an answer to the
13 The Multi-Industry Finance industry has a positive and significant coefficient as well. However, due to the
Table 10: Regression model on the relationship between firm complexity and analyst informativeness separated by industry
Variable Banking Finance Finsvc Insure Invest Multfinl Savloan
INTERCEPT 5.41e-3*** (9.61) 8.94-3*** (8.03) 8.23e-3*** (12.42) 1.12e-3 (0.59) 7.80e-3*** (24.33) 5.36e-3*** (7.93) 8.23e-5*** (9.91) INV4FIRMRATIO 3.00e-4*** (7.61) 1.66-4** (2.43) 4.88e-4*** (7.39) 7.40e-4** (2.06) 1.54e-4*** (5.93) 1.66e-4** (2.38) -6.20e-5* (-1.73) STDRETURNS -1.41-4*** (-5.69) -2.69e-4** (-2.51) 2.43e-5 (0.76) -3.76e-4*** (-5.26) -1.60e-4*** (-10.97) 4.86e-4** (2.29) -3.32e-4*** (-5.04) LOGFIRMSIZE -2.66-4*** (-4.13) -8.53e-4** (-7.17) -1.04e-4*** (-10.37) 6.58e-5 (0.38) -5.84e-4*** (-14.44) -4.16e-4*** (-3.47) -5.28e-4** (-4.38) SALESGROWTH 8.10e-6 (9.61) -1.99e-6*** (8.03) -1.39e-6*** (-3.02) 2.50e-5*** (4.32) 8.21e-6*** (6.10) -1.30e-7 (-0.60) -3.65e-6 (-0.78)
Fixed Effects Yes Yes Yes Yes Yes Yes Yes
Observations 1,663 382 687 252 2,456 47 456
R-squared 0.09 0.24 0.15 0.19 0.15 0.34 0.12
F-statistic 40.35 29.61 29.09 13.60 108.41 4.56 15.46
Prob 0.000 0.000 0.000 0.000 0.000 0.000 0.000
*, **, *** p<0.1; p<0.05; p<0.01, respectively.
This table presents coefficient estimates and t-statistics from the regression of analyst informativeness on complexity variables and control variables separated by industry.
