The role of innovation, past performance and intangible assets in explaining firm performance

(1)

UNIVERSITY OF GRONINGEN

The role of innovation, past performance and

intangible assets in explaining firm performance

Master’s Thesis Finance Anouk van den Berg

S1919199 June 2013

Keywords: Industry effects, firm effects, innovation effects, past performance effects, intangible assets, value based performance measures, explained and unexplained firm performance

JEL Codes: G10 & G11

ABSTRACT

(2)

1 1. Introduction

The assessment of firm performance is an important input for the valuation of companies. The market value of a firm is the sum of the net present values of future expected cash flows of the firm (Koller et al., 2010). Firm value and firm performance are related since any management action or decision that increases cash flows, indeed performance, increases firm value. A change in firm performance causes a change in the value of the firm (Koller et al., 2010). Financial measures provide CFOs the most precise insight in the performance of the firm (Van der Stede et. al., 2006). This study examines which factors influence market-to-book values and indeed explain firm performance. Performance is seen from an investor’s perspective or from an operating perspective in this study. As a result, the outcome of this study contributes to a better understanding of what explains firm’s value and enables CFO’s to make the right capital allocating decision (Koller et al., 2010).

The debate in the literature regarding performance studies argues whether performance is merely driven by industry factors or firm characteristics. In the industry organization theory, industry structure is the main determinant of firm performance (Porter, 1980). However, how come that some firms that face identical industrial conditions of supply and demand and operate in the same market structure, still perform better than other firms (Nelson, 1991)? To better explain this, Barney (1991) comes up with a different theory regarding firm performance, namely the resource based view. In the resource based theory, the firm’s rare resources determine firm performance. Though both views highlight the importance of different factors that explain performance of a firm, the two theories should be seen as two sides of the same coin.

(3)

2

From previous studies I conclude that I cannot expect a general firm fact in combination with a general industry fact to completely explain the variance in firm performance (Schmalensee, 1985; Rumelt, 1991; McGahan and Porter, 1997; Hawawini et al., 2003). In order to lower the unexplained variance of performance, I extend the model of Hawawini et al. (2003) by adding an extra factor, firms’ innovativeness. The empirical literature finds that innovative firms earn large part of the industry’s profits, especially in this rapidly changing environment (Kim and Mauborgne, 1997; Gadies and Gilber, 1998). In addition, Teece et al. (1997) find that recent winners in the global world are firms that demonstrate timely responsiveness, rapid and new innovation. Therefore, I expect a relation between firm’s innovativeness and performance. However, the long term persistence of those excess profits generated by a rare resource is doubtful. Koller et al. (2010) find that for the long term, returns on invested capital (ROICs) gradually converge to an industry average for US firms between 1965 and 2010. Nissim and Penman (2001) also find partly convergence of excess returns to an economy wide average for US companies between 1963 and 1999. Both empirical researches find industry structure to be a main determinant in the variation of a firm’s ROIC and support the industry organization view. However, Nissim and Penman (2001) find quicker convergence than Koller et al. (2010) did. Following this line of reasoning I expect reversal of performance, either in the long or short term and therefore a relation between past and current firm’s performance. In order to improve the explanation of the variance in firm performance, I add a proxy for past performance to the model of Hawawini et al. (2003).

In this study, I use market-to-book ratios as the performance measures. Market-to-book ratios can be measured in many ways and different ways of measuring provide insight in different aspects of performance. Investors are mostly interested whether they earn a decent return on their investment. From their perspective, book values in market-to-book ratios should include the book value of acquired goodwill, the cumulated amortization and the impairments of acquired goodwill and intangible assets if market-to-book ratios are used as the performance measure. In this calculation, market-to-book ratios consider external growth generated by mergers and acquisitions. When assessing operating performance, acquired goodwill should be excluded from the invested capital (Koller et al., 2010).

(4)

3

Amazon.com was the best performing firm of all the firms in S&P 500 in terms of unadjusted market-to-book ratios. However, Amazon.com didn’t belong to the top-10 best performing firms anymore once the book values in the market-to-book ratios were adjusted for the value of acquired goodwill. Based on this finding, I use both performance measures to examine whether firms’ performance can be better explained when book values in market-to-book ratios are adjusted for acquired goodwill. Adjusting the book values in the market-to-book ratios for the book value of intangible assets is the third extension of this study since previous performance studies use the unadjusted book ratios, hereafter named market-to-book ratios, as performance measure (Hawawini et al., 2003).

In short, this study contributes to the literature by exploring whether the addition of a proxy for past performance and innovativeness to the model of Hawawini et al. (2003) and adjusting book values in the market-to-book ratios for the book value of acquired goodwill, as I expect, decreases the unexplained firms’ performance; a relevant issue to solve since insight in this problem improves capital allocating decisions of CFOs. The study is based on data of S&P 500 from 1990-2012.

The rest of the paper is organized as follows. In the next section we give a brief overview of the relevant literature. Section 3 documents on the dataset. The methodology is discussed in section 4. In section 5 we report, interpret and discuss the results. Concluding remarks including suggestion for further research are presented in section 6.

2. Literature review

2.1 Performance measurement

(5)

4

should therefore be considered in assessing firm performance (Koller et al., 2010). The advantage of market value based measures is that these measures measure the net present value of the expected future profits. Contrary to accounting measures, such as return on assets, market value based measures do filter one-time occurring events which is preferable. Value based measures are especially preferable in this study since value based measures partly solve the problem that there is time between the moment R&D is expensed and the moment R&D generates its economic profits. Because the effects of R&D expenditures are time lagged, their effect on firm performance is hard to measure (Gunday et al., 2011). The market value of a firm reflects the future expected returns to R&D (Toivanen et al., 2002). Investors’ expectations of R&D returns are only influenced by new information about R&D expenditures and patenting (Griliches, 1981). As a result, past R&D expenditures are already reflected in the market-to-book values. Therefore, this study can relate R&D expenditures and market-to-book values of the same year when market value based performance measures are used (Tidd, 1991; Hall, 1993). As a result of these arguments, we prefer value based measures even though Bergsma (2012) finds that the choice between accounting and market value based measures doesn’t affect the magnitude of the effect sizes and unexplained variance in firm performance.

This study uses two different enterprise market-to-book ratios to measure firm performance. In general, market-to-book ratios larger than one imply that the present value of the expected future economic profits is positive. Both ratios are based on the enterprise market value instead of equity market value. Firms with different levels of debt but similar operating activities have equal enterprise market values but different equity market values. Since firm value is determined by operating activities, enterprise values are used in this study.

The first measure used in this study is the firm’s total market value per capital invested, where the total market value is the sum of the firm’s market capitalization of equity and the firm’s market value of debt and where capital invested is the sum of the book value of debt and equity. This measure assesses firm performance from the perspective of investors.

(6)

5

The adjustment of book values for acquired goodwill makes the indicator useful to evaluate different aspects of performance and decisions of managers. The unadjusted market-to-book ratio tells whether investors earn a decent return on their capital invested. The market-to-book ratios in which book values are corrected for acquired goodwill provide information on the competiveness of the underlying business. This measure can better be used to compare firm operating performance with competitors since this measure is not distorted by firms’ acquisition activities (Koller et al., 2010). In addition, implementation of new accounting systems changed the way companies have to account for acquired goodwill. For this reason, adjusting book values in market-to-book ratios for acquired goodwill improves the comparison of performance over the years.

Our dataset supports the fact that both indicators measure a different aspect of performance. Based on the results of the Wilcoxon signed-ranking test, I conclude that the ranking of best performing companies is significantly different for the two performance measures (P-value =0.000).

2.2 The model of Hawawini et. al. (2003)

In this study I extend the model of Hawawini et al. (2003) by adding a proxy for firms’ past performance and innovativeness to the model and adjusting book values in the market-to-book ratios for the market-to-book value of acquired goodwill. Since this study is an extension to the model of Hawawini et al. (2003), I explain the model that Hawawini et al. (2003) use to measure the contribution of industry and firm factors first. What is exactly meant by the contribution of industry factors and firm factors in their study is also explained and important to know for a good understanding of this study. Thereafter, I elaborate on the effects that I add to the model, namely past performance and innovativeness.

(7)

6

The contribution of industry factors in explaining firm performance is based on the industry organization view. In this framework, the structural industry characteristics are the primary determinants of firm performance (Cabral, 2000). The industry organization theory is based on the structure-conduct-performance framework. This framework shows that the market structure, also explained as the concentration of buyers and sellers within the industry, influences the behavior of firms, called the conduct of firms. The conduct of firms determines firm performance. A development of the structure-conduct-performance framework led to the five forces model of Porter. In the model of Porter (1980), firm performance is explained by five forces: competitive rivalry, the threat of substitute products, the threat of new entrants, the bargaining power of suppliers and the bargaining power of customers (Porter, 1980). In the industry organization view, industry structure effects the overall performance of the industry and therefore the performance of firms within the industry (Galbreath and Galvin, 2008).

If industry factors mostly explain firm performance, why do some firms within the same industry perform better than other firms? The industry organization perspective isn’t able to explain the difference between performances of firms that operate in the same industry. For this reason, Wernerfelt (1984) explores the usefulness of analyzing firms in terms of their resources rather than in terms of their products and external factors. This resulted in a shift in focus from industry factors to individual firm factors for the analysis of firms’ performance in the 1980s (Hoopes et al., 2003).

The main reason for intra-industry differences is according to Nelson (1991) quite simple and can be explained by the fact that different choices are profitable for different firms due to differences in initial conditions of firms.

(8)

7

Galbreath and Galving (2008) argue that firm factors become more important than industry factors in explaining variance of firm performance due to the current rapidly changing environment. Industry boundaries become more blurry, converge and overlap continuously (Bettis and Hitt, 1995). Firms nowadays can outperform when they have the capability to focus on leveraging and integrating their resources with external factors rather than being positioned in attractive industries.

(9)

8

Table 1: Results component variance analysis of relevant previous studies

Study Measure Number of industries used Industry effects Firm effects Year effects Industry-Year effects Country effects Unexplained firm performance Schmalensee (1985) ROA 242 19.5 0.6 NT NT NT 80.5 Rumelt (1991) ROA 242 4.0 45.8 NT 5.4 NT 44.8 McGahan and Porter (1997) ROA 628 18.7 36.0 2.4 NT NT 48.4 Hawawini et al. (2003, Full sample) ROA 55 8.1 35.8 1.0 3.1 NT 52.0 ROIC-WACC 55 6.5 27.1 1.9 4.2 NT 60.3 M/B 55 11.4 32.5 1.3 2.9 NT 51.9 Hawawini et al. (2003, Excluding top 2 out- & underperforming firms) ROA 55 17.6 12.4 1.9 5.8 NT 61.0 ROIC-WACC 55 17.0 30.2 2.5 5.0 NT 45.3 M/B 55 16.7 16.0 1.1 4.1 NT 62.1 Bergsma 2012 (2012, Full sample) ROA 18 5.4 29.3 1.8 3.1 5.8 54.5 ROIC 18 0.6 21.3 0.0 0.0 1.0 77.5 ROIC-WACC 18 0.5 21.4 0.0 0.0 1.0 78.0 M/B 18 0.0 12.1 0.0 0.0 0.0 87.9 M/B2 18 3.1 63.4 0.7 0.8 0.9 31.1 Tobin's Q 18 14.2 40.5 1.6 3.6 1.1 39.1 Bergsma (2012, Excluding top 5 leaders and losers from each industry) ROA 18 9.6 14.1 2.7 4.5 1.6 65.7 ROIC 18 0.5 27.0 0.6 3.3 1.0 58.6 ROIC-WACC 18 0.6 73.1 0.6 1.1 0.5 19.6 M/B 18 0.0 28.9 0.9 2.8 0.5 50.9 M/B2 18 10.4 23.0 3.7 4.7 1.5 52.1 Tobin's Q 18 22.6 22.2 2.3 5.2 1.0 42.6

ROA, return on assets; ROIC, return on invested capital; WACC, weighted average cost of capital; ROIC-WACC, equal to the economic value added; M/B, unadjusted market-to-book ratio at enterprise level; M/B2, unadjusted market-to-book ratio at equity level; Tobin's Q, enterprise value divided by replacement value of total assets; NT, not tested. Results of the effects are presented as a percentage of total variance of firm performance.

(10)

9

firm effect is measured in this study. Schmalensee’s (1985) only measure of firm effects is market share while in other studies firm effects cover more than just market-share. He reports a positive, but negligible, effect between market share and firm performance. Rumelt (1991) extends the study of Schmalensee (1985) by using data on the same firms over more than year, respectively covering the period from 1974-1977. In the study of Rumelt (1991) firm effects cover all business-unit effects and not just market-share effects. This explains the fact that Rumelt (1991) finds firm effects to dominate industry effect.

As shown in table 1, industry effects also become larger when more industries are used in the study. When there are more industries used in a study, the study uses a narrow definition of the industry. A narrow definition would lead to a strong- industry effect while a broad industry definition would result in a less significant industry effect (Hawawini et al., 2003). For this reason, the definition of an industry is a topic of discussion and the number of industries used in a study needs to be taken into consideration when interpreting the results. In addition, the database used in the several studies, may also explain the mixed results of the empirical literature. Contrary to the studies of Schmalensee (1985) and Rumelt (1991), the other studies presented in table 1 base their study on Compustat data instead of data from US Federal Trade Commission line of business dataset (FTC). Compustat reports contain information on more firms, not on only large manufacturing firms as the FTC dataset contains, and on business segments instead of business units. Therefore, studies that used Compustat data need to rely on a broader classification system, named the SIC system, for industry classification and this may further diminish industry effects. Appendix A contains an overview of the data and methods used in the relevant previous studies that are shown in table 1.

Hawawini et al. (2003) also examine the role of out- and underperformers within the industry. They find that firm effects are in particular the dominant factors for the out- and underperforming firms within the industry. For firms that are neither out- nor underperforming firms, called firms with an average performance, industry effects become relatively more important although firm effects remain the dominating factor.

(11)

10

interest of this study. Management practice also indicates that corporate strategies can play a negative role and since variance-based analysis cannot estimate negative effects, this study cannot properly measure the corporate effect (Hawawini et al., 2003). For these reasons, corporate effects are not separately included in our studies and these effects are part of the firm effect.

2.3 New effects: Innovativeness and past performance

In the previous section I explained the model of Hawawini et al. (2003). In this model firm performance is explained by an industry and a firm factor. Based on the large unexplained performance in this model, I can’t expect a general firm fact in combination with a general industry fact to completely explain the variance in firm performance (Schmalensee, 1985; Rumelt, 1991; McGahan and Porter, 1997; Hawawini et al., 2003). In order to lower the unexplained variance in performance, I further specify the firm and industry effects in the model of Hawawini et al. (2003) by adding two effects, namely the innovativeness and past performance of the firm.

2.3.1 The innovation effect

A decision to innovate is a net present value decision. A firm has to decide whether they want to invest in their unique resources (Nelson, 1991). Investing in the unique resource can generate excess returns for the firm, at least in the short run. When legal rights prevent competitors from imitating the resource, even longer excess returns can be generated. When the expected generated future returns are higher than the cost of the investment, firms will decide to invest in research & development (R&D) since the investment will increase the market value of the firm.

(12)

11

A decrease in costs can be achieved by more efficient processes. More efficient processes can be explained with the concept of the learning curve. Firms that have high frequency and consistency in adopting innovation progress faster along the learning curve and become more efficient than firms that focus less on innovation, at least on the short term (Mahoney and Pandian, 1992; Subramanian, 1996). The efficiency advantage is sustainable for the long run and explains some variance in firm performance when again the improvement in either a process or technology cannot be imitated by competitors (Diericks and Cool, 1989; Wernerfelt amd Montgomery, 1988).

Thus, unique resources are a necessary condition to earn these excess returns, but not a sufficient condition. Market conditions determine whether the unique resource is valuable. When market conditions determine that the unique resource is valuable, investment in the resource increases profit margins and positively affects the market value of a firm (Bosworth and Rogers, 2001). For this reason, I expect innovation to play a role in the explanation of firm performance.

The challenging question is how firms’ innovativeness can be properly measured. Past research examines whether either R&D expenditures, number of patents, value of patents or a combination of these measures can best be used to measure firms’ innovativeness. The results of these studies are mixed. Most studies agree on the fact that the indicator, R&D expenditures to sales, is a proper proxy for innovation. The variable reflects how important it is for a firm to devote resources to R&D and invest unique resources which is a necessary condition to earn excess returns (Chan et al., 2001).

(13)

12

than R&D expenditures for the high-technological industries (Tidd, 2001). Tidd (2001) confirms a positive relationship between R&D expenditures to sales and firm performance in terms of market-to-book values.

Other studies find that patenting activity always adds information above and beyond that obtained from R&D expenditures and find a positive relationship between increasing R&D expenditures, patenting activity and the market value of a firm (Griliches, 1981; Chan et al., 2001; Anandarajan et al., 2008; Franko 1989; Bosworth and Rogers, 2001; Megna and Klock, 1993). In those studies the number of patents measures the “success” of a R&D program (Hall et al., 2005).

Thus, empirical literature is not clear about what measures are the best proxies for innovation. Theoretical literature implies to use both measures as a proxy for innovation. The measure, R&D expenditures to sales, indicates the amount that is invested in a unique resource. The number of patents granted indicates whether resources are protected and whether the investment is successful. Since investing in a unique resource and the protection of this resource, is both a necessary condition for innovation to result in excess returns, theoretical literature implies to use both measures as a proxy for innovation. Due to the deviating implications in the empirical literature I use a model in which innovation is proxied only by R&D expenditures to sales, only by patents granted and a model in which innovation is proxied by R&D expenditures to sales and the number of patents granted.

2.3.2 The past performance effect

(14)

13

the return on invested capital. Koller et al. (2010) find that returns of invested capital revert toward to an average industry level on the long run. Companies earning high returns on invested capital tend to see their returns fall and companies earning low returns see them rise over time (Koller et al., 2010). Penman and Nissim (2001) find even quicker reversal of returns to an overall economic average. Both studies find reversal of returns and state that returns of extreme out- and underperforming firms don’t completely revert to the mean, but do increase or decrease towards a mean level. The fact that excess returns are mean reverting implies a relationship between the past, current and future performance of a firm. For this reason, I expect past performance to explain, either on the short or long term, variance in firm performance. Therefore, we add a past performance effect to the model of Hawawini et al. (2003) in order to diminish the unexplained variance in firm performance.

3. Data

The data used to calculate market-to-book ratios and R&D expenditures to sales are extracted from the Thomson Reuters Datastream database. Information about the number of patents per firm and industry is extracted from Orbis. The dataset contains US financial and non-utility firms of the S&P 500 index covering the years from 1990 till 2012. Table 2 contains information about the sources and definitions of the different variables used in this study. Table 3 documents the calculations of the variables used in this study.

(15)

14

total firm performance (Hirschey and Wichern, 1984). Though, we conclude that both measures are non-optimal proxies for firm performance, this study examines which measures of the two better explains the variance in firm performance.

Table 2: Variable definitions and sources*

Variable name Symbol Sources

Market capitalization common equity

MC Datastream code WC08001

Book value debt BVD Datastream code WC03255

Market value debt MVD Book value debt

Book value minority interest MI Datastream code WC01501

Book value preferred stock PS Datastream code WC03451

Book value common equity CE Datastream code WC03501

Book value intangible assets IA Datastream code WC02649

Book value acquired goodwill AG Book value intangible assets

Cash and cash equivalents CCE Datastream code WC02001

R&D expenditures to sales per firm, j and year, t

Datastream code WC08341

R&D expenditures to sales per industry, i and year, t

Datastream code WC08341

Sales Sales Datastream code WC01001

Number of patents granted to an

industry, i, and in year, t, Orbis

Number of patents granted to a

firm, j, and in year, t, Orbis

* All data is in US dollars; balance sheet and market data are end of year data.

Table 3: Variable definitions and calculations*

Variable Name Symbol Calculation

Working Cash WC 0.02 * Sales

Excess cash EC CCE - WC

Market-to-book ratio M/B

Adjusted market-to-book ratio Adjusted M/B R&D expenditures to Sales per

firm scaled by R&D expenditure of the industry

R&D to Sales

Patent ratio per firm per year PR

(16)

15

As argued in the literature section, innovation is proxied by the number of patents granted to a firm and by R&D expenditures of a firm. For a comparable measure of patenting activity, the patents granted to a firm should be related to the total number of patents granted to the entire industry of that firm, because patenting activity differs per industry (Cockburn and Griliches, 1988). As presented in graph 1, relative more patents are granted to the travel & leisure-, oil & gas and health care industry. Because of the difference in patenting activity, the variable, hereafter named the patent ratio, is calculated by dividing the total number of patents granted to a firm per year by the total number of patents granted to the industry of that firm for the same year. The value of patents would have been a better proxy for innovation since value of patents can change per year and per patent (Hall et al., 2005), but no data is available this variable yet.

Graph 1: Average number of patent granted to an industry for the years 1991 till 2012, scaled by the number of firms within each industry

Different industries imply different deprecation rates of R&D investment and different firm sizes spent different amounts on R&D in absolute terms as shown by graph 2. Firms within the health care and the technology industry spent on average more on R&D than firms within the other industries. For this reason, R&D expenditures of a company should be scaled by the average R&D expenditures of the industry and by the total sales of the firm to make this variable comparable between firms (Griliches and Cockburn, 1988; Franko, 1989).

13.15

7.95

3.14 2.88 2.72 2.65

(17)

16

Graph 2: Average R&D expenditures to sales (%) per industry for the years 1991 till 2012

Firms are classified into industries according to the ICB framework. This framework, developed by FTSE and Dow Jones in 2005, segregates firms into groups within the macro economy and contains four levels of classification. The classification of the first level divides firms into 10 industries. The second level segregates firms into 20 so called super sectors. The third and fourth level of the ICB framework respectively contains 41 and 114 sectors. Due to the relatively small sample size of our data, I’m only able to examine industry effects based on level 1 and level 2 classifications. If I would use a more narrow industry classification, some industries would contain not enough data for further statistical analysis and would have to be left out, which is undesirable. Whether firms where classified according to a level 1 or a level 2 classification didn’t alter the conclusions. Hawawini et al. (2003) segregated firms into 55 industries. To make our study as comparable as possible with Hawawini et al. (2003), I classified firms according to the level 2 classification instead of the level 1 classification. So firms are segregated into 20 super sectors. Appendix B shows how many firms are classified to a certain industry.

A limitation of the ICB frameworks is that this system classifies firms based on the firms’ major source of revenue. This system is supply side oriented and ignores the demand side dimension. Ignoring other dimensions such as the consumer segmentation on the demand side can result in a wrong classification of strategically relevant industries (McGahan and Porter, 1997). Therefore, one can argue whether the estimates of the industry effects on performance are reliable when industries are classified following the ICB classification. There are no other options that do not suffer from similar problems available. Therefore, I, just as other empirical

(18)

17

studies (see table 1 for the most relevant ones) did, have to depend on the ICB system for industry classification. As a consequence, results regarding the size of industry effect should be interpreted with some caution (Hawawini et al., 2003).

I make some adjustments to the dataset. First, I exclude firms within the utilities and financial sector. Firms within the utility sector are regulated and firms within the financial sector have a different balance sheet. For these reasons, firms within the utility and financial sector are deviating too much from the other sectors to include in the analysis. All observations for which values are missing are excluded. In a few cases, adjusting the book value for intangible assets led to negative adjusted market-to-book ratios. These negative observations are excluded from the dataset (Nissim and Penman, 2001). Secondly, I remove the statistical outliers from the dataset following the guidelines for estimating variance component effect of Cohen et al (2003). Data points that are more than 3 standard deviations away from the median of the entire sample of the variable are classified as statistical outliers and removed from the dataset.

This study focuses on the performance of the average firm. For that reason, out- and underperforming firms are excluded. In line with Hawawini et al. (2005), I classify firms that have an average market-to-book ratio, which is more than three standard deviations deviated from the median of either the adjusted or unadjusted market-to-book ratio of the industry, as an economic outlier. In total I exclude all observations of 9 firms from the dataset. The remaining dataset contains 8.614 observations spread over 355 firms, 14 industries and 22 years. Table 4 reports statistics describing the data sample.

(19)

18

The large skewedness statistics indicate that the distribution is not symmetric around its mean value for all variables. The skewedness statistics are positive which means that the right tail is longer than the left tail of the distribution. All observations containing negative market-to-book ratios are excluded from the dataset and the dataset contains only firms that survived during the entire period. As a result, this study, just as almost all performance studies, encloses an inevitable survivor bias. Because of this, most observations of market-to-book ratios are concentrated on the left. The kurtosis statistics are larger than three which implies that the distribution has a sharper peak and fatter tails than a normal probability distribution for all variables (Brooks, 2008). All Kolmogorov-Smirnov test statistics are significant. The total data sample is not normally distributed and this implies that there are significant differences between groups of the variables. This test doesn’t test whether the sub-samples aren’t non-normal distributed either. Therefore, the assumption of parametric tests that data of the sub-groups of the effect should be normally distributed isn’t violated by the outcome of this Kolmogorov-Smirnov test.

Table 4: Descriptive Statistics of the data sample used in this study

Variable* Minimum Maximum Mean Std. Dev.

Median Skewness Kurtosis Kolmogorov-Smirnov** Market-to-book ratio 0.54 17.54 3.15 2.23 2.49 2.31 7.08 10.59 Adjusted market-to-book ratio 0.67 324.44 5.97 11.26 4.08 17.13 381.77 22.23 R&D expenditures to sales 0.00 95.72 4.36 6.54 1.62 2.82 16.44 12.28 Patent Ratio 0.00 1.00 0.05 0.13 0.00 4.41 21.30 24.82 Past Performance market-to-book ratio 0.54 18.02 3.21 2.35 2.50 2.35 7.28 10.82 Past Performance adjusted market-to-book ratio 0.66 302.68 5.66 8.88 4.01 18.60 497.34 20.15

(20)

19

Graph 3: Trend market-to-book ratios from 1991 till 2012

M/B, market-to-book ratio calculated as shown in table 3. Correlation coefficient is 0.24.

Lastly, I change the ratio data on R&D expenditures to sales, patent ratio and past performance to categorical data. The component variance analysis can only estimate the sizes of the effects when data of the effects is categorical data (Chakravarti, Laha and Roy, 1967). For this reason, I categorized the data of R&D expenditures to sales, the patent ratio and past performance of firms into ten groups. Firms were given an index deviating from 1 till ten for each variable. Larger or smaller category sizes did not alter the results and conclusions. To consider the robustness of our findings, I perform the same analysis as described in the next section on a dataset for which I define the statistical outliers differently using the same method as Nissim and Penman (2001). The upper and lower one percent of the observations per variable is excluded from the dataset. This remaining dataset contains 3075 observations spread over 222 firms, 14 industries and 21 years.

4. Model & methodology 4.1 Model

I base my analysis on the same model as Hawawini et al. (2003), described in section 2.2, used in their study to explain the variation in firm performance. In order to decrease the unexplained variance in firm performance I extend the component variance analysis model of Hawawini et al. (2003) by adjusting the performance measure and by adding a R&D effect, patent effect and past performance effect as independent variables to the model.

0 1 2 3 4 5 6 7 8 9

(21)

20

I use the following model to explain the firm performance:

= (1)

The dependent variable is the performance measure for firm j’s operations in industry i and year t: adjusted market-to-book and market-to-book ratios. In this model is a constant;

is the stable industry effect; is the stable firm effect; covers the year effect;

indicates the patent ratio; is the R&D effect; is the past performance effect; is the interaction between the main year and industry effects and is the unexplained variance in firm performance.

4.2 Methodology

The component variance analysis used in the study of Hawawini et al. (2003) is the statistical methodology in this study to measure the effect sizes. The size of an effect indicates how much of the variance in the dependent variable, the market-to-book ratio, is explained by the independent variable such as the R&D effect. However, the component variance analysis doesn’t test whether the independent effects are significant. For this reason, I first test whether there is a significant difference between groups of the several effects. If the differences between groups of an effect are statistically significant, the effect is included in the component variance analysis.

The first test I use to test for significance of the independent effects is the One-way ANOVA. This test assumes data of each group to be normally distributed and variances between the groups to be equal. If the distribution of the data of a group is non-normal, one needs to use the non-parametric version of the One-way ANOVA, the Kruskal Wallis test, to test whether the ranks of means of market-to-ratios are significant different between the groups. In most cases, non-parametric tests apply the same procedure as parametric tests do; as a difference, non-parametric tests use ranks of the data instead of the data itself. This rank transformation makes the test less sensitive to outliers (Conover and Iman, 1981). You can use a Kolmogorov-Smirnov test to assess the normality of the data. Since testing for a non-normal distribution of each sub-group is time consuming, I decide a priori to use parametric and non-parametric tests to test for significance of the independent effects.

(22)

21

whether there are significant differences between groups (Brown and Mood, 1951; Kruskal and Wallis, 1952; Gibbons and Chakraborti, 2003). The Mood’s median test tests whether the median rank of the market-to-book ratios differs between the groups. This test has even less assumptions than the Kruskal Wallis test and besides normality this test doesn’t require same distributions of the groups either; though, this test is less powerful than the Kruskal Wallis test (Gibbons and Chakraborti, 2003).

A non-parametric Levene’s test is used to test whether the variance of variables of interest are the same for all sub-samples. The non-parametric Levene’s F-test has more statistical power than the median based Levene’s test when population distributions are skewed and sample sizes are unequal. Because of the large skewness statistics shown in table 2, the non-parametric Levene’s test is preferable in our study (Norstokke and Zumbo, 2010). When the p-value of the non-parametric Levene’s test is smaller than 0.05, the hypothesis of homogenous variability should be rejected.

By using a One-way ANOVA, Kruskal Wallis test and Mood’s median test, I hedge against non-normality of the sub-groups and unequal variances between the sub-groups. If all three tests result in a p-value smaller than 0.05, the independent effects are statistically significant. Only the statistically significant effects are included in the component variance analysis. In line with Hawawini et al. (2003) I measure the separate effect sizes with the component variance analysis. This analysis estimates the relative contribution of each random effect to the total variance of the dependent variable. The equation of the component variance analysis is based on the descriptive model of equation 1 and assesses the variance of the dependent variable, i.e. performance measures, as the sum of the variance of the separate effects.

(2) The are the main effects, is the interaction effect and is the unexplained firm performance. The symbols in equation 2 have the same meaning as the symbols in equation 1. All the effects described in equation 2 follow a normal random distribution with a mean of zero and a constant variance (Cox and Solomon, 2003).

(23)

22

The component variance analysis provides insight in what areas to focus on in order to reduce variance. The main focus of this study is to decrease the unexplained variance in firm performance, indicated by the term in equation 2. The estimates of the individual effect sizes are of indirect importance. Component variance analysis allows effects to be fixed or random, as long as at least one of the factors is random. Effects are random when observations are attributable to an infinite set of levels of a factor, of which a random sample occurs in the dataset. Effects are fixed when the effects are attributable to a finite set of levels of a factor that all occur in the dataset (Searle, Casella and McCulloch, 1992).

The disadvantage of a model without fixed effects is that this model does not test whether the effects are statistically and economically significant. I use the One-way ANOVA, Kruskal Wallis and Mood’s median test to test whether there are statistically significant differences between groups. I use the size estimate of the effect, measured by the component variance analysis, as a proxy for the economic significance of the effect (Roquebert et al., 1996). In that way the disadvantage of a model without fixed effects is partly tackled.

The advantage of the component variance analysis only comprising random factors is that the results may be generalized over the whole population even though the dataset used only contains a sample of the population and unequal groups with missing values (Hawawini et al., 2003). For this reason, I prefer to assume all effects to be randomly generated, which is in line with the assumptions of the component variance analysis and with the studies presented in table 1 (Schmalensee, 1985; Rumelt, 1991; McGahan and porter, 1997 and Hawawini et al., 2003).

5. Results

(24)

23

ANOVA and Kruskal Wallis test if you want to be sure that the independent effects are indeed significant.

Table 5: Are the means of firm performance between the groups of the effects significant different as tested by the One-way ANOVA test?

Performance measure Statistic Year effect Industry effect Firm effect R&D effect Patent effect Past performance effect Market-to-book ratio F-Statistic 9.42 36.62 14.86 8.48 5.05 877.54 P-Value 0.00 0.00 0.00 0.00 0.00 0.00 Adjusted market-to-book ratio F-Statistic 2.86 9.95 3.27 2.34 0.99 207.29 P-Value 0.00 0.00 0.00 0.00 0.45 0.00

Sample size is 4.681 observations in both scenarios.

Table 6: Are the means of firm performance between the groups of the effects significant different as tested by the Kruskal Wallis test?

Performance measure Statistic Year effect Industry effect Firm effect R&D effect Patent effect Past performance effect Market-to-book ratio Chi- statistic 243.92 544.65 2908.84 73.51 87.54 3019.23 P-Value 0.00 0.00 0.00 0.00 0.00 0.00 Adjusted market-to-book ratio Chi- statistic 209.18 726.58 3084.44 117.30 52.62 1095.36 P-Value 0.00 0.00 0.00 0.00 0.00 0.00

Table 7: Are the medians of firm performance between the groups of the effects significant different as tested by the Mood's median test?

Performance measure Statistic Year effect Industry effect Firm effect R&D effect Patent effect Past performance effect Market-to-book ratio Chi- statistic 169.84 398.31 2342.56 42.27 52.83 2276.49 P-Value 0.00 0.00 0.00 0.00 0.00 0.00 Adjusted market-to-book ratio Chi- statistic 148.71 559.41 2442.63 99.03 41.49 555.30 P-Value 0.00 0.00 0.00 0.00 0.00 0.00

(25)

24

this model are presented on the first row in table 8. To solely show the effects of the addition of the new effects, the models in which the book values in market-to-book ratios are adjusted, are compared to the model that includes the same effects as Hawawini et al. (2003) but in which book values are also adjusted. This model is presented on the ninth row in table 8. The first point to discuss is that our unexplained variance is much lower than the unexplained variance in the literature. In our model, the unexplained variance is 36.98% whereas the unexplained variance in the literature in Hawawini et al. (2003) is more than 50.0%. In line with the literature, I find that firm effects dominate industry effects although I find larger firm effects and smaller industry effects than Hawawini et al. (2003). These deviating results could be explained by the number of industries used in this study and by a different sample of years. I use more recent data than Hawawini et al. (2003) did which can cause firm effects to become more important (Bettis and Hitt, 1995). The use of a smaller data sample in this study could also explain the deviating findings. In addition, 14 industries are used in this study whereas Hawawini et al. (2003) used 55 industries in their study which can explain the lower industry effects found in this study (Claver et al. 2002).

(26)

25

Table 8: Effects in percentage of the total variance of the dependent variable as measured by the component variance analysis

Performance measure

Industry effect Firm effect Year effect Transient industry effect

R&D effect Patent effect Past Performance effect Unexplained variance in firm performance Decrease in unexplained variance compared to Hawawini et al. (2003) Market-to-book ratios 8.44 47.68 3.49 3.42 36.98 0.00 8.20 45.65 3.41 3.29 3.72 35.72 1.26 8.76 47.66 3.46 3.37 0.11 36.69 0.29 0.31 2.14 1.04 1.37 83.25 11.89 25.09 8.47 45.53 3.38 3.25 3.81 0.11 35.45 1.53 0.23 2.63 1.02 1.34 1.99 81.73 11.57 25.41 0.31 2.14 1.04 1.37 0.00 83.25 11.89 25.09 0.32 2.03 1.02 1.34 1.99 0.00 81.73 11.57 25.41 Adjusted market-to-book ratios 2.79 12.77 1.49 0.00 82.94 0.00 2.79 12.67 1.48 0.00 0.00 82.89 0.05 2.79 12.77 1.49 0.00 0.00 82.94 0.00 0.07 0.36 0.04 0.00 91.18 8.35 74.59 2.86 12.65 1.48 0.00 0.20 0.04 82.77 0.17 3.29 5.53 0.83 0.36 0.23 81.98 7.78 75.16 0.07 0.36 0.04 0.00 0.00 91.18 8.35 74.59 0.06 0.36 0.04 0.00 0.02 0.00 91.19 8.34 74.60

(27)

26

Adding the R&D effect and the patent effect to the model hardly reduces the unexplained variance. Since I assume the size of the estimates as a proxy for economic significance, I state, based on table 8, that the small transient industry-, year-, R&D- and patent effect estimates, varying between 0.00 and 3.81%, are insignificant in economic terms (Roquebert et al., 1996). With our proxies for innovation, only the creation of innovative products and processes is considered. Even though, a large part of innovation doesn’t have to be self-created, as long as new ideas from outside are adapted within the organization. It might be the case that larger innovation effects are found when the creation as well as the adaption of innovative products and processes is considered (Subramanian, 1996). Only considering the creation of innovative products could be a reason for the unexpected small sizes of the R&D and patent effect. Another issue that could explain the small effect size of patents is that the patent ratio of a firm in this study is based on the number of patents granted to firm. This variable doesn’t consider the value of a patent even though this value can differ per year and patent. Patents that are more valuable, generate more excess returns for a firm than other patents. Therefore, value of patents granted to a firm might be a better proxy for innovation and better explain variance in firm performance than the number of patents granted to a firm (Hall et al., 2005). For this reason, the actual effect size of granted patents might be larger than is measured in this study. Further research could reconsider the patent effect when the value of the patents granted to a firm is taken as a proxy for innovation.

(28)

27

Table 9: Decrease in unexplained performance compared to the model of Hawawini et al. (2003) caused by the adjustment of book values for the book value of intangible assets.

Model Effect size

Model of Hawawini et al. (2003) -45.96

Model of Hawawini et al. (2003) + R&D effect -47.17

Model of Hawawini et al. (2003) + Patent effect -46.25

Model of Hawawini et al. (2003) + Past performance effect 3.54 Model of Hawawini et al. (2003) + R&D and patent effect -47.32 Model of Hawawini et al. (2003) + R&D and past performance effect 3.79 Model of Hawawini et al. (2003) + Patent and past performance effect 3.54 Model of Hawawini et al. (2003) + R&D, patent and past performance effect 3.23 Effect sizes are presented in percentages. Note that positive estimates indicate an improvement in unexplained performance and negative estimates indicate deterioration compared to the same model in which the book value in the market-to-book ratio is not adjusted for the book value of intangible assets.

When firms’ past performances are not included in the model, adjusted market-to-book ratios worsen the explanation in variance of firm performance. If the adjustment of book values does improve the explanation of firm performance, it only decreases the variance in unexplained firm performance with approximately 3%. For this reason, I’m not convinced that the adjusted market-to-book ratios better explain variance in firm performance. Though, the results of the sensitivity check presented in appendix D, find a more pronounced improvement in the explanation of variance in firm performance when book values are adjusted. Further research is necessary to understand the effect of adjusting the book values for the book value of intangible assets.

(29)

28 6. Concluding Remarks

This study examines whether the addition of an innovation and a past performance effect and the adjustment of firms’ book values for the firms’ book value of intangible assets in market-to-book ratios improves the explanation of firms’ performance. The main question is whether these extensions to the model of Hawawini at al. (2003) decrease the unexplained variance in firm performance.

Unadjusted market-to-book ratios and market-to-book ratios in which the book values are adjusted for the book value of intangible assets are used as the two measures for firm performance. Our dataset consists out of 355 non-financial and non-utility US firms, covering 14 industries from 1990 till 2012. This study focusses on the performance of average firms and excludes firms for which the average market-to-book ratios are more than 3 standard deviations deviated from the industry median. The One-way ANOVA, Kruskal Wallis test and Mood’s median test are used to test whether the independent effects are significant. In line with Hawawini et al. (2003), I use the component variance analysis as the main methodology to measure the sizes of the effects and unexplained variance in firm performance.

Based on the results of the component variance analysis, I conclude that the addition of the past performance effect decreases the unexplained firm performance compared to the model of Hawawini et al. (2003) with 25.41%. The addition of an innovation effect is negligible. The adjustment of firms’ book values for the book value of intangibles slightly improves the unexplained variance, but only when the past performance is included in the model as an explaining factor. The results are not convincing enough to conclude that the adjusted market-to-book ratios better explain the variance in firm performance and that they are the preferred performance measure.

Past performance explains approximately 80% of the variance in firm performance and is the dominant factor in explaining firm performance. The addition of past performance nullifies all other effects. Therefore, past performance should be the main focus for CFOs in capital allocating decisions. The findings of this study imply that a linear regression model, similar to the model used by Schmalensee (1985) and shown in equation 3, might be a more appropriate model to use for capital allocating decisions than the component variance analysis model described in section 4.1:

(30)

29

The strength of this model is that one can easily focus on the importance of an individual effect. An interesting topic for further research is to obtain insight in the size and the sign of the alpha in equation 3. The conclusion that past performance is the dominating effect, possibly supports the theory of Koller et al. (2010) and Nissim and Penman (2001) that over time excess returns and therefore firm performances revert to a mean level. On the other hand, the dominating past performance effect could also imply that a firm could maintain its competitive advantages and a ROIC higher than a certain mean level. Therefore, it would be valuable to explore whether the alpha is more firm or industry specific.

As emphasized in section 5, this research has some limitations which form topics for future research. First, as the dataset only contains large US firms, my findings cannot be generalized on small and medium firms. Results may only be generalized on all large US firms because industries, firms and years are a random selection from their underlying population (Hawawini et. al., 2003). Future research should find out whether the same conclusions apply to small and medium sized firms.

In addition, the proxies for innovation leave room for improvement. Replicating the research and taking the value of patents granted to firm and taking the extent to which a firm adapts and creates innovation as proxies for firms’ innovativeness, would be a good topic for future research to verify if innovation effects are irrelevant compared to other firm and industry effects.

(31)

30 References

Anandarajan, A., C. Chin, H. Chi, and P. Lee, 2008, The effect of innovative activity on firm performance: The experience of Taiwan, Advances in accounting 23, 1-30.

Barney, J., 1991, firm resources and sustained competitive advantage, Journal of Management 17, 99-120.

Bergsma, F., Is performance driven by firm- or industry specific factors? Unpublished Master

Thesis, Faculty of Economics and Business, University of Groningen, 1-41.

Bettis, R. A., and M. A. Hitt, 1995, The new competitive landscape, Strategic Management

Journal 16, 7-19.

Bosworth, D., and M. Rogers, 2001, Market value, R&D and intellectual property: An empirical analysis of large Australian firms, The Economic Record 77, 323-337. Brooks, C., 2008, Introductory econometrics for finance, (Cambridge university press,

Cambridge, UK).

Brown, G.W. and A.M. Mood, 1951, Proceedings of the Second Berkeley Symposium on

Mathematical Statistics and Probability (University of California Press, Berkeley, CA).

Cabral, L. M. B., 2000, Introduction to Industrial Organization (MIT Press, Cambridge, MA). Chakravarti, I. M., R. G. Laha, and J. Roy, 1967, Handbook of Methods of Applied Statistics

(John Wiley and Sons, New York, NY).

Chan, L. K. C., J. Lakonishok, and T. Sougiannis, 2001, The stock market valuation of research and development expenditures, Journal of Finance 56, 2431-2456.

Claver, E., J. Molina, J. and J. Tarí, 2002, Firm and industry effects on firm profitability: A Spanish empirical evidence, European Management Journal 20 (3), 321-328.

(32)

31

Cohen, J., P. Cohen, S. G. West, and L. S. Aiken, 2003, Applied Multiple Regression

Correlation Analysis for the Behavioral Sciences (Lawrence Erlbaum Associate,

Mahwah, NJ).

Conover, W.J. and R.L. Iman, 1981, Rank transformations as a bridge between parametric and nonparametric statistics, The American Statistician 35(3), 124-129.

Cox, D.R. and P.J. Solomon, 2003, Components of Variance (Chapman &Hall, London, UK). Dierickx, I. and K. Cool, 1989, Asset stock accumulation and the sustainability of competitive

advantage, Management Sciences 35(2), 1504-1511

Franko, L. G., 1989, Who's winning, who's losing, and the R&D factor as one reason why,

Strategic Management Journal 10, 449-474.

Gadiesh, O. and J.L. Gilbert, 1998, Profit pools: A fresh look at strategy, Harvard Business

Review 76(3), 139-147.

Galbreath, J., and P. Galvin, 2008, Firm factors, industry structure and performance variation: New empirical evidence to a classic debate, Journal of Business Research 61, 109-117. Gibbons, J.D. and S. Chakraborti, 2003, Non Parametric Statistical Inference (Marcel

Dekker, New York, NY).

Griliches, Z., 1981, Market value, R&D, and patents, Economic Letters 7, 183-187.

Gunday, G., G. Ulusoy, K. Kilic, and L. Alpkan, 2011, Effects of innovation types on firm performance, International Journal of Production Economics 133 (2), 662-676 Hall, B. H., A. Jaffe, and M. Trajtenberg, 2005, Market value and patent citation, Rand

Journal of Economics 36, 16-38.

Hall, B. H., 1993, The stock market's valuation of R&D investment during the 1980's,

American Economic Review 83, 259-264.

(33)

32

Hawawini, G., V. Subramanian, and P. Verdin, 2005, Is performance driven by industry- or firm specific factors? A reply to McNamara, Aime and Vaaler, Strategic Management

Journal 26, 1083-1086.

Hirschey, M. and W. Wichern, 1984, Accounting and market-value measures of profitability: Consistency, determinants, and uses, Journal of Business & Economic Statistics 2 (4), 375-383.

Hoopes, D.G., T.L. Madsen, and G. Walker, 2003, Why is there a resource-based view? Toward a theory of competitive heterogeneity, Strategic Management Journal 24(10), 889-902.

Kim, C.W. and R. Mauborgne, 1997, Value innovation: The strategic logic of high growth,

Harvard Business Review 75, 103-11

Koller, T., M. Goedhart, and D. Wessels, 2010, Measuring and Managing the Value of

Companies (John Wiley & Sons, Inc, Hoboken, New Jersey).

Kruskal, W. H., and W. A. Wallis, 1952, Use of ranks in one-criterion variance analysis,

Journal of the American Statistical Association 47.

Mahoney, J.T. and R.J. Pandian, 1992, The resource-based view within the conversation of strategic management, Strategic Management Journal 13(5), 363-380

McGahan, A. M., and M. E. Porter, 1997, How much does industry matter, really? Strategic

Management Journal 18, 5-14.

Megna, P., and M. Klock, 1993, The impact of intangible capital on Tobin's q in the semiconductor industry, American Economic Association 83, 265-269.

Nelson, R. R., 1991, Why do firms differ, and how does it matter? Strategic Management

Journal 12, 61-74.

(34)

33

Nordstokke, D. W., and B. D. Zumbo, 2010, A new nonparametric Levene test for equal variances, Psicológica 31, 401-430.

Porter, M. E., 1980, Competitive Strategy (Free Press, New York, NY).

Roquebert, J. A., R. L. Philips, and P. A. Westfall, 1996, Market vs. management: What 'drives' profitability? Strategic Management Journal 17, 653-664.

Rumelt, R. P., 1991, How much does industry matter? Strategic Management Journal 12, 167-185.

Schmalensee, R., 1985, Do markets differ much? American Economic Review 75, 341-351. Searle, R. S., G. Casella, and C. E. McCulloch, 1992, Variance Components (John Wiley and

Sons, New York, NY).

Subramanian, A., 1996, Innovativeness: Redefining the concept, Journal of Engineering and

Technology Management 13, 223-243.

Teece, D.J., G. Pisano, and A. Shuen, Dynamic capabilities and strategic management,

Strategic Management Journal 18(7), 509-533.

Tidd, J., 2001, Innovation management in context: environment, organization and performance, International Journal of Management Reviews 3, 169-183.

Toivanen, O., P. Stoneman, and D. Bosworth, 2002, Innovation and the market value of UK firms 1989-1995, Oxford Bulleting of Economics and Statistics 64 (39), 39-61.

Van der Stede, W.A., C.W. Chow and T.W. Lin, 2006, The use and usefulness of nonfinancial performance measures, Behavioral Research in Accounting 18, 185-206

Wernerfelt, B., 1984, A resource-based view of the firm, Strategic Management Journal 5, 171-180.

(35)

34

Appendix A: Data, methods and findings of relevant previous studies

Table A.1: Methods and data used in relevant previous studies

Study Data used Method Used Findings

Schmalensee (1985) US FTC, only manufacturing firms, 1975, 1.775 observations

OLS and CVA Firm effects .62%, industry effects 19.46%, unexplained performance 81% Rumelt (1991) US FTC, only manufacturing firms, 1974-1977, 6.932 observations

ANOVA and CVA Firm effects 46%, industry effects 4%, unexplained performance 45%

McGahan and Porter (1997)

US Compustat business segments reports,1981-1994, 72.742

observations

ANOVA and CVA Firm effects 36%, industry effects 19%, unexplained performance 49%

Hawawini et al. (2003, Full sample, M/B)

Stern Stewart data US firms,1987-1996, 5.620 observations

CVA Firm effects 33%,

industry effects 11%, unexplained performance 52%

Hawawini et al. (2003, Excluding top 2 out- & underperforming firms, TMV/CE)

Stern Stewart data US firms, 1987-1996, 3.420 observations

Bergsma 2012 (2012, Full sample, M/B)

S&P 1200 global index, 1990-2012, 20.242 observations

Bergsma (2012, Excluding top 5 leaders and losers from each industry, M/B)

Thomson Reuters Database 1990-2012, 20.022 observations

CVA Firm effects

12%,industry effects 0%, unexplained performance 88%

M/B, market-to-book ratio at enterprise level; OLS, ordinary least square; CVA, component variance analysis; ANOVA, analysis of variance; FTC; US Federal Trade Commission line of business dataset; S&P 1200, Standard's and Poor's 1200 global index.

(36)

35

Appendix B Descriptive statistics of the dataset used in this study Table B.2: Division of firms into industries

Industry name Number of firms classified to the industry

Automobiles & Parts 31

Basic Resources 14

Chemicals 10

Construction & Materials 6

Food & Beverage 64

Health Care 7

Industrial Goods & Services 23

Media 25

Oil & Gas 46

Personal & Household Goods 39

Retail 15

Technology 13

Telecommunications 8

Travel & Leisure 54

(37)

36

Table B.2: Descriptive Statistics per year of the data sample used in this study

Year Variable Number of

observations Mean

Std.

Deviation Median

1991 _{Market-to-book ratio} ₁₀₄ _3.26 _2.53 _2.56

Adjusted market-to-book ratio 104 3.94 2.87 3.36

R&D to sales 104 3.19 3.92 1.84

Patent ratio 104 0.05 0.14 0.00

Past performance market-to-book ratio 104 2.49 1.69 2.04 Past performance adjusted market-to-book ratio 104 3.00 1.98 2.45

1992 Market-to-book ratio 124 3.01 1.98 2.50

R&D to sales 124 3.50 4.48 1.74

Patent ratio 124 0.05 0.15 0.00

R&D to sales 139 3.31 4.39 1.49

Patent ratio 139 0.04 0.12 0.00

R&D to sales 145 3.46 4.78 1.91

Patent ratio 145 0.04 0.12 0.00

R&D to sales 145 3.37 4.60 1.69

Patent ratio 145 0.04 0.12 0.00

R&D to sales 151 3.53 4.83 1.69

Patent ratio 151 0.05 0.14 0.00

R&D to sales 159 3.55 4.88 1.61

Patent ratio 159 0.05 0.15 0.00

(38)

37

Table B.2: Descriptive Statistics per year of the data sample used in this study (continued)

Year Variable Number of

observations

Mean Std. Deviation

Median

R&D to sales 167 3.60 5.03 1.74

Patent ratio 167 0.05 0.15 0.00

R&D to sales 172 3.36 4.54 1.47

Patent ratio 172 0.05 0.15 0.00

R&D to sales 185 3.43 4.47 1.70

Patent ratio 185 0.05 0.15 0.00

R&D to sales 204 4.62 6.74 1.83

Patent ratio 204 0.05 0.15 0.00

R&D to sales 241 5.35 9.35 1.65

Patent ratio 240 0.05 0.13 0.00

R&D to sales 260 5.16 7.63 1.80

Patent ratio 259 0.05 0.13 0.00

R&D to sales 268 4.58 6.71 1.57

Patent ratio 267 0.04 0.13 0.00