Does CSR performance affect analyst forecast accuracy?
Master thesis 2017/2018
Name: Thijs Pronk
Student number: 10198571
Thesis supervisor:dhr. dr. W.H.P. Janssen Date: May 26, 2018
Word count: 11810
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Thijs Pronk who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This study examines the effect of corporate social performance of firms on forecast accuracy of financial analysts. The purpose is to test whether the social performance influences the forecasts of financial analysts. By performing a panel regression it is tested whether this effect actually takes place. It is hypothesized to have a negative relationship between CSP and accuracy, which means when CSP is increasing that the error margin of analysts is decreasing. This expectation is based on different theories, which are outlined in the literature and hypothesis. The results show that there is a significant negative effect of CSP on forecast accuracy, this result holds in different checks.
Keywords: Accuracy, Analyst Forecast Accuracy, Corporate Social Performance, Corporate Social Responsibility, Financial Analysts.
Acknowledgement
I would like to thank my thesis supervisordr. W.H.P. Janssen of the University of Amsterdam for helping me during the thesis process. He allowed me to make this my own research, while also providing me guidance through the whole process and feedback on questions I had or things I could do to improve my thesis. I also want to thank my family for the support I received during this writing process.
Table of Contents
1. Introduction ...7
2. Literature review ... 10
2.1 Financial analysts ... 10
2.2 Forecast accuracy ... 12
2.3 Corporate Social Responsibility ... 13
2.4 Corporate Social Performance ... 15
2.5 Stakeholder theory ... 17 3. Hypothesis development... 18 4. Research method ... 20 4.1 Sample selection... 20 4.2 Dependent variable ... 21 4.3 Independent variable ... 21 4.4 Control variables ... 22 4.5 Statistical model ... 25 4.6 Sample mutations ... 25 5. Results ... 26 5.1 Descriptive statistics ... 26 5.2 Additional analysis ... 29 5.3 Regression results... 31 5.4 Robustness tests ... 33
6. Discussion and conclusion ... 35
6.1 Discussion ... 35
6.2 Limitations ... 36
6.3 Future research ... 36
6.4 Conclusion ... 37
List of tables
Table 1: Sample history ... 25
Table 2: Descriptive statistics ... 26
Table 3: Mean per year ... 27
Table 4: Percentiles ... 28
Table 5: Industry overview ... 28
Table 6: Correlation matrix ... 29
Table 7: Variance inflation factor ... 30
Table 8: Hausman test ... 31
Table 9: Regression results ... 31
Table 10: Regression with loss obs. only ... 33
Table 11: Regression with profit obs. only ... 34
Table of figures
Figure 1: Overview of percentile table (ACC) ... 281. Introduction
What is known as “dieselgate” was the scandal where software had been installed in diesel cars that could detect when cars were being tested (Hotten, 2015). When the cars were tested, the software could manipulate the settings of the car in such way that the test result showed a much better emission compared to what the actual value should have been without the software. This was cheating by Volkswagen, although eventually, it came out that many other car manufacturers did the same thing with their diesel cars. After this incident, Volkswagen sales declined a lot and many cars had to be replaced, which caused high costs for Volkswagen (Hotten, 2015). This learns companies that getting caught for bad social issues leads to high costs and reputation damage. To control for these costs and prevent reputation damage, companies are focusing more on sustainability. Sustainability is the term for the concern with “well-being of future generations and the concern of
irreplaceable natural resources” (Kuhlman & Farrington, 2010, p. 3436). Addressing social issues is becoming important for companies since the more direct involvement of society with companies because of social media and current interests. Nowadays more people are holding companies accountable for their activities and objectives.
In addition to the environment, many other areas are important in doing sustainable business what is demanded by investors and stakeholders. These social activities for doing sustainable business are called Corporate Social Responsibility (CSR). The base principle of CSR is the balance between social, environmental and economic issues (Kuhlman &
Farrington, 2010). These topics are also called the 3 P’s of sustainability: People, Planet and Profit. The goal of CSR is that companies are not simply to make a profit, but to be of service to society through its actions and make a profit (Williamson, 2008).
Currently, CSR is becoming more important for companies and especially for big companies (Hahn & Kühnen, 2013), because these type of companies can be more easily held to account and face more pressure from society. Besides that, bigger companies have lower marginal costs compared to smaller companies to report for non-financial
information. For example, Kolk (2003) found a significant increase in sustainability reporting by multinationals.
To be accountable, companies are publishing more non-financial information besides their financial statements, with for instance integrated reports (Hahn & Kühnen, 2013) or
standalone CSR reports (Dhaliwal et al., 2012). Non-financial information is also demanded and used by many different stakeholders. For this research, the focus will be on one specific stakeholder: financial analysts.
Financial analysts are processing relevant information about companies to make earnings forecasts, these forecast reports provide recommendations to other stakeholders. Analysts monitor the financial situation of firms and incorporate as much information into their reports, often resulting in a buy, hold or sell advice. The information analysts use were traditionally more financial orientated, however, because there is more non-financial information available nowadays, non-financial information is also taken into account (Previts, Bricker, Robinson & Young, 1994; Orens & Lybaert, 2007).
Because analysts have increased attention whether companies are performing socially responsible (Luo, Wang, Raithel, & Zheng, 2015), this research will focus on the relationship between CSR performance and analyst forecast accuracy. The goal is to measure whether social performance of companies influences the predictions of financial analysts. This is done by testing company level CSR performance scores on analyst forecast accuracy. Therefore, the research question is: Does CSR performance affect analyst forecast accuracy?
The paper of Ghoul et al. (2011) examines whether CSR performance has an effect on the cost of equity with regards to stock prices and analysts’ earnings forecasts. High
performing CSR firms should have a lower cost of equity capital compared to lower performing CSR firms because low CSR firms have a reduced investor base and higher perceived risk according to Ghoul et al. (2011). So if the research question of this research is answered with yes, it’s interesting for analysts to know that they should adjust their
calculated cost of capital for CSR performance. For companies, it’s interesting to know when they conduct CSR whether their perceived risk is incorporated in analyst reports.
Dhaliwal et al. (2012) made a distinction between companies who issued a separate CSR report besides their financial statements and companies without. CSR reports were used as a proxy for disclosure of non-financial information. They found an association between the issuance of a CSR report and lower forecast errors or better accuracy. Also, companies with CSR disclosure have a higher CSR performance (Dhaliwal et al., 2012; Mahoney et al., 2013). This research tries to link whether higher CSR performance leads to more accurate analyst forecasts.
Because CSR data is hard to quantify in a valuation of a company, CSR often isn’t present in the analyst reports. According to Orens & Lybaert (2007), there is no mentioning in the frequency or quantity of non-financial information in analyst reports. As a
recommendation for future research, they state that it would be an interesting topic to examine whether information that analysts use, but which not appears in their reports, has an influence on forecast accuracy. The data I’m going to use for CSR performance is publicly available, so analysts are able to use the information. Though, most analysts won’t use it in their reports because it’s hard to be interpreted. This research is trying to give a meaning to how analysts could interpret CSR performance.
On the one hand, this research will benefit users of analysts’ reports, because if the results show that CSR performance is a good indicator for lower accuracy errors. Then users of analyst reports know when a company is performing well in CSR, the earnings can be predicted more accurately on average. On the other hand, this will benefit analysts, they know whether they can or can’t rely on CSR performance information when estimating earnings.
The result from this research is that CSR performance does indeed lead to lower forecasting error, resulting in better accuracy for analysts. Although CSR performance is significantly related to forecast accuracy, the effect of CSR performance on the total variance is limited. This is explained by the relatively low explained variance.
By means of quantitative empirical research, it is tested whether the effect of company CSR performance influences the accuracy of financial analysts. Section two discusses the relevant literature and theories. In the third section, the hypotheses will be developed using relevant theories. Subsequently, section four contains the research method, sample
selection, research design and variables explanation. After that, section five is about the results. Subsequently, in section six the discussion, limitations, future research and conclusion are described. Finally, section seven contains the reference list.
2. Literature review
The following sections will outline the concepts of Financial analysts (§2.1), Forecast
accuracy (§2.2), Corporate Social Responsibility (§2.3), Corporate Social Performance (§2.4) and Stakeholder theory (§2.5).
2.1 Financial analysts
Analysts provide an unbiased expectation of how a firm will perform in the future, therefore analysts are information intermediaries by acting between companies and the information users of that company. Financial analysts use both public and private
information to make an informed recommendation towards the financial world whether to buy, hold or sell the relevant stock (Brown et al., 2015). The vision that financial analysts have a crucial role in the public market as information intermediaries is broadly backed by literature (Gleason & Lee, 2003; Frankel et al., 2006; Ioannou & Serafeim, 2010; Luo et al., 2015).
Analysts add value to the capital market according to Healy & Palepu (2001). Analyst reduce the information asymmetry between firms and financial stakeholders, because the information financial analysts produce is seen as verified and objective. Another argument why financial analysts have this authority comes from the research of Kasznik & McNichols (2002). The researchers found that the market assigns a higher value to firms that meet expectations consistently. The market valuates a premium in the share price for meeting the analyst forecasts, therefore it’s important for companies to meet the forecasts. This means that financial analysts have a significant impact on the financial markets and stock prices of companies (Healy & Palepu, 2001; Kasznik & McNichols, 2002).
The methods financial analysts use to make their forecasts differ to what kind of information they use and produce. Block (1999) conducted a survey under financial analysts to measure which methods analysts use to make their forecasts. Only 54,3% used present value techniques, Block thinks that future cash flows are hard to estimate and the cost of capital is hard to measure without inside company information. The survey also found that future stock price is hard to measure. And lastly, 287 out of 295 analysts were neutral or negative towards the efficient market hypothesis, which implies that analysts don’t rely very heavily on current stock prices. Although analysts make predictions for the market and some
of their calculations are based on market values, analysts don’t believe that the market is always efficient (Block, 1999).
So, on the one hand, analysts don’t think there is an efficient market. On the other hand, they contribute to an efficient market. Because analysts’ forecasts are used by the shareholders of the company and shareholders do find to have better information value when it’s provided by analysts (Healy & Palepu, 2001; Gleason & Lee, 2003). Shareholders are informed with verified information about firm performance and earnings expectations, which reduces the information asymmetry between the company and users of the
information. The company itself also profits from the reduction of information asymmetry, as this leads to a lower risk for capital providers and therefore reducing the cost of capital.
2.2 Forecast accuracy
Analyst forecast accuracy is the term for the error margin of a forecast made by an analyst. Accuracy is the difference between the real value and the predicted value of the earnings forecast, therefore it’s also called the forecast error. For analysts it’s important to have the best accuracy or lowest forecast error as possible, that would mean that they have done their jobs excellent and provide punctual predictions to their information users.
Important aspects for an analyst to produce accurate forecasts are reliable numbers, personal contact with management of the forecasted company (Brown et al., 2015) and industry knowledge (Healy & Palepu, 2001; Brown et al. 2015). To know how well a company is performing, it’s important for analysts to know how comparable companies perform and what the industry characteristics are (Brown et al., 2015). It’s also important for analysts to measure that earnings are backed by cash flow and whether the cash flow is repeatable and sustainable for the future. To gather additional information from the company, most
analysts have contact with the CEO or CFO of the company which they forecast for (Brown et al., 2015).
Forecast accuracy is improved when companies have more informative disclosures. Being more informative gives a higher chance of more analyst following (Healy & Palepu, 2001). More following creates less dispersion in analyst forecasts and less volatility in forecast revisions. With more analysts following, companies will have less volatile earnings predictions and a better analyst accuracy (Alford & Berger, 1999). This is explained because more following creates more insights into how a firm works and what the prospects are. With this in mind, big stock exchanges will have the least amount of errors according to Alford and Berger (1999), because these exchanges have the most analyst following.
Another factor that improves forecast accuracy is regulation and the enforcement of the regulation (Hope, 2003). Enforcement pushes managers to follow the rules and
therefore create a strong guideline which everyone needs to follow. With strong
enforcement regulation, there is not much space to deviate from the rules. Therefore it’s easier for analysts to forecast for these companies because of a better comparability (Barth, Landsman, Lang & Williams, 2012). US GAAP is used for accounting regulation in the USA, Europe mostly uses IFRS. While US GAAP relies more on rules, IFRS relies more on principles (Barth et al., 2012). Because rules enforce more than principles, it’s logical that forecast accuracy is better for US GAAP analysts compared to IFRS analysts.
2.3 Corporate Social Responsibility
CSR is a broad concept and therefore it’s not really possible to summarize it in a single definition. According to Carroll (1979), Corporate Social Responsibility can only be described if it consists the following four obligations companies have towards society:
1. Economic expectations to deliver services or products society needs and companies which make profits.
2. Legal responsibilities in obeying the law and regulations.
3. Ethical expectations to follow norms and values society expects from companies. 4. Discretionary expectations to exceed current ethical expectations from society. These four terms are ordered on importance from 1 being the most important to 4 being the least important, which is intuitive, since making a profit is the core objective of a commercial company.
In addition to the four obligations, McWilliams et al. (2006) state that CSR consists of the actions of companies that contribute to social welfare, more than what is necessary for profit maximization or enforcement of the law. This is supported by Williamson (2008), they state that the goal of CSR within companies is to be of service for society and make a profit. Both definitions agree with Carroll that economic is the most important factor and society’s expectations should also be included. CSR activities consist of, among other things, social characteristics or functions in products and production processes, progressive views in the personnel policy, working on a better environment or co-operating with the objectives of civil society organizations (McWilliams et al., 2006).
Garriga and Melé (2004) add to economic expectations that some companies use CSR as an instrument to achieve profits. There are three objectives why companies are doing that: maximizing shareholder value, gaining competitive advantage and using CSR as a marketing tool. This is in line with traditional thinker Friedman (1970) who stated that “the social responsibility of business is to increase its profits”. Friedman argues that seeking for profit through CSR is short-term oriented because a company only thinks about the
economic expectations.
Friedman’s statement can be seen as traditional thinking because that way of thinking is mainly from the past. In the past companies did get lower recommendations by analysts, because CSR used to be seen as value destructing (Ioannou & Serafeim, 2010). Although eventually, CSR is more perceived to contribute to long-term value creation.
Therefore companies which are active with CSR and sustainability trigger more positive recommendations by analysts nowadays.
In addition to the economic expectations, Porter & Kramer (2006) find out that CSR programs are mostly focused on the reputation of companies. They also argue that CSR has a limited connection to activities of companies, therefore justification of CSR will be hard to carry on in the long-term. Schaltegger et al. (2012) add to this that current established business models are hard to adapt because companies that already perform well already have met the economic expectations. This is only reachable when companies get more proactive in sustainability at all levels of the firm. The focus on profit is contradicting the goal of CSR, but the problem is that short-term gains tend to have a higher perceived value
compared to long-term sustainable gains (Dyllick & Hockerts, 2002). Because time is valuable due to the discount rate, long-term social and environmental costs are being neglected too much. Therefore it’s important that firms also maintain the social and environmental capital in order to maintain the needs in the future.
Maximizing long-term shareholder value is the goal of commercial companies, but short-term value creation regularly has more priority than the long-term due to two factors (Rappaport, 2005). Firstly, to be able to meet short-term expectations from for instance financial analysts, management sometimes ignores value-creating investments. Secondly, managers try to use accounting to push revenues from the current year at the cost of the next year, eventually leading to not meeting expectations in the long run (Rappaport, 2005). In contradiction to the short term economics: legal, ethical and discretionary obligations are more long-term oriented, by doing well for society.
CSR can be used for long-term value creation, therefore it will always be the question whether the company is doing this for a long-term benefit for the society or whether the company wants to have long-term value creation by doing CSR. The fact is that CSR is more and more demanded by society and companies have to be increasingly active in social activity, therefore CSR will benefit both society and business in the long run. Porter & Kramer (2006) indicate that, thanks to their existence, companies generate prosperity in both economic and social terms, so that there is already a positive effect of companies on society.
2.4 Corporate Social Performance
Corporate Social Performance (CSP) is the concept which measures how a company is performing on social activities. According to Wood (1991), CSP is the social achievement from doing CSR activities, this is in accordance with Ioannou & Serafeim (2012). They argue that a company first needs to have a certain amount of CSR activities before a company is involved in social performance.
In order to get a reliable measure of CSP, many factors have to be included from different categories, such as the four obligations from Carror example. The research of Wartick & Cochran (1985) adds to this that CSP needs to address “principles of social
responsibility, process of social responsiveness and policies of issues management” (p. 767). When these requirements have been met, there is an integrated perception of corporate social involvement with their stakeholders.
There are ways within companies to use CSP to estimate how much a company contributes to social objectives through CSR activities. Although it’s hard to check how a company is performing socially, there are some external agencies which are measuring many metrics to get performance scores. Agencies like Thomson Reuters ESG, Bloomberg and KLD. How a company is ranked on the CSP index could lead to influences on consumer
preferences, analysts, investors and organizations themselves (Ioannou & Serafeim, 2012). Therefore it’s important for stakeholders and especially shareholders to perform well on CSR. Although this is a good way to measure companies’ CSR practices, it possibly takes away the proactive side of CSR from companies. Because most companies may only want to score high on the CSP indexes and don’t integrate CSR activities in their business models like Schaltegger et al. (2012) advocates companies to do. On the other hand, a lot of companies aren’t proactive at all (Schaltegger et al., 2012), therefore it’s better to activate companies rather than doing nothing.
Researchers found out that CSP does provide corporate financial profit (Waddock & Graves, 1997; Margolis et al., 2009; Harjoto & Jo, 2011; Barnett & Salomon, 2012). Although, the research of Margolis et al. (2009) state that the overall research done in the past
demonstrates a small but significant effect in the relation between CSP and profit.
Companies aren’t rewarded a lot, that’s partly because it’s hard to quantify how much CSR really connects to profit. Nevertheless, in the case of a negative event regarding CSR, the penalty will be higher compared to positive events. This is also explained by Sen &
Bhattacharya (2001), who found out that customers more heavily react to negative CSR information compared to positive information.
In addition to that, McWilliams & Siegel (2001) argue that it’s more a question of supply and demand. Firms are expected to have a certain level of CSR activities and these firms should comply with those levels. The researchers also reason that, in order to reach profit maximization, managers should make decisions in CSR activities the same as they do it for all decisions. With these arguments, the authors suggest that there is not a real
connection from CSP towards corporate financial performance. Although there are more researchers that don’t find a connection Margolis et al. (2009) did a meta-analysis and found a significant small effect of CSP on profit, therefore it’s realistic to assume that there is an effect of CSP on profit.
There are several reasons why CSR might has an impact on analysts’ forecasts
according to Ioannou & Serafeim (2010). First, CSR affects long-term financial performance by creating or destroying value for a lot of stakeholders. These value changes are being incorporated by analysts directly if known. The second reason is that mutual funds tend to invest more and more into socially responsible firms and thus creating demand for CSR. The money invested creates higher stock prices and therefore analysts expect higher earnings from these companies. Lastly, agencies like KLD and Thomson Reuters rank companies on their CSR programs, this leads to growing demand by investors that these companies have a minimum level of social engagement.
If shareholders attach value to CSR programs from companies, share prices are influenced by CSR information and CSR performance. Analysts need to incorporate this and therefore need to know how a company is performing in CSP. Higher CSP scores do lead to better performance evaluations by analysts compared to lower scoring companies (Ioannou & Serafeim, 2012). The impact of these higher scores on forecast accuracy will be
2.5 Stakeholder theory
Stakeholders are people and groups of people that don’t necessarily have economic interests in a company but do have interests in the activities that the company undertakes. Stakeholders can be different interest groups, such as NGO’s, consumers, employees, suppliers, shareholders, financial analysts and others. CSR is often related to stakeholder theory because stakeholders are demanding companies to take their social responsibility (McWilliams & Siegel, 2001).
The meaning of the stakeholder theory is that firms are responsible for all stakeholders (Carroll, 1991). Because there is more demand from stakeholders and more attention for stakeholder issues nowadays, firms are willing to spend more money on their business in social and sustainable activities like environment, employees and community. This theory can help to understand the needs of the stakeholders and what a company should do to meet this needs.
What is connected with stakeholder theory is the legitimacy theory. This theory offers a perspective on how much impact the stakeholders should have in the operational
management and what influence the stakeholders already have inside a company (Carroll, 1991). Legitimacy theory is defined as how attention should be divided around the different stakeholders. When management of a company makes decisions, stakeholders who are directly involved should play a more important role than indirectly involved stakeholders. By making a distinction between how important groups of stakeholders are, it is easier for companies to make decisions. This is because fewer interest groups have an influence on the company according to their legitimacy, which means that a consensus can be found more quickly among the most important stakeholders. A reason why the legitimacy theory is important is that ignorance of stakeholders can reduce long-run profits resulting in things like lower sales or being a less attractive employer.
Although for some companies the financial benefit of conducting CSR is limited (Barnett & Salomon, 2012), the goal of CSR isn’t just profit but to be of service to society through its actions and make a profit (Williamson, 2008). Besides that, there are spillovers from conducting CSR to for example customer loyalty (Sen & Bhattacharya, 2001), lower cost of capital (Ghoul et al., 2011) and employee happiness (Singhapakdi et al., 2015). Therefore companies and their stakeholders are able to benefit from CSR while it doesn’t directly show up in the financial statements.
3. Hypothesis development
According to Luo et al. (2015) analysts have a mediating role in the relationship between CSP and corporate financial performance. Analysts reduce the CSP information asymmetry by using CSP in their forecasts, therefore CSR information is used to make potentially improved estimates of a company. The research by Luo et al. (2015) didn’t investigate whether CSP actually led to better analyst results or better accuracy, but only focused on whether analysts use the CSP information and decrease the agency problems surrounded with corporate social performance. This research will attempt to address whether CSP leads to better accuracy.
There are two indirect relationships between CSR performance and analyst forecast accuracy. First, companies with higher CSR performance are better off when they make more publicity about their CSR program because firms with more visibility on their CSR program have better analyst expectations compared to firms without (Ioannou & Serafeim, 2010). So better performing CSR companies should disclose more CSR information, and more CSR disclosure leads to more accurate forecasts (Dhaliwal et al., 2012). Second, as
mentioned earlier, CSR performance on average leads to higher profits for companies. Therefore a company which has higher CSR performance has more chance of making a profit. Because analysts are better in predicting profit-making firms compared to loss-making firms, it’s intuitive to hypothesize that CSR performance has a positive relationship with forecast accuracy. These relationships lead to the following hypothesis:
Higher CSR performance leads to more accurate analyst forecasts.
In the past, CSR used to be seen as value destructive by analysts and CSR had a negative impact on investment recommendations (Ioannou & Serafeim, 2010). At present this is different, nowadays CSR performance has a positive impact on analyst
recommendations because CSR strengths are perceived as value creating. Besides that, CSP also impacts the financial performance positively. Since both financial performance and analyst buy recommendations go up because of CSR performance, it’s assumed that CSR performance leads to higher forecast accuracy.
In addition to that, companies which are making losses are harder to make estimates from (Ciccone, 2005). This is supported by Duru & Reeb (2002), they find that analyst’ predictions for loss-making firms are on average less accurate compared to predictions for firms that made a profit. Therefore, companies that are performing well are easier to estimate for analysts and companies that are performing well on CSR have more chance to have improved profits. Because of this, it’s expected to have a lower amount of errors when a company is scoring higher on the CSR performance index.
When the hypothesis is significant there is a positive effect of CSR performance on forecast accuracy, whereas the CSR performance results in lower forecast error. If the hypothesis isn’t significant, the hypothesis is rejected and no effect is recognized between CSR performance and forecast accuracy.
4. Research method
This section provides an overview of the research design and how the research will be conducted, including which variables are being used. Firstly, the data collection method and the sample will be discussed (§4.1). Secondly, the dependent (§4.2), independent (§4.3) and control variables (§4.4) are described. After that, the statistical model (§4.5) and the variable names will be treated. Finally, section §4.6 contains the sample mutations.
4.1 Sample selection
This study only uses the annual EPS predictions, because CSR performance is also measured yearly and therefore it’s less useful to utilize quarterly data. The timespan of the sample will be from 2010 up to 2015. The starting year will be 2010 because data can be influenced by the financial crisis from 2007. That did result in major stock market reactions and could have biased the data used (Jacob, Lys & Neale, 1999; Hutira, 2016). In order to diminish that, 2010 will be the starting year. 2015 will be the ending year because for some data it was only possible to gather it up to 2015. Because this is time series data, the dataset will be treated as panel data and different type of testing is needed compared to regular pooled OLS regression. There will be a choice between fixed effect and random effect regression for panel data (Bell & Jones, 2015), this will be treated in section 5.3 regression results.
Only American listed companies are being used, because these companies have the most analyst coverage and are represented in both the CSR performance index and in the IBES database. Because analysts provide verified information to the market, more analyst coverage means a better information and improved liquidity (Irvine, 2003). Therefore it will be suitable to only use companies which have at least been forecasted by 3 analysts because this will provide more accurate information.
In order to answer the research question, an empirical research will be conducted. The tests will be done in Stata, by using analyst accuracy (ACC) as the dependent variable, CSR performance (CSP) as independent variable and control variables: company size (SIZE), analyst following (AF), forecast horizon (FH), loss (LOSS) and panel variables (firm and time). To try to answer the question: Does CSR performance affect analyst forecast accuracy?
4.2 Dependent variable
Analyst accuracy. The data for analyst accuracy comes from the Institutional
Brokerage Estimate System (IBES), this data is published by Thomson Reuters. In the IBES database, there are lots of historical forecasts from financial analysts from large U.S. firms. Analyst accuracy is measured by the absolute value of actual Earnings Per Share (EPS) -/- the median forecast of the EPS predictions done in each month from all analysts divided by the stock price at the end of the year. The method of calculating accuracy is also called Absolute Forecast Error and is similar to Lang & Lundholm (1996), Hope (2003) & Dhaliwal et al. (2012). For clarification, this is the definition:
Analyst accuracy = |(median EPS - actual EPS) / Stock price|
Absolute value of forecast error is being used, as this research wants to measure whether analysts do deviate from the actual value. Therefore, it doesn’t matter whether this difference is positive or negative. This is in line with the common literature about financial analysts (Hope, 2003; Dhaliwal et al., 2012).
Predictions from different analysts in each month are taken into the calculation of the median. The median will be used in the accuracy equation, because the forecast bias is nonexistent nowadays when using median values (Hovakimian & Saenyasiri, 2010; Gu & Wu, 2003). While for using the mean, there is still a little bias (Hovakimian & Saenyasiri, 2010), therefore median is preferred above mean.
Lastly, this research will have one accuracy observation per month per firm. Because of this, there is time series data and panel regression is necessary, which will be discussed later.
4.3 Independent variable
CSR Performance (CSP). The data used for CSR performance is gained through
DataStream and is also published by Tomson Reuters. The data comes from the Thomson Reuters ESG database. This dataset has been developed by using more than 400 company level ESG measurements, resulting in 178 performance indicators on Environmental, Social and Governance determinants (Thomson Reuters ESG Scores Methodology, 2017). These determinants result in one score for each category and these combined provide the ESG score per company. The ESG score will be used as a proxy for CSR performance in this research and will be the independent variable.
The data from Thomson Reuters (2017) is comprehensive enough to use as a proxy for CSR performance, therefore, no other variables will be used that measure CSR performance. For clarification, this definition will be used as proxy for CSP:
CSP = Thomson Reuters ESG score / 100
4.4 Control variables
Firm Size. There will be controlled for firm size as described in McWilliams & Siegel
(2001), Behn, Choi & Kang (2008), Udayasankar (2008) and Ioannou & Serafeim (2012) because CSR participation can be influenced by the scale of the firm. Besides that, firm size also has an influence on the accuracy of analyst’ forecasts according to Brown (1998). The research states that larger firms tend to forewarn analysts more often when losses will occur, therefore larger firms can be estimated more accurately and timely by analysts. The natural logarithm of the total assets is used as a proxy for firm size in Gleason and Lee (2003) and Dhaliwal et al. (2012). Because really big companies could potentially influence the sample too much, the natural logarithm is used to provide an unbiased variable for firm size. The natural logarithm of the value of assets in U.S. Dollars at the end of the previous year will be used as proxy.
Analyst following. The number of analysts following a company has been researched
significantly within the financial analyst research field. This variable has been widely used as a control variable in different research papers (Alford & Berger, 1999; Hope, 2003; Behn et al., 2008; Dhaliwal et al., 2012; Hutira, 2016). Analyst following is calculated as the natural logarithm of the number of analysts which followed the company (Behn et al., 2008;
Dhaliwal et al., 2012). Analyst following also has been used in the calculation of the median value, each analyst counted as one observation and the median was taken from all these observations. Observations with lower than three analysts following were deleted, because at least three observations are needed to calculate the median.
Forecast horizon. The forecast horizon is the average number of days between the
forecast and the actual earnings announcement. The assumption is that the closer the forecast has been made to actual announcement, the more accurate the forecast will be. This assumption is intuitive because more information about the company and its
environment is known at the end of the year and positive or negative surprises could have occurred during the year. All the information during the year is incorporated by analysts,
therefore predictions should be more accurate at the end of the year (Behn et al., 2008). Forecast horizon has also been widely used by analyst accuracy literature (Behn et al., 2008; Dhaliwal et al., 2012; Hutira, 2016). This research uses the same proxy as Behn et al. (2008) for forecast horizon: the natural logarithm of the difference in days between the actual announcement date and the median forecast date. There is a need to control for the
forecast horizon in this research, because of different due dates and different dates in when a forecast has been done.
Loss. Another control variable is a loss occurrence dummy in the year which has been
forecasted for. This variable has an influence on analyst accuracy according to the research of Brown (1998), Duru & Reeb (2002), Hope (2003), Behn et al. (2008) and Dhaliwal et al. (2012). The dummy variable LOSS will have a value of 1 if the firm made a loss in the year that the prediction has been made for, in case of a profit the variable will have a value of 0. Controlling for loss lowers the bias of the sample size and reduces the chance for less accurate and too optimistic analyst’ forecasts (Duru & Reeb, 2002). Negative earnings are needed to control for because it’s harder to estimate for loss-making firms compared to firms with positive earnings. This implies that losses would lead to a lower forecast accuracy. The possible reason is that negative earnings are on average less common than positive earnings (Hwang, Jan & Basu, 1996). Therefore, analysts have less experience with losses. Besides that, analysts tend to be optimistic (Duru & Reeb, 2002) due to the commissions they could generate by issuing buy recommendations (Hwang, Jan & Basu, 1996).
For clarification, these definitions are being used for the control variables:
Firm Size = ln(asset value)
Analyst Following = ln(number of analysts in median EPS calculation)
Forecast Horizon = ln(number of days between predicted and actual EPS)
Loss = 0 if profit, 1 if loss
Size, analyst following, forecast horizon and loss are almost always taken into
account with analyst forecast accuracy testing, therefore these are used as control variables. Other variables may also have an effect on CSR performance or forecast accuracy, but those are not practical to take into account or for some variables it’s simply not possible to get the data.
With panel data regressions it’s necessary to give definition to a group variable and to a time series variable to distinguish the different panel observations from the same year or from the same company. To distinguish between the different groups within the panel, the company identifier is the group variable.
Because of the panel dataset a time series variable is also necessary to define the panel itself. Due to the fact that there has been chosen to have accuracy measured per month, it’s necessary to also have the time series based on each month. Therefore each month is set as a different value and in this way a distinction in time was made for the panel.
Although the relatively low number of control variables could be a limitation, control variables on the country level are not necessary for this research. The reason that the
country level variables are not necessary is that this research only uses data from companies in the United States. This means that there are no different countries in the sample and controlling for country-level differences are redundant. Examples of variables which aren’t relevant in this research are enforcement level, accounting methods and type of law (Hope, 2003; Dhaliwal et al. 2012). Because these variables aren’t necessary for this research, it’s expected to have the same explained variance as in comparable analyst forecast research.
Different checks will be necessary before the model can be tested in a multiple regression. Descriptive statistics are used in order to give an insightful overview of the sample. Besides that, it’s needed to make sure that observations don’t contain errors and variables aren’t correlated.
4.5 Statistical model This formula will be used:
H1: ACC = β0 + β1CSP + β2SIZE + β3AF + β4FH + β5LOSS +ε
ACC Forecast Accuracy = Absolute value of (actual EPS -/- predicted median EPS divided by actual EPS)
β Constant
CSP Corporate Social Performance (or CSR performance)
SIZE Natural logarithm of the total assets at previous year end (31st of December)
AF Analyst Following: the natural logarithm of the number of analysts who made their prediction and are included in the calculation of the median.
FH Forecast Horizon: the natural logarithm of the difference in days between the median forecast date and the actual EPS announcement date.
LOSS Dummy variable of loss: 1 if a loss occurred during the predicted year, 0 if profit
ε Error term of the regression model
4.6 Sample mutations Table 1: Sample history
Observations Obs. dropped Description
39,312 Master data: ACC, AF, FH & Loss variables from Excel
Merge with SIZE
2 Deleted 2 unmatched master data observations
Merge with CSP
2,299 Dropped the observations with only CSP data 3,190 Dropped the observations with CSP missing
38 Deleted CSP = 0 observations
248 Deleted duplicates
Merge with stock prices
78 Dropped the observations without stock price
33,457 Final number of observations
Firstly, the master data file was created with input about accuracy, analyst following and forecast horizon. Loss is a dummy variable which has been created by sorting on profit which has been given a value of 1 to each loss observation 0 when profit occurred. These variables have been merged with company size data, 2 observation didn’t match and were deleted. Thereafter, the master file has been merged with CSP and missing values, duplicates and errors have been removed. The removal of the observation which weren’t complete resulted in a drop of 5,575 observations. Subsequently, some observations were removed that did not have stock price information, which resulted in a final sample of 33,457 observations.
5. Results
In the following section, the results of the statistical tests are presented and explained. This includes an overview of the descriptive statistics (§5.1), additional analysis (§5.2) for
normality, the results of hypothesis testing (§5.3) and robustness checks (§5.4).
5.1 Descriptive statistics
This section attempts to make the raw data transparent and understandable. Some descriptive tables are analyzed and have the aim to make the data insightful.
Table 2: Descriptive statistics
Variable n Minimum Maximum Mean Std. Deviation
ACC 33,457 0 0.03 0.01 0.01 CSP 33,457 0.05 0.97 0.68 0.26 SIZE 33,457 5.54 14.76 9.61 1.38 AF 33,457 1.10 4.03 2.81 0.50 FH 33,457 0 5.92 4.92 0.93 LOSS 33,457 0 1 0.03 0.18
On a first note, the dependent variable ACC is winsorized at a p-value of 0.05, this is explained in section 5.2 additional analysis in assumption 3 Significant outliers.
The final sample contains 33,457 observations after all adjustments. All these
observations are from 489 groups, which means that there are 489 different companies in 6 different years each month. This means that the maximum amount of observations in this sample with these companies could have been 35,208, which means that there are a total of 1,751 gaps in this sample. Gaps in the meaning that some observations are missing in the sample from a certain company or some months from a company. These gaps occurred because of missing data, duplicates or contradictory observations as shown in 4.6 Sample mutations.
The dependent variable ACC has a mean of 0.005, so this is the average margin between an analyst forecast and the actual value divided by the stock price. The standard error of ACC is 0.007.
The average CSP by the sample companies is 0.683, meaning that on average U.S. listed companies score relatively high in social performance because 0.50 should be an average score. This is probably caused by the fact that this sample only contains big
companies, these companies have more funds to invest in CSR related projects and that causes better scores.
The average company size is 14,868 million in assets (= e9.6070), the average number
of Analysts Following a company is 17 (= e2.8144) and the average Forecast Horizon is 137 days
(= e4.9227).
Table 3: Mean per year
Year N ACC CSP SIZE AF FH LOSS Groups
2010 5,317 0.007 0.66 9.40 2.70 4.91 0.034 460 2011 5,477 0.005 0.67 9.50 2.79 4.92 0.026 473 2012 5,614 0.006 0.64 9.57 2.83 4.92 0.031 478 2013 5,612 0.005 0.66 9.66 2.86 4.93 0.031 481 2014 5,701 0.004 0.69 9.72 2.85 4.93 0.025 485 2015 5,736 0.004 0.77 9.77 2.85 4.93 0.052 485
The average per year shows that ACC decreased over time, meaning that stock prices went up, analyst forecasts improved or both. Besides that, 2013 to 2015 shows an increasing CSP, meaning that companies are performing better socially, especially in 2015. SIZE is growing consistently throughout the years, AF to a lesser extend as well.
The growing numbers of CSP, SIZE and AF could be an impact of the growing economy after the financial crisis in the years before 2010, which causes more CSR related investments, bigger companies and more analyst covering. Nevertheless, 2015 appears to be a difficult year, because more than twice as many losses have been reported in 2015 (5.23%) compared to 2014 (2.47%). The reason that there are more observations in later years comes from the fact that there were some companies without CSP score in 2010. Besides that, adjustments were also done in the research, which also has an impact on the
observations per year. The number of groups in the table shows how many unique
observations there are present in each year. The lower number of observations in the earlier years causes the lower number of groups.
To predict whether results could be significant percentiles are taken from the sample, which could be found in the table and figure below:
Table 4: Percentiles Figure 1: Overview of percentile table (ACC)
Figure 2: Overview of percentile table (CSP)
The table and figure provide insight into how CSP and ACC are related to each other. While ACC is increasing more each time, CSP increases less over time. This indicates that there is an inverse relation between CSP on ACC. SIZE, AF and FH are in natural logarithm, therefore the plot for these 3 variables should look like CSP. Meaning that the higher the percentile, the slower the variable increases. Loss shouldn’t be plotted, as it’s a dummy variable which becomes 1 at the 96.67th percentile.
Although industry can’t be used in this regression, it’s still useful to understand which industries the sampled companies are from. In the following table the frequency and
percentage of the total are given: Table 5: Industry overview
Type Frequency Percentage
Mining 1,994 5.96%
Construction 353 1.06%
Manufacturing 12,431 37.16%
Transport, communication,
electric, gas and sanitary 4,170 12.46%
Wholesale 499 1.49%
Retail trade 2,833 8.47%
Finance, insurance and
real estate 5,645 16.87% Services 3,409 10.19% Not specified 2,123 6.34% Total 33,457 100.00% Percentiles ACC. CSP. 1 0.00007 0.124 10 0.00022 0.278 25 0.00069 0.470 50 0.00204 0.767 75 0.00597 0.918 90 0.01564 0.955 99 0.02857 0.964
More than 37% comes from the manufacturing business, followed by Finance (16.9%), Transport etc. (12.5%) and Services (10.2%). The group not specified could have been caused by two reasons: companies don’t fit in the description of the corresponding categories or they aren’t classified in the Compustat database.
5.2 Additional analysis
To be able to conduct multiple regression, several conditions are being checked:
1. No multicollinearity
Multicollinearity occurs when there is a linear association between independent variables. If there is an issue, it means that two variables are heavily correlated and one of the two variables should be deleted from the test. There could be an indication of multicollinearity when R2 is high, this research has an R2of 0,11, therefore no indication of multicollinearity is
expected. The Pearson correlation matrix provides an indication whether independent variables have a linear association. There is multicollinearity when the correlation is above 0.9 and when the p-value is lower than 0.05.
Table 6: Correlation matrix
Variables ACC CSP SIZE AF FH LOSS
ACC 1 CSP -0.09** 1 SIZE 0.09** 0.35** 1 AF -0.13** 0.17** 0.27** 1 FH 0.22** -0.01** 0.00** 0.01** 1 LOSS 0.27** -0.12** -0.04** -0.08** -0.00 1 ** p<0.01
In this table most correlation numbers are significant at one percent, the highest correlation number is between LOSS and ACC. Although this is significant, the correlation of 0.35 is below the 0.9 threshold and is probably not a concern of multicollinearity. To test whether this statement is true, another method will be used. This method is called Variance Inflation Factor (VIF), there is no multicollinearity if the VIF tolerance (1/VIF) is below 10.
Table 7: Variance inflation factor
Variable VIF 1/VIF
CSP 1.16 0.86
SIZE 1.20 0.83
AF 1.09 0.92
FH 1.00 1.00
LOSS 1.02 0.98
As shown in the table VIF, the VIF Tolerance is always below 10. This means that there is no multicollinearity happening in this sample.
2. Significant outliers
Accuracy is capped by winsorizing this variable at the 5% at both sides of the extreme values. This is done because accuracy is the only variable with huge outliers which influences the dataset for a certain amount. With this method, all results above a certain percentage are capped at the value of the extreme at the percentage level chosen. In this case, every value of accuracy exceeding 0.0286 is capped at 0.0286. In the unadjusted sample, ACC had a mean of 0.08, a standard deviation of 0.05 and a maximum value of 3.46. With a maximum value which is more than 43 times higher than the mean, it shows that outliers are present in the sample. These extreme results implies that the dataset could be biased and therefore winsorizing is justified. For example, the company Peabody Energy Corporation (BTU ticker) had huge unexpected losses and profits during the sample period, this resulted in extreme outliers of up to 31 times the standard deviation (20/0,65 = 31). These observations did influence the dataset too much.
After winsorizing, ACC has a mean of 0.005, a standard deviation of 0.007 and a maximum value of 0.0286. The standard deviation dropped 86% from 0.05 to 0.007. In addition, the R2 increased from 0.02 to 0.11 due to the adjustment. These improvements
show that the outliers had a big impact on the total sample.
Other variables show no outliers, therefore no further measures are necessary. The reason that no other outliers are present is that natural logarithm calculations are used in AF, FH and SIZE. With using the natural logarithm extremes scores are already limited in how much the value can deviate from the mean. Besides that, LOSS is a dummy variable and CSP is an index number between 0 and 1. All these variables can’t be subject to outliers, because they are limited in which value they can have.
5.3 Regression results
There is a choice for Fixed Effects (FE) or Random Effects (RE) with panel data regression. A FE model only adds estimates for within model effects, while RE models also estimate the higher-level and external effects (Bell & Jones, 2015). A commonly used method to make a choice between the preferred models is the Hausman test (Bell & Jones, 2015). Under the null hypothesis there is no significant difference measured between FE and RE, therefore FE model testing is the best method when the Hausman test is significant. While RE model testing is the preferred method when the null hypothesis is rejected. The results are presented below in the table:
Table 8: Hausman test
Hausman Chi2 296
p-value <0.01
The null hypothesis is not rejected and therefore it’s assumed that the FE regression model is the preferred method to test for internal effects according to Bell & Jones (2015). Compared to the pooled OLS regression, the FE model adds that internal model effects are taken into account in the regression (Bell & Jones, 2015). Below the FE model output is presented in table 8:
Table 9: Regression results
Variables Effect of variable on ACC (Constant) 0.0188** (0.0013) CSP ( -0.0016** -(0.0003) SIZE (-0.0019** -(0.0011) AF -0.0014** -(0.0001) FH ( 0.0018** + (0.0000) LOSS (0.0043** + (0.0002) Time ✓ Firm ✓ Observations 33,457 Number of groups 489 R-squared 0.11
Standard errors in parentheses ** p<0.01 & * p<0.05
The total model is significant with a F-score of 818, which means that the model is significantly influenced by one or more variables. This is true because all variables are significant with a p-value below 0.01. As described earlier, the R2 of this model is only 0.11,
this means that only 11% of the variance in accuracy is explained by the variables used in this regression model. However, the R2 of this research is about the same as comparable
research: 0.12 in Hope (2003) and 0.12 as well in Dhaliwal et al. (2012). In addition to that, the reason that the explained variance in ACC is only explained for this low percentage is that there is a variety of information sources which are being used by analysts to produce forecasts. It’s not possible to take all factors into account when making an archival research because some variables are not quantifiable and for a lot of variables there is no data available or measurable. Note that year and industry dummies are only included in the regression result as checkmark, because these are panel variables, although industry has been omitted.
This research hypothesizes that higher CSR performance leads to better forecast accuracy by analysts. The results show that there is a significant but small negative effect of CSP on ACC, meaning when CSP increases the accuracy amount will be lower. A lower accuracy amount results in a smaller error margin and more accurate predictions from financial analysts. This means that the hypothesis is not rejected and it’s assumed that assumed that there is an effect of CSR performance on financial analysts.
SIZE is another variable which is in line with theory, because when SIZE increases ACC decreases as well. This negative relationship implies that accuracy becomes better when companies are larger, which does correspond to the literature as described in 4.4 control variables.
AF has a negative relationship as well, implying that when more analysts are
following the company, the accuracy becomes better and error margin declines. This is also in line with theory.
FH and LOSS both have a positive relationship with ACC, which means that both relationships of the variables with ACC are also in accordance with the theory. A longer Forecast Horizon (FH), results in a higher ACC. This means that a longer forecast horizon leads to a reduced forecast accuracy. Besides that, when companies are making losses, accuracy goes up as well, implying that earnings are harder to predict during loss years.
5.4 Robustness tests
As mentioned in section 3.1 CSP and forecast accuracy, loss-making firms are harder to make estimates for (Ciccone, 2005). Therefore analysts are providing better information value in forecasting loss firms compared to forecasting profit firms. But as the information value from forecasts during loss years goes up, it’s likely that the error margin goes up as well because it’s harder to predict during a loss year. This is supported by Hope (2003), the research argues that losses make earnings more volatile in common law countries (the U.S. is a common law country). When losses occur, it’s also likely to affect CSR investments and CSP scores. This indicates that there is less predictability of CSP on ACC during loss
observations. To tests this, profit-making observations are discarded from the sample. Therefore 32,338 observations are dropped and only 1,119 loss-making observations are kept in the sample. Duru & Reeb (2002) already found that loss years are insignificantly different from zero. Meaning that there is an effect of a loss year on forecast accuracy. In the following table only loss observations are taken into regression:
Table 10: Regression with loss obs. only Variables (Constant) 0.0143* (0.0160) CSP -0.0042 (0.0026) SIZE -0.0013 (0.0017) AF -0.0019* (0.0009) FH ( 0.0043** (0.0002) Time ✓ Firm ✓ Observations 1,119 Number of groups 47 R-squared 0.255
Standard errors in parentheses ** p<0.01 & * p<0.05
In this sample, the effect of keeping only loss observations means that the model stays significant with an F-score of 91. The lower score compared to the original model is explained by having fewer observations. Besides that, variable CSP becomes insignificant with a p-value of 0.109. This means that the effect of CSP on accuracy isn’t significant
anymore with loss making observations only. Therefore, CSP isn’t an indicator of ACC in loss making years. Both the constant and variable SIZE also become insignificant at p-values of respectively 0.370 and 0.433. The R2 goes up from 0.11 in the main sample to 0.26 in this
sample. This means that the variables AF and FH are explaining the variance of accuracy more accurate compared to the main sample.
Reversely, it’s tested with profit-making years only, which means that all loss observations are deleted from the original sample. We already know that loss reduces the predictability of CSR performance on accuracy, therefore it’s expected to have a somewhat more significant score for CSP in this regression.
Table 11: Regression with profit obs. only
Standard errors in parentheses ** p<0.01 & * p<0.05
This model hasn’t changed much compared to the original regression, because the R2
is a little lower (9.9%) compared to the original model (11.0%). Besides that, all variables of the model stay significant at a p-value below 0.01 and have the same positive or negative coefficient as in the original regression. Although the F-score slightly increased from 817 to 874 while there are fewer observations, the model hasn’t changed much overall. This increase in F-score is probably a result from the fact that accuracy is worse for loss-making observations and relatively better for profit-making observations.
Variables (Constant) 0.0190** (0.0013) CSP - .-0.0013** (0.0003) SIZE -0.0019** (((((((0.0001)) AF -0.0011** (0.0001) FH ( 0.0017** (0.0000) Time ✓ Firm ✓ Observations 32,338 Number of groups 486 R-squared 0.099
6. Discussion and conclusion
In the following sections the discussion (§6.1), limitations (§6.2), opportunities for future research (§6.3) and the conclusion (§6.4) will be treated.
6.1 Discussion
The goal of this research was to investigate whether Corporate Social Responsibility performance does influence the forecasts of financial analysts. By using a panel model with corrections for time and firm effects the results are robust and provide insight to the hypothesized relationship.
The hypothesized relationships from CSR performance to analyst forecasts are both indirect. Firstly, firms with higher CSP should disclose their CSR program more clearly, because more CSR disclosure leads to a better forecast accuracy by financial analysts. And Secondly, CSR performance leads to more profitability nowadays and profit-making firms are more accurately predictable compared to loss-making firms. Based on this theory there has been hypothesized that higher CSR performance leads to more accurate analyst forecasts.
From the tests, a significant association has been found between CSP and forecast accuracy. A negative effect has been found between CSP and the error margin in forecasts, that was also what has been hypothesized. Meaning that the higher the CSP score is, the better the accuracy of analysts become, or the lower the error margin of their forecasts. After testing these results for loss-making observations only, CSP wasn’t significant anymore. With profit-making observations only, the results were comparable. This means that during loss CSP isn’t an accurate predictor for accuracy and during profit it is.
The fact that CSP has a significant impact on forecast accuracy is probably relatively limited, because all other variables have a higher significance score. Despite CSP only having a limited effect, it’s still useful for practitioners to understand there is an effect. The reason is that analysts use many different information sources to make their forecasts. Although it’s not probable that analysts use CSP ratings, CSR reports do increase forecast accuracy
according to Dhaliwal et al. (2012). Therefore analysts might be using CSP ratings indirectly to figure out how reliable the firms’ CSR report is, this could explain the impact of CSP on forecast accuracy.
6.2 Limitations
This research has only taken companies from the US into consideration. By limiting to one country, country-level effects are not measured. To test in one country and not test multiple countries was done deliberately to simplify the research because it was not
practical to implement more control variables. This research has produced a comparable R2
score. Although the R2 score is in line with comparable research, it’s possible that
country-specific variables would result in a different outcome. Therefore the results may not be possible to reproduce when doing this study in other countries, because of the different country characteristics.
There is only an indirect theoretical relationship between corporate social
responsibility performance and analyst forecast accuracy, because of limited literature on this subject. This is because most research involving CSR or analysts are covering a different scope. The results of this research indicate that CSP only affects analyst accuracy for a small part, therefore researchers may came to the conclusion that the relative impact of CSP on accuracy isn’t as important as many other variables.
6.3 Future research
The theory on which this research has been based is indirectly related to each other. Therefore more research is needed to confirm that CSP does have a significant effect on analyst accuracy. This is doable by using data from different countries and correcting for country effects. A good starting point will be that there are enforcement differences
between US GAAP and IFRS countries because it’s expected to have a higher accuracy in the US compared to IFRS countries. Which means that this could lead to differences between countries and getting other results because of these differences.
Another possibility for future research is to elaborate on which information financial analysts use in their reports to make forecasts. This information can be used to incorporate in research about forecast accuracy, like this study. With more reliable variables in a
regression, more uncertainty about results could be taken away and therefore more variance can be explained and results will be more robust.
6.4 Conclusion
This research explored the effect of CSP on the forecasting ability of financial analysts. By taking data from IBES, Compustat and Thomson Reuters the full dataset could have been created. The focus of this research was answering the research question, formulated in section 1. introduction: “Does CSR performance affect analyst forecast accuracy?” To answer this question: A significant association has been found between CSP and forecast accuracy, the effect found was negative and this does correspond to the expectations formulated in the hypothesis. The results implies when firms are having higher CSP scores, the accuracy becomes better for financial analysts.
Other research made insightful that CSR information leads to more awareness by financial analysts (Ioannou & Serfeim, 2010; Dhaliwal et al., 2012 & Luo et al., 2015). By taking that theory a step further, this study focused on CSR performance specifically. With the results of this research, the impact of CSP on accuracy is relatively small compared to forecast horizon. Although in conclusion, Corporate Social Performance does have an effect on forecast accuracy.
Reference list
Alford, A.W. & Berger, P.G. (1999). A Simultaneous Equations Analysis of Forecast Accuracy, Analyst Following, and Trading Volume, Journal of Accounting, Auditing & Finance, 14 (3), 219-240.
Barnett, M.L. & Salomon, R.M. (2012). Does it pay to be really good? addressing the shape of the relationship between social and financial performance, Strategic Management Journal, 33 (11), 1304-1320.
Barth M.E., Landsman, W.R., Lang, M. & Williams, C. (2012). Are IFRS-based and US GAAP-based accounting amounts comparable? Journal of Accounting and Economics, 54 (1), 68-93. Behn, B.K., Choi, J.H. & Kang, T. (2008). Audit Quality and Properties of Analyst Earnings Forecasts,
The Accounting Review, 83 (2), 327-349.
Bell, A. & Jones, K. (2015). Explaining fixed effects: Random effects modeling of time-series cross-sectional and panel data*, Political Science Research and Methods, 3 (1), 133-153. Block, S.B. (1999). A Study of Financial Analysts: Practice and Theory, Financial Analysts Journal,
55 (4), 86-95.
Brammer, S., Brooks, C. & Pavelin, S. (2006). Corporate Social Performance and Stock Returns: UK Evidence from Disaggregate Measures, Financial Management, 35 (3), 97-116. Brown, L.D. (1998). Managerial Behavior and the Bias in Analyst’ Earnings Forecasts, Georgia
State University, available at: https://ssrn.com/abstract=113508 or http://dx.doi.org/10.2139/ssrn.113508
Brown, L.D., Call, A.C., Clement, M.B. & Sharp, N.Y. (2015). Inside the “Black Box” of Sell-Side Financial Analysts, Journal of Accounting Research, 53 (1), 1-47.
Carroll, A.B. (1979). A Three-Dimensional Conceptual Model of Corporate Performance,
Academy of Management Review, 4(4), 497-505.
Carroll, A.B. (1991). Corporate The pyramid of corporate social responsibility: Toward the moral management of organizational stakeholders, Business Horizons, 34 (4), 39-48.
Ciccone, J. (2005). Trends in analyst earnings forecast properties, International Review of
Financial Analysis, 14 (1), 1-22.
Dhaliwal, D.S., Radhakrishnan, S., Tsang, A. & Yang, Y.G. (2012). Nonfinancial Disclosure and Analyst Forecast Accuracy: International Evidence on Corporate Social Responsibility Disclosure, The Accounting Review, 87 (3), 723-759.
Duru, A. & Reeb, D.M. (2002). International Diversification and Analysts’ Forecast Accuracy and Bias, The Accounting Review, 77 (2), 415-433.
Dyllick, T. & Hockerts, K. (2002). Beyond the business case for corporate sustainability, Business
