Abstract
This paper uses a worldwide sample to investigate the
‘uniqueness paradox’, which affects managers who pursue long term value through executing a unique strategy. This results in lower coverage by stock analysts, which reduces the market valuation of the firm. Additional attention has been given to digital platform firms in this study, as the winner-take-all structure of the market is expected to increase the value of unique strategies, but requires investors and analysts to spend
more effort to evaluate the firm. This study finds limited evidence for the expected negative relationship between unique
strategies and analyst coverage, and does not find the positive relationship between analyst coverage and the firm’s valuation.
However, we did find evidence that the uniqueness paradox is stronger for digital platform firms, and we observe a direct negative relationship between strategy uniqueness and firm market value. These findings show that the information cost of
evaluating strategy is a significant driver of the valuation of a firm.
Keywords: Uniqueness paradox, strategy uniqueness, digital platform firms, analyst coverage, market valuation.
The uniqueness paradox for digital platform firms
Do investors and analysts appreciate unique digital platform strategies?
Master Thesis Tim Hendriks, BSc.
S2390612 +31 6 83 20 74 52 T.Hendriks.3@student.rug.nl MSc BA, Change Management Faculty of Economics and Business
University of Groningen
Thesis supervisor: Prof. dr. J. Oehmichen Word count: 10373
1
1. Introduction
The widespread adoption of the internet and increased connectivity of consumers over the past decades have paved the way for the rise of digital platform firms in today’s economy. Examples of this are Uber, Airbnb, eBay and Facebook, which have created a different business model from that of traditional firms. Digital platform firms are “businesses that create significant value through the acquisition, matching and connection of two or more customer groups to enable them to transact” (Reillier & Reillier, 2017). This can be contrasted with traditional firms, who create value by providing their clients with the goods and services themselves. Digital platform firms provide a platform or market place, where both supply and demand of goods or services comes from the users of that platform (Cennamo & Santalo, 2013; Evans, 2003; Gawer, 2010; Rochet &
Tirole, 2006; Zhao, von Delft, Morgan-Thomas & Buck, 2019). The online nature of platforms allows rapid expansion and a very low marginal cost per user (Zhao et al., 2019). The platform market has grown rapidly in the past decade (Zhu & Liu, 2018). In 2018, the global platform economy was estimated to be worth over 7 billion dollars, up from 4.3 billion in 2016 (KPMG, 2018).
One of the main requirements for the success of a platform firm is its user base. An online platform requires large amounts of users to supply and demand goods and services. The more users can be attracted, the more value can be created. These network effects are particularly strong in platform firms, as additional users provide a greater diversity and volume of supply and demand which the platform cannot generate in any other way (Boudreau & Jeppesen, 2015; Cusumano, Yoffie &
Gawer, 2019). In contrast to traditional firms therefore, a platform’s user base is one of its key resources in generating a durable competitive advantage. This leads to fierce competition for the user base between rival platforms, in which the platform that manages to attract most users often comes out on top: the winner takes all (Boudreau & Jeppesen, 2015; Cusumano et al., 2019; Zhao et al., 2019).
These market conditions lead to a situation in which aggressively growing and acquiring a large amount of users is the key to success (Boudreau & Jeppesen, 2015; Gawer & Consumano, 2002;
Schilling, 2002). If a young platform firm acquires a small advantage that boosts user adoption early in its lifecycle, learning curve effects together with the development of complementary technologies can give the platform an ever-increasing advantage as the userbase increases (Schilling, 2002). Given this situation, it can be expected that platform firms would strongly benefit from the development of a strategy that is unique from their competitors, in order to find an advantage that gives the platform an edge over its rivals.
2 However, research by Litov, Morton & Zenger (2012) found that firms face a ‘uniqueness paradox’
in selecting a strategy. Because unique strategies require analysts and investors to devote more resources to understanding and valuing the firm than existing strategies, analysts are less likely to cover the firm, which led to a lower valuation. The management of a firm therefore faces a choice between utilizing a common strategy which results in a higher firm value in the short run, or pursue a unique strategy that has more potential of paying off in the future: after all, a unique strategy is also necessary for generating durable economic rents (Litov et al., 2012). Previous studies, however, provided an insight in firms in general, and did not make specific predictions for digital platform firms.
This provided the avenue into the main research question of this paper. Firstly, we attempt to reproduce the general findings of Litov et al. (2012). Secondly, due to the winner-take-all structure of the platform market, a unique strategy is likely to be a more crucial factor for success for digital platforms than for a traditional firms.
Does the platform business model of a firm influence the relationship between strategy uniqueness and its market valuation?
This paper aims to answer this question using a sample of 3.348 firms drawn from the MSCI World Index between the years 2008 and 2017. This dataset allows us to validate the results of Litov et al. (2012), taking firms from outside the U.S. into account as well. Answering this question helps to further understand the dynamic between the valuation of long term earnings potential and the information cost of choosing a unique strategy, particularly in the context of platform firms, which have quickly become large and valuable players in today’s economy.
After testing our hypotheses, our findings differ from Litov et al. (2012), as we did not find evidence for the influence of strategy uniqueness on analyst coverage, and we could also not conclude that analyst coverage drove the market valuation of the firm. However, we did find a direct negative relationship between strategy uniqueness and market value, as well as grounds for concluding that unique strategies lead to a stronger market value discount and larger analyst coverage reductions for platform firms.
Firstly, this paper will provide a short overview of the literature on network effects, specifically in the context of platform firms, as well as research on the uniqueness paradox. Secondly, the hypotheses and methodology are presented. In the next section, the results of this research are shown. After that the implications of the results and the limitations of this research are presented, followed by the conclusion.
3
2. Literature review
2.1 The value of users
Following the Resource Based View of the firm, a firm can create value by creating a unique combination of resources and capabilities which are rare, valuable and difficult to imitate or substitute (Barney, 1991; Wernerfelt, 1984). An important difference between traditional firms and digital platform firms is the presence of large network effects in platform firms, as the value of the platform increases with the amount of users (Katz & Shapiro, 1985). Network effects are not just limited to platform firms, as traditional firms might experience network effects through additional compatibility and the development of complementary products by other firms. The difference for platform firms comes from the fact that users themselves actively create a substantial amount of the value for other users on the platform.
On one hand, an increase in platform users who create the supply of products and services, ‘the complementors’, adds value. A larger group of complementors can supply a more diverse range of goods and/or services on the platform (Cennamo & Santalo, 2013). On the other hand, an increase in buying users increases, the demand for the products of complementors. Besides this, many platforms also accommodate ‘unpaid crowd complementors’ (Boudreau & Jeppesen, 2015). These users provide additional functionality or services to the platform, such as add-ons, apps or modifications, or additional information such as product reviews or (news) reports. The crowd complementors are often motivated intrinsically and heterogeneously, driven for example by their own learning or use-cases, or esteem and capability-signaling motivations towards other users, instead of by financial gain (Boudreau & Jeppesen, 2015). Crowd complementors provide another avenue through which a platform’s userbase provides value to the platform,
For platform firms, this means that their userbase is an important resource for the creation of value and their eventual success, to the extent that platforms that manage to attract a critical mass of users first, often emerge as the dominant platform due to the positive feedback loop created by the additional users (Cennamo & Santalo, 2013; Schilling, 2000; Zhu & Liu, 2018). Competition between platform firms will therefore be strongly focused on attracting as many users as possible in order to become the dominant platform (Cusumano et al., 2019). Platforms attempt to differentiate themselves by providing additional features, reduce the friction of operating on the platform, and match users more effectively (Zhao et al., 2019). The need for differentiation encourages platforms to follow a different strategy to those of their competitors.
2.2 The uniqueness paradox
Following a unique strategy from rivals is not just important for platforms. All firms create value by creating unique combinations of resources and activities (Barney, 1986; Litov et al., 2012).
4 Foresight by managers regarding these unique combinations allows them to buy the resources on the market for below their worth in the managers’ strategy and use these resources in activities in order to generate economic rents (Barney, 1986; Litov et al., 2012). Choosing a unique strategy that is resistant to replication is necessary to prevent imitation by other firms, which would eliminate the rents through competition (Litov et al., 2012).
Following this reasoning, it could be expected that the stock market will value firms capable of generating economic rents through unique strategies higher than a comparable firm which does not pursue a unique strategy. However, investors require information and understanding of the firms that they can invest in, which is costly (Litov et al., 2012). In reality, the monitoring of firms is often delegated to stock market analysts by investors.
Stock market analysts are important market intermediaries. Their recommendations have been found to influence stock prices (Barber, Lehavy, McNichols & Trueman, 2006), and play a role in the dismissal of CEO’s (Wiersema & Zhang, 2011). Furthermore, analysts provide monitoring and information to investors. Investors, especially small shareholders, often face a collective action problem when monitoring the firms they have invested in (Chung & Jo, 1996; Macey, 1997).
Monitoring a firm is important for shareholders to prevent agency problems and mismanagement, but comes at the cost of time and resources. While it is beneficial for all shareholders if monitoring takes place, the cost of monitoring is not worth the benefits to an individual investor, which discourages shareholders from monitoring (Chung & Jo, 1996). Analysts can play an important role in solving this problem, as they provide an independent, external analysis of the firm, decreasing the need for shareholder monitoring and the agency costs associated with the collective action problem (Wiersema & Zhang, 2011). Analysts are often employed by brokerage firms, and are compensated with transaction fees from the brokerage (Litov et al. 2012).
This leads stock market analysts to also face a trade-off when deciding which firms to monitor.
Uncovering and understanding unique strategy comes at an information cost to analysts, as unique strategies require more time and expertise in order to understand correctly, compared to strategies that are well-known to them (Litov et al., 2012). The managers of a firm will have a better insight in the value of their resource-activity combination than other market participants;
after all, if this value was immediately apparent to the rest of the market, it would be trivial for others to copy the strategy, diminishing its value (Barney, 1986; Litov et al., 2012).
Analysts are generally incentivized to cover as much firms, as accurately as possible (Litov et al.
2012). This also explains the tendency of stock market analysts to specialize in specific industries (Zuckerman, 2000). Their incentives would lead analysts to focus on firms which are less costly to follow, as it allows them to cover more firms without compromising accuracy (Litov et al. 2012).
5 This leads to lower coverage of ‘difficult’ firms by analysts as a whole. In turn, this decreases the amount of information about firms with unique strategies available in the market to (potential) investors, making it more difficult to recognize and value investment opportunities in a firm and increasing monitoring costs. The result is decreased demand for the firm’s stock and a lower stock price (Litov et al., 2012). Litov et al. (2012) and Chaffkin (1999) also propose more anecdotal evidence that analysts that do perform a costly analysis on difficult to value firms would analyze more extensively and also produce higher ratings if the firm were to adopt an easier to analyze strategy.
This leads to the uniqueness paradox as described by Litov et al. (2012). Due to the behavior of analysts, managers face a trade-off. They can choose a common strategy, which will increase available stock market information for its firm and have a positive effect on their valuation, but is less likely to generate value over the long term. On the other hand, they can choose a unique strategy with benefits regarding long-term value generation, which will receive less coverage by analysts. Litov et al. (2012) found that in their sample of 7,630 U.S. firms, choosing a unique strategy led to higher valuations on its own, but also lead to lower analyst coverage, which decreased the valuation.
3. Hypothesis development
The first aim of this paper is to validate the results in the study of Litov et al. (2012) by reproducing the main relationship it has found between strategy uniqueness and the amount of coverage received by analysts, They also showed a relationship between strategy uniqueness and the market-to-book value of a firm. The contribution from this paper in this regard stems from the inclusion of firms based around the world in the sample. As Litov et al. (2012) only considered U.S. firms in their sample, this study would allow for further generalization of their findings.
Furthermore, because this study differs in some respects to Litov et al. (2012) in terms of model specification and control variables used, it becomes possible to see if conclusions made in regards to the uniqueness paradox are robust to changes in the model specification. We will therefore test the following hypotheses, based on the conclusions of Litov et al (2012):
H1: There is a negative linear relationship between the uniqueness of a firm’s strategy and the amount of coverage of the firm by stock market analysts.
H2: There is a positive linear relationship between the amount of analyst coverage of a firm and the market-to-book ratio of this firm
H3: There is a positive linear relationship between the uniqueness of a firm’s strategy and the market-to-book ratio of this firm.
6 Furthermore, this study aims to contribute to the understanding of the uniqueness paradox by making a distinction between digital platform firms and other firms. Platform firms can be expected to be more sensitive to the positive effect of following unique strategies on long term value generation due to their reliance on differentiation to attract new users quickly. This would also make the (costly to analyze) unique strategy a more important subject in the analyst’s valuation than it would be for a traditional firm, decreasing the analyst’s incentives to cover the platform firm. This drives the hypothesis that the relationship between strategy uniqueness and stock analyst coverage is moderated by a platform business model. Investigating these relationships contributes to the understanding of the importance of the uniqueness paradox for platform firms in shedding light on the information costs of analysts in their coverage of platform firms, and the valuation the market places on the unique strategy.
H4: The relationship between the uniqueness of a firm’s strategy and the amount of stock analyst coverage it receives is moderated by whether or not the firm is a platform firm, such that the relationship is stronger for platform firms
H5: The relationship between the uniqueness of a firm’s strategy and the market-to-book ratio of that firm is moderated by whether or not the firm is a platform firm, such that the relationship will be stronger if the firm is a platform firm.
Lastly, this study investigates the relationships between strategy uniqueness and the average analyst recommendation the firm receives. While there is anecdotal evidence that difficult to value firms - such as firms with unique strategies - receive lower recommendations, this study aims to validate these claims with statistical evidence. It is also expected that higher recommendations lead to a higher market-to-book value, as has been established in other studies. Due to the hypothesized increased value of unique strategies for platform firms, we expect that analysts also recognize this added value, which would moderate the relationship between strategy uniqueness and the recommendation of its stock.
H6: There is a negative linear relationship between the uniqueness of a firm’s strategy and its average recommendation by analysts.
H7: The relationship between the uniqueness of a firm’s strategy and the recommendation of its stock by stock market analysts is moderated by whether the firm functions as a digital platform. This relationship is weaker when the firm functions as a digital platform firm.
H8: There is a positive linear relationship between the average analyst recommendation of a firm and its market-to-book value.
7
4. Methodology
4.1 Sample
In order to test the hypotheses, firm data from Compustat was used to create an unbalanced sample of 3.348 firms covered in the MSCI World Index between 2008 and 2017, for a total of 21.046 firm years. On average, 6.3 yearly observations per firm are recorded in the dataset. 65 firms in this sample were identified as platform firms, with 386 observations in total. This dataset provided accounting data, as well as details regarding the sectors the firm has been active in, and how their revenues have been divided between sectors. This sample was matched with analyst data from I/B/E/S/, which contained information about analyst coverage and recommendations of the firms in the sample. The sample contains firms from around the world. About 36% of firms are based in Asia, 30% in North America, 20% in Europe, 7% from Middle or South America.
Australian, Middle Eastern and African firms make up the rest of the sample. An issue with the dataset came from missing accounting data used as control variables in the model. This resulted in roughly 60% of the firms being dropped from models using the control variables, which made it more difficult to draw conclusions.
4.2 Method
Due to the panel structure of the dataset, we require a statistic method that allows drawing a conclusion about relationships between variables across all firms, while taking into account the unbalanced time-series structure and the fact that the error terms within observations of the same firm are likely to be correlated. In many cases, a fixed effects GLS regression model fits these requirements well (Brooks, 2014). Fixed effects regressions are also able to estimate the explanatory value of the model, and its coefficients are generalizable to individual firms in the sample. In the case of this study, we use fixed-effects regressions in order to estimate all coefficients in models without influence of platform firms. That is because the dummy variable used to mark platform firms does not vary over time, because a firm is either marked as a platform firm or not throughout the dataset. This means that it is not possible to use a fixed effects model to estimate the expected moderating effect of platform firms.
For this reason, a Generalized Estimation Equations (GEE) method was used in order to make inferences about relationships involving platform firms. The GEE method is well suited for drawing inferences about a population from a sample with panel data with multiple observations per firm (Liang & Zeger, 1986), but cannot predict effects on individual firms.
GEE allows for treating the correlation within observations of one firm as noise (Liang & Zeger, 1986), which lets GEE make inferences about the population without influence of correlated observations within firms. It does this by first calculating a standard linear regression, which
8 comes with the assumption that all errors are independent. Then, a correlation matrix is generated based on the results of the first regression, which is used to adjust the regression model in the second iteration. This continues until the tolerance of the model is minimized. GEE uses heteroskedasticity robust standard errors by default.
An issue with this method is the fact that GEE is not completely robust to data that is missing not at random. Because the entry and exit of firms from the MSCI World Index can be based on data reported in the dataset itself (such as firm size), this can create a bias in the GEE estimations.
Furthermore, it is not possible to assess the explanatory value or combined significance of all variables in the model. In order to combat these weaknesses, we have specified models with and without interaction effects, and inspected the results between fixed effect regression outcomes and GEE outcomes. We did not find evidence that the GEE specification introduced a noticeable bias.
4.3 Independent variables
4.3.1 Strategy uniqueness. This study follows the methodology established by Litov et al.
(2012) for the calculation of the measure of strategy uniqueness. They make the assumption that, because analysts are often specialized in a certain industry, their familiarity with the strategy of a firm depends on the degree to which the strategy of the firm varies from its industry peers.
For every firm i, a vector is created for its revenues generated in year t across all market segments N, as follows:
𝑠𝑖,𝑡 = [𝑠𝑎𝑙𝑒𝑠1,𝑖,𝑡 . . 𝑠𝑎𝑙𝑒𝑠𝑁,𝑖,𝑡]′
Like Litov et al. (2012), the segments are defined as all 4 digit SIC codes in which any revenue was generated by any firm during the duration of the sample. In our case, firms across the sample were active in 778 of such segments, therefore N = 778. This vector is then normalized to unit length by dividing it by
∑ 𝑠𝑎𝑙𝑒𝑠𝑗,𝑖,𝑡 𝑗
where j indicates the set of segments N that were present in year t. This vector now contains the distribution of revenues over all different segments that firm i is active in.
Then, the primary industry of all firms is determined, by finding the segment in which a firm generated the largest part of its revenue. The primary industry of a firm is defined as 𝑗∗. This primary industry is used in the vector
𝑠𝑗∗,𝑡 = [∑ 𝑠𝑎𝑙𝑒𝑠1,𝑖,𝑡
𝑖 . . ∑ 𝑠𝑎𝑙𝑒𝑠𝑁,𝑖,𝑡
𝑖 ]
9 which includes for every segment j, all firms that have 𝑗∗ as their primary industry. This vector is also normalized to unit length by dividing it by
∑ ∑ 𝑠𝑎𝑙𝑒𝑠𝑗,𝑖,𝑡
𝑖 778
𝑗=1
This vector now contains the average distribution of revenues over all subsegments that all firms with primary industry 𝑗∗ are active in. The two unit length vectors are then used in order to create a uniqueness variable, which measures the deviation of the distribution of its revenues compared to other firms in the same primary industry as follows:
𝑈𝑛𝑖𝑞𝑢𝑒𝑛𝑒𝑠𝑠𝑖,𝑡 = (𝑠𝑖,𝑡− 𝑠𝑗∗,𝑡)′∗ (𝑠𝑖,𝑡− 𝑠𝑗∗,𝑡)
The difference between the revenue share of the firm and its peers per segment is squared, and the resulting outcome is summed across all industries that firms with the same primary industry are active in, in order to arrive at a Uniqueness score. This measures how a firm generates revenue in different markets compared to the average of its closest peers (Litov et al., 2012). Note that this does not mean that a more diversified firm
necessarily receives a higher uniqueness score: if most firms in a primary industry are diversified over the same segments with roughly equal shares, they receive a low uniqueness score, while a firm in the same industry that collects all of its revenue in its primary industry will receive a high uniqueness score. In order to make sure that firms that are alone in their SIC4 segment do not influence the result (these firms would receive a strategy uniqueness score of zero,
regardless of their strategy), these firms are deleted from the sample. This reduced the sample size by 1,103.
In our sample, the majority of firms receive a low strategy uniqueness score, with more unique strategies tapering off in occurrence (see figure 1). This shows that the measure is behaving as intended in rating the uniqueness of strategies, as by definition, more unique strategies should be less likely to be employed by firms.
Measuring strategy uniqueness in this way limits this study to investigating corporate strategy, as product or business strategy would not be captured by revenue in other SIC segments. Secondly, Figure 1: This histogram shows the distribution of the strategy uniqueness score over the sample.
10 unique corporate strategy that does not influence the distribution of sales over SIC segments can also not be captured. Thirdly, the primary industry in which a firm falls can influence its strategy uniqueness score, which is relevant if a firm’s revenues are almost equally divided over two segments, in which case a small difference in the revenue distribution would mean that the firm would be compared to another primary segment.
4.3.2 Platform firm indicator. As defined in the introduction, platform firms are
“businesses that create significant value through the acquisition, matching and connection of two or more customer groups to enable them to transact” (Reillier & Reillier, 2017). This definition from the literature is helpful in understanding platform firms, but provides little guidance in selecting platform firms in the sample. Therefore, this definition is further operationalized with the help of the WisdomTree Modern Tech Platforms Index Fund. WisdomTree is an exchange traded fund sponsor and asset manager. WisdomTree has published their criteria for inclusion in the ETF, and a list of the companies that make up the ETF, which made it a reliable source to start in labelling platform firms in the dataset. The aforementioned fund is an ETF consisting of 69 mid- and large cap firms that WisdomTree has included in their platform ETF. The following criteria have been formulated by WisdomTree in determining whether or not a firm should be regarded as a platform firm, and have also been used as criteria in further manual selection of platform firms in this study (WisdomTree, 2019):
- business has direct relationship with a user group who consumes value (e.g., product, service, content, etc.).
- business has direct relationship with an external user group who supplies the value to be consumed by another user group. In some cases, this producer user can be the same person as the consumer group, but they engage in a separate set of activities related to creating value when acting as a producer.
- the value being consumed by the consumer user group is being supplied by a third-party and is not directly controlled by the company, and that supply does not sit on company’s balance sheet.
- there is a positive network externality between the consumer and producer user groups, meaning that the demand for the platform from one user group is dependent upon the number of users on the other side of the platform. The company will either explicitly mention it is subject to network effects or include a statement indicating that the more consumers that use the platform, the more value each producer will get from the platform, and vice versa.
- The company owns the network by which the consumers and producers directly connect. If the company services multiple customer groups but does not own the underlying network
11 that connects them, it may not qualify (i.e., it is a service provider to that network, and not a platform business itself).
- The company derives revenue from its platform business unit(s). There is direct platform revenue, which includes money captured as revenue by the platform as a percentage of the monetary value of each transaction it facilitates (e.g., a take rate). There is also indirect platform revenue, which includes revenue generated by providing products or services related to the facilitation of value exchanges on the platform (e.g., advertising, fulfillment services, additional software features that enable the user to transact on the platform).
34 of the firms that are part of the platform firm ETF were present in the sample. All these firms have been marked as platform firms in the dataset. Besides that, in all 2-digit SIC segments in which WisdomTree had identified platform firms, a manual search through the dataset has been conducted in order to find additional platforms that had not made it into the ETF. This approach for manually labelling platform firms was taken for two reasons: Firstly, it was expected that platform firms would cluster around particular SIC segments, because many SIC segments would not be able to contain platform firms due to the segment definition. Secondly, it was not feasible to manually evaluate every firm in the dataset for matching the platform firm criteria. Still, this approach leaves room for platform firms that have not been labelled as such by the ETF or the manual search, which is a limitation of this study.
4.4 Dependent variables 4.4.1 Analyst coverage
Analyst coverage is measured by summing all analyst recommendations from I/B/E/S/ for one firm over one year. If an analyst produced more than one report for the same firm in one year, all reports would be counted. In contrast to Litov et al. (2012), we do not construct multiple analyst measures in order to control for analyst specialization, analyst effort, specific coverage or adjusted coverage. For this study, the available data was not exhaustive enough in order to make valid conclusions regarding, for example, the total number of firms covered by one analyst, or the segments in which analysts were specialized. Our sample only consisted of firms in the MSCI World Index, and stock market analysts that rated firms included in this study cannot be assumed to only, or mostly, rate firms in our sample. We therefore felt that it was not appropriate to construct measures from this likely limited data. This can be a limitation of this study, as it is possible that these more detailed measures are required in order to be able to draw the same conclusions as Litov et al. (2012)
4.4.2 Market value of the firm
12 Tobin’s Q is used to measure the relative market value of a company in terms of market-to-book values of firms (Chung & Jo, 1996; Litov et al., 2012; Morck, Schleifer & Vishny, 1988). Tobin’s Q indicates the level to which a firm’s market value is above the replacement value of the asset.
When viewed over the entire market, the expected mean Tobin’s Q would be 1, else the entire market would be over- or undervalued. In our sample, the mean Tobin’s Q is 1.71, but our sample reflects an index of successful companies from over the world, not a cross section of the world market.
4.4.3 Average analyst recommendation. Analyst recommendations in I/B/E/S/ are reported on a scale of 1 to 5, which correspond to a “strong buy” and a “strong sell” rating, respectively. In order to make the results more intuitive, the average recommendation is recoded in order for better recommendations to receive a higher score. The recommendations of all analysts are averaged per firm for each year that they appear in the dataset. Average recommendations are approximately normally distributed, as can be seen in figure 2, but the distribution shows more observations around full numbers. This is caused by the fact that the recommendations are recorded as integers, and firms with fewer analysts covering them are more likely to end up with a round number as their average score.
By using the analyst recommendations as a dependent variable, this study expands the study of Litov et al. (2012), This study chooses to evaluate the average recommendation, because it gives an insight in the beliefs of analysts on whether the firm is an attractive investment.
In this study, the analyst recommendation is treated as interval data. This could pose a risk to the validity of this study, as analyst recommendations can be interpreted as more ordinal than interval, as a “strong buy”
recommendation (5) does not necessarily have to be twice as far removed from a “hold”
rating (3) as the “buy” recommendation (4).
However, in other literature concerning analyst recommendations, their ratings are routinely treated as interval variables for the purposes of calculating average
recommendations (e.g. Jegadeesh & Kim, 2006; Lin & Tai, 2013; Wiersema & Zhang, 2011), which suggests that treating recommendations as interval data does not introduce bias.
Figure 2: This histogram shows the distribution of the average analyst recommendation in the sample.
13 Figure 2 shows that analysts are biased towards rating stocks favorably, a phenomenon that is well-known and studied in the literature (Barber et al., 2006; Jegadeesh & Kim, 2006), and is explained by findings that analysts are subject to incentives to be optimistic (Barber et al., 2006;
Hong & Kubik, 2003).
4.5 Control variables
In the selection of control variables, this study followed the approach of Litov et al. (2012) and Barth, Kasznick and McNichols (2001) where the available data allowed for this. In order to avoid influence of outliers, all accounting data is winsorized at 2.5% at each tail. Furthermore, the natural logarithm of the accounting data is taken in order to normalize the variables. Table 1 shows the descriptive statistics of all control variables, as well as the dependent and independent variables
Firm size and growth. Wiersema & Zhang (2011) recommend controlling for firm size as larger firms are more closely examined by the market. Market capitalization is therefore added as a control variable. Barth et al. (2001) further advise controlling for total common equity, revenue, assets and sales growth over the last three years due to findings in previous research on variables that influence on analyst coverage.
Table 1: Descriptive statistics of model variables
Mean Median Min Max Std. Dev N
Strategy uniqueness score 0.075 0.038 0 0.710 0.095 19,943
Platform firm indicator 0.019 0 0 1 0.138 19,943
Average analyst recommendation 3.439 3.462 1 5 0.441 19,943
Analyst coverage 17.189 15 1 106 11.178 19,943
Std. dev. of analyst recommendations 0.904 0.908 0 2.828 0.247 19,681
Tobin’s Q 1.707 1.302 0.823 5.541 1.043 19,840
Log of market capitalization 15.716 15.620 13.822 18.253 1.061 19,845
Log of common equity 15.007 14.934 12.708 17.643 1.138 19,654
Log of net sales 15.340 15.290 12.790 18.108 1.306 19,868
Sales growth over the last three years 0.870 0.211 -9.892 2923.671 25.092 19,229
Return on equity % 15.241 12.772 -22.090 72.766 16.635 19,748
Share of intangible assets 0.140 0.046 -0.007 1.893 0.190 19,655
Amortization of intangible assets 10.153 10.274 5.759 13.809 1.947 14,138
R&D expenses as percentage of total 0.052 0.020 0 0.301 0.076 9,058
Depreciation expenses as percentage of total 0.081 0.051 0.005 0.370 0.084 19,644
Diversification dummy – 1 segment 0.354 0 0 1 0.478 19,943
Diversification dummy – 2 segments 0.273 0 0 1 0.445 19,943
Diversification dummy – 3 segments 0.179 0 0 1 0.383 19,943
Diversification dummy – 4+ segments 0.151 0 0 1 0.358 19,943
This table shows the descriptive statistics of the dependent. Independent and control variables. Accounting variables have been winsorized at the 95%
14 Stock return. The return of a stock is perhaps the most interesting aspect of a stock for investors and will therefore influence the recommendation of analysts. This study therefore follows Litov et al (2012) in controlling for return on equity.
Intangible assets. The amount of intangible assets present in a firm has been found to increase the difficulty and uncertainty of valuing a firm by analysts (Barth et al., 2001). This study hypothesizes that the same is true for unique strategies. In order to avoid confounding these effects, the share of intangible assets in total assets is controlled for, as well as the total amortization of intangible assets. We further follow the recommendations of Barth et al. (2001) to control for R&D expenditure as a share of total expenditure due to its contribution to the creation of intangible assets, and depreciation expenditure, as that measures the tendency of a firm to use tangible rather than intangible assets.
Strategy complexity. Firms that are active in more segments are more difficult and costly to analyze than single segment firms (Litov et al., 2012). This means we add strategy complexity in order to distinguish its effect on recommendations from the influence of strategy uniqueness. We therefore add dummies to identify firms that are active in 1, 2, 3 or 4 or more segments
Analyst measures. The rating of analysts can be influenced by their peers. In other studies, correlation has been found between the amount of coverage a firm has received and the average recommendation it receives (Wiersema & Zhang, 2011). Similarly, the variance of analyst ratings can also influence the average recommendation of a firm (Wiersema & Zhang, 2011). Therefore, the variance of analyst recommendations and analyst coverage are controlled for in models testing for the effects involving average analyst recommendation and the market value of a firm.
Year, industry and country. Lastly, effects from the year, industry and country that a firm is active in are controlled for. Time should be accounted for due to the variations in stock market over time (Wiersema & Zhang, 2011). The country that a firm is active also influences its valuation by analysts (Luo & Zheng, 2018). Lastly, the performance of peers in the same industry have an impact on analyst recommendations (Wiersema & Zhang, 2011). Dummy variables have been added to the model for each year, each 2 digit SIC code and every country in the dataset.
A number of factors that could not be controlled for due to the lack of available data, are variables related to stock liquidity, stock issuances, and measures related to market share. Litov et al.
(2012) argue for their inclusion when investigating relationships influencing analyst coverage.
This means that some of the effects of these omitted variables can be present in the model through their effect on analyst coverage. However, it is possible that an omitted variable bias is introduced due to the exclusion of these control variables.
15
5. Results
The following section presents the results of this study. Firstly, the results regarding the hypotheses which followed the findings of Litov et al. (2012) are covered, as well as the influence of the moderating effect of a platform business model. Secondly, the relationships regarding the average analyst recommendation are presented. Lastly, the impact of uniqueness, platform firms, analyst coverage and average recommendation on the valuation of the firm is discussed.
5.1 Analyst coverage
In order to test our first two hypotheses, we have reproduced the study of Litov et al (2012) regarding the influence of strategy uniqueness on analyst coverage within the constraints of data availability. Four models have been used to measure these effects. The first three models used a fixed effects GLS regression, clustering at the company level. The last model used GEE in order to be able to include the interaction term between platform firms and strategy uniqueness. The results obtained using these four models can be found in table 2. Model one contains only the independent and dependent variable. This model has very little explanatory power, producing an R-squared of 0.00. Nonetheless, it does find a significant negative relationship between strategy uniqueness and analyst coverage. The second model contains just the control variables. The control variables are able to explain roughly 20% of the variance in analyst coverage. It is noticeable that the dataset contains missing data for the control variables, decreasing the observations in the dataset to 1.286 firms over 6.383 firm years. Also noticeable is the significant negative coefficient for the share of R&D expenses in this model. R&D expenses, like strategy uniqueness, are expected to increase the difficulty of producing a recommendation for analysts.
Model three contains the full model except for the influence of platform firms. This model does not manage to explain more variance than the control variables alone, and the significance of strategy uniqueness is also lost. The last model also models the influence of platform firms.
Noticeable are the highly significant positive coefficient for platform business models, and the significant negative interaction term for strategy uniqueness and platform firms. This indicates that platform firms with common strategies receive more coverage by analysts, while platform firms with unique strategies receive less coverage than equivalent traditional firms. The sign of the coefficient for the diversification dummies are also unexpected, as Litov et al. (2012) found significant negative coefficients for firms that were more diversified. The results are not very convincing regarding H1: Strategy uniqueness is only significant in absence of all control variables, and has very little explanatory power. Model four shows also no significance of strategy uniqueness for traditional firms. It is therefore not possible to reject the null hypothesis in this
16 Table 2: Results for the effects on analyst coverage
(1) Fixed
effects GLS (2) Fixed
effects GLS (3) Fixed
effects GLS (4) GEE
Analyst
coverage Analyst
coverage Analyst
coverage Analyst coverage
Strategy uniqueness -3.80*** -0.88 -0.95
(0.76) (1.36) (1.23)
Indicator for platform firms 7.21***
(1.57) Interaction term - strategy
uniqueness and platform firms -13.73**
(6.64)
Log of common equity 0.96* 0.95* 1.43***
(0.50) (0.50) (0.37)
Log of market capitalization 0.60* 0.60* 1.26***
(0.34) (0.34) (0.27)
Log of net sales 1.88*** 1.88*** 1.96***
(0.69) (0.69) (0.41)
Log of total assets -0.01 -0.01 -1.53***
(0.74) (0.74) (0.50)
Return on equity -0.03*** -0.03*** -0.03***
(0.01) (0.01) (0.01)
Share of intangible assets -1.37 -1.43 -4.42***
(1.98) (1.98) (1.23)
Amortization of intangible assets 0.29 0.29* 0.08
(0.18) (0.18) (0.14)
Share of depreciation in total
expenses 4.27 4.19 9.41**
(6.05) (6.05) (3.68)
Share of R&D in total expenses -12.92** -12.84** 8.62***
(6.44) (6.44) (3.11)
Sales growth in the last three years -0.82*** -0.83*** 0.36
(0.31) (0.31) (0.24)
Diversification dummy - 1 segment 1.63** 1.67** 1.33**
(0.73) (0.73) (0.65)
Diversification dummy - 2 segments 1.85** 1.91** 0.80
(0.76) (0.76) (0.67)
Diversification dummy - 3 segments 1.33* 1.41* 0.30
(0.77) (0.78) (0.68)
Diversification dummy – 4+ segments 0.79 0.88 -0.72
(0.83) (0.85) (0.72)
Constant 17.47*** -29.79*** -29.88*** -16.87***
(0.07) (7.77) (7.77) (5.70)
Observations 19,943 6,383 6,383 6,383
R-squared 0.00 0.20 0.20
P – value F test 0.00*** 0.00*** 0.00***
Number of companies 3,272 1,286 1,286 1,286
Standard errors reported in parentheses. Industry, year and country are controlled for but not reported in this table. GLS regressions used fixed effects at the firm level.
*** p<0.01, ** p<0.05, * p<0.1. All accounting variables are winsorized at the 95% level
17 case. There is more support evidence for a relationship to be found in the interaction term of model 4. Both coefficients regarding platform firms are significant and they also have the expected sign, leading to the rejection of the null hypothesis for H4.
5.2 Average analyst recommendations
The average analyst recommendation is the second dependent variable of interest. Again, we use four models, containing respectively only the variables of interest, a control model, the full model without interaction effects and a full model with interaction effects. The results are presented in table 3
The first model does not manage to explain variance in the analyst recommendation and produces an R-squared of 0.00. Strategy uniqueness has the expected sign, but is only marginally significant.
Modelling only control variables has more explanatory power and results in an R-squared of 0.09.
We observe that the standard deviation of analyst recommendations has a significant negative coefficient, which implies that stock analysts on average produce lower recommendations if the recommendations from their colleagues are more varied. The coefficient for analyst coverage is highly significant in this model, but does not have the expected sign, as analyst coverage had been found to increase market value in other studies, but implies a lower recommendation in our findings.
The third model shows significant negative coefficients for both explanatory variables. Again, this is an unexpected result for analyst coverage, but the coefficient for strategy uniqueness is in line with the anecdotal evidence quoted in Litov et al. (2012), expecting a lower recommendation for firms with a unique strategy, when the control variables are accounted for. The R-squared of this model remains at 0.09.
The last model shows no significant interaction term, even if the observed signs are in line with hypothesis 7. When modelling the interaction effect, strategy uniqueness remains marginally significant.
Based on these results, there is enough evidence to reject the null hypothesis for hypothesis 6, which stipulated a negative relationship between strategy uniqueness and the average recommendation, based on the significant result in model 3, which remained marginally significant when modelling the interaction. The results, however, provide no support for the interaction effect in hypothesis 7, which means that the null hypothesis cannot be rejected in that case.
18 Table 3: Results for the effects on average analyst recommendation
(1) Fixed effects
GLS (2) Fixed effects
GLS (3) Fixed effects
GLS (4) GEE
Analyst
recommendation Analyst
recommendation Analyst
recommendation Analyst recommendation
Strategy uniqueness -0.06 -0.14** -0.09*
(0.04) (0.07) (0.06)
Indicator for platform firms -0.03
(0.05) Interaction term - strategy uniqueness and
platform firms 0.24
(0.33)
Analyst coverage -0.00*** -0.00*** -0.00***
(0.00) (0.00) (0.00)
Standard deviation of analyst
recommendations -0.16*** -0.16*** -0.22***
(0.02) (0.02) (0.02)
Log of market capitalization 0.19*** 0.19*** 0.08***
(0.02) (0.02) (0.01)
Log of common equity 0.00 0.00 0.02
(0.02) (0.02) (0.01)
Log of net sales -0.05 -0.05 -0.05***
(0.03) (0.03) (0.02)
Log of total assets 0.03 0.03 0.03*
(0.04) (0.04) (0.02)
Return on equity 0.00*** 0.00*** 0.00***
(0.00) (0.00) (0.00)
Share of intangible assets 0.22** 0.21** 0.14***
(0.10) (0.10) (0.05)
Amortization of intangible assets -0.01 -0.01 -0.01
(0.01) (0.01) (0.01)
Share of depreciation in total expenses 0.16 0.15 -0.42***
(0.30) (0.30) (0.14)
Share of R&D in total expenses 0.25 0.26 -0.09
(0.32) (0.32) (0.11)
Sales growth in the last three years 0.05*** 0.05*** 0.10***
(0.02) (0.02) (0.01)
Diversification dummy - 1 segment 0.04 0.04 0.09***
(0.04) (0.04) (0.03)
Diversification dummy - 2 segments 0.02 0.03 0.09***
(0.04) (0.04) (0.03)
Diversification dummy - 3 segments 0.02 0.03 0.11***
(0.04) (0.04) (0.03)
Diversification dummy – 4+ segments 0.01 0.02 0.10***
(0.04) (0.04) (0.03)
Constant 3.44*** 1.15*** 1.13*** 2.14***
(0.00) (0.39) (0.39) (0.19)
Observations 19,943 6,320 6,320 6,320
R - squared 0.00 0.09 0.09
P – value F test 0.00*** 0.00*** 0.00***
Number of companies 3,272 1,280 1,280 1,280
Standard errors reported in parentheses. Industry, year and country are controlled for but not reported in this table. GLS regressions used fixed effects at the firm level.
*** p<0.01, ** p<0.05, * p<0.1. All accounting variables are winsorized at the 95% level
19 5.3 Market value of the firm
In this section, we test all hypotheses that predict an effect on the market valuation of a firm. The same structure of four models as in the previous sections is used. The outcomes of the models are presented in table 4. The first model again has a low explanatory value with an R-squared of 0.00.
Both the average recommendation by analysts and analyst coverage are significant in this model.
As expected, the coefficient of the average analyst recommendation is positive, but the coefficient for analyst coverage is negative. Strategy uniqueness does not yield significant coefficients and has an unexpected sign.
The control model adds a lot of explanatory power, resulting in an R-squared of 0.59. We see that large firms are negatively influencing the market-to-book value, and that having a high share of intangible assets also drives the market-to-book ratio down.
In the full model, the significance of the explanatory variables changes around compared to the first model. Now, strategy uniqueness has a significant negative coefficient, which is not what was expected. The average recommendation of analysts and their coverage are no longer significant.
When modelling the interaction effect, we find that the uniqueness coefficient remains significant, while the coefficient for platform firms is positive and also significant. The interaction term is marginally significant with a negative coefficient. This means that platform firms with a more common strategy are valued higher than a traditional firm would be, while platform firms with unique strategies are valued lower than an equivalent traditional firm.
In terms of hypotheses, we can conclude that our results did not meet our expectations based on the literature. H2 and H3 proposed a positive relationship between analyst coverage and market valuation, and strategy uniqueness and market valuation, respectively. In both cases, we observe the opposite relationship. The coefficient for analyst coverage was not significant beyond the first model, while the coefficient for strategy uniqueness was significant in both full models. In both cases, we cannot reject the null hypothesis. H8 predicted a positive relationship between the average analyst recommendation and the market valuation, but this coefficient fails to be significant beyond the first model, even though it does have the expected sign. Again, the null hypothesis cannot be rejected. Lastly, the interaction model expected the relationship between strategy uniqueness and the market valuation predicted by H3 to be weaker if the firm in question was a platform firm. In our results, we see that the unexpected negative relationship between uniqueness and the market to book ratio is stronger for platform firms. Therefore, the null hypothesis for H5 is also not rejected.
