Optimal Levels of Industrial Diversification: Observed for Dutch, Belgian and German listed firms

(1)

Optimal Levels of Industrial Diversification:

Observed for Dutch, Belgian and German listed

firms

MSc IFM Thesis

Author: Arno Nijrolder, 3033775

University of Groningen

Faculty of Economics and Business MSc International Financial Management Supervisor: Adri de Ridder

(2)

2 Table of Content

Abstract ... 3

1. Introduction ... 4

2. Literature review ... 6

2.1 The development of Industrial Diversification over time ... 6

2.1 Benefits and drawbacks of international Diversification... 7

2.3 Measurement of industrial diversification ... 9

2.4 Valuation effects from Industrial Diversification ... 10

2.5 Performance effects from Industrial Diversification ... 12

2.5 Hypothesis development ... 14

3 Methodology ... 17

3.1 Industrial diversification measure ... 17

3.2 Tobin’s q method ... 18

3.3 Fama French 3 factor model (1992) method ... 20

4. Data collection ... 22

4.1 Data collection ... 22

4.2 Descriptive Statistics ... 24

5. Results ... 31

5.1 Regression analysis of firm value on industrial diversification, by the Tobin’s q method ... 31

5.2 Regression analysis on the shape of the relationship of firm value on industrial diversification, by the Tobin’s q method ... 33

5.3 Primary Three-Factor Regression analysis of performance ... Fout! Bladwijzer niet gedefinieerd. 5.3 Primary Three-Factor Regression analysis of performance ... 34

5.4 Additional tests for Three-Factor Regression analysis of performance ... 37

5.5 Secondary Three-Factor Regression analysis of performance ... 39

6. Conclusion ... 43

(3)

3 Abstract

This thesis analyzes the effect of industrial diversification on firm value for samples of firms from Germany, the Netherlands and Belgium, in an independent fashion. A unique

methodology is applied to study how non-diversified firms and firms of different degrees of diversification affect firm value. The first part of the two-fold methodology observes firm market valuation by Tobin’s q in a cross-sectional setting, in the second method the Three-Factor Model of Fama and French (1992) is applied to analyze the unexplained returns from five portfolios ranging from no diversification to high levels of diversification. While the literature provides numerous articles identifying how diversification is on average value destructive, such as Denis et al. (2002), this research is set out to identify where and how diversification can enhance value. Empirical evidence is presented of a positive relation between lower levels of diversification and firm value for German firms. In the Netherlands, non-diversified firms are negatively related to firm performance, realizing an average annual discount of 5.76 percent yearly. Hereby the results hint at the existence of optimal trade-offs between the benefits and costs of diversification, such as synergetic benefits and

(controllable) levels of managerial complexity. Furthermore, for Dutch firms it appears that wide levels of diversification outperforms narrow levels of diversification, while for German firms evidence is presented, significant at the 10 percent level, for a negative relationship between firm performance and lower levels of R&D intensity amongst diversified firms. The value destroying effects from lower levels of R&D for diversified firms can be explained by Tallman and Li (1996), who argue that managers of diversified firms can resort to strategies emphasizing on financial controls while sacrificing strategic investment in innovation.

Field Key Words: Industrial Diversification, Firm value, Firm performance, Portfolio

(4)

4 1. Introduction

“Over the past decade, an extensive academic literature has developed documenting the

causes and consequences of industrial diversification. Studies in this literature report that, on average, diversified firms are valued at a discount relative to a portfolio of comparable, single-segment firms. This value discount appears to stem, in part, from inefficient investment policies. “ – “Global diversification, industrial diversification, and firm value” Denis, D. J.,

Denis, D. K., Yost, K., 2002. The Journal of Finance, 57, 1951-1979

Since 2002 the number of publications studying industrial diversification on industrial diversification effects drastically reduced, while findings on value reducing effects are empirically supported for several major world economies next to the U.S., though not in Germany, in Lins and Servaes (1999).

Meanwhile in 2016, 25 percent of the listed firms in Germany and 33 percent in the

Netherlands and Belgium are segmented at the 2-digit level, based on NACE Rev. 2 industry codes. Such a company is Siemens AG, active in 12 unique industries according to the Bureau van Dijk Orbis database, which includes: ‘Manufacture of computer, electronic and optical products, Manufacture of general — purpose machinery, Civil engineering, Information and communication - Publishing activities, Financial service activities, except insurance and pension funding, Architectural and engineering activities; technical testing and analysis.’ The company is ranked as the world’s 50th_{largest company, with a US$ 109.8 Billion market} capitalization in May 2017 (Forbes, 2017), it is successfully active as conglomerate while it is also identified as the most diversified firm in this study.

In several studies, such as Denis et al. (2002) and Berger and Ofek (1995), the effect of industrial diversification on firm value is concluded to be negative on average. This study takes a closer look at this relationship, by aiming to create understanding of why some diversified firms perform better than others.

The following research question is determined: How does industrial diversification influence

firm valuation?

(5)

5

industrial diversification related stock performance, the level of concentration of diversification and the level of R&D intensity amongst diversified firms.

Particularly, the aim of this study is to identify if the benefits can outweigh the costs of diversification when industrial diversification is present at particular levels. To do so, the methodology is build two-fold. The effect on firm value is measured by a method incorporating Tobin’s q as dependent variable. Firm performance is studied by applying the 3-factor model of Fama and French (1992), after creating five portfolios of firms which are categorized by their level of industrial diversification the sample will be categorized into portfolios ranging from no- to high levels diversification. Herewith both a cross-sectional and a time-series methodology is followed.

A research gap is shown for reviewing the effect of industrial diversification on firm value and performance, in a recent and non-U.S. sample. Studying German firms will be

particularly interesting when comparing the results to findings such as those by Lins and Servaes (1999), who found no discount for segmented firms in Germany. However in other markets such as the U.S., U.K. and Japan a discount was empirically determined.

(6)

6 2. Literature review

2.1 The development of Industrial Diversification over time

Pointed out by Denis et al. (2002) the use of industrial diversification by U.S. firms has decreased since the 1980s. Liebeskind and Opler (1994) identified a relationship between corporate diversification and agency costs by studying privately held firms. They explain their results as follows, the increase of competition by globalization has lead firms to focus on their core lines of business, leading to a decrease in industrial diversification. Jensen (1993) argue that the strengthening of corporate control in the 1980s has lead firms to withdraw from industrial diversification strategies in the following years.

(7)

7 2.2 Benefits and drawbacks of international Diversification

Denis et al. (2002) conducted a research to study excess firm value in relation to international and industrial diversification, leading to three conclusions. First, their results showed that reductions in industrial diversification lead to an increase in value, between 1984 and 1997 in the U.S. Second, global and industrial diversification are correlated on the firm level.

Implying that these two strategic options tend to be used as complements rather than

substitutes. Third, they report that industrial diversification has decreased over 1984 to 1997 within the U.S. Their results show that industrially diversified firms are valued at a discount to comparable stocks.

They theorize that industrial diversification is viewed as a facilitator for increasing agency problems, as it can facilitate managers to pursue motivations such as increased power and prestige, compensation arrangements, personal risk reductions, leading to inefficient cross-subsidization of less profitable business units, according to Denis et al. (2002). Theoretical support of inefficient investing by diversified firms over segments is provided by the models of Rajan et al. (2000) and Scharfstein and Stein (2000), and empirical evidence in Rajan et al. (2000) and Scharfstein (1998). Scharfstein (1998) also notes to increase in managerial

complexity resulting from diversification.

Berger and Ofek (1995) observed the effects of industrial diversification to be a value

discount of 13 to 15 percent in the US market. Lins and Servaes (1999) found the effects to be similar within the UK, smaller in Japan, with a 10 percent value discount and nonexistent in Germany. The authors found that inside ownership positively and size negatively affect value improvement effects of industrial diversification within Germany. Furthermore, he authors state the difficulties of industrial diversification are rooted in the agency costs issue, while power struggles amongst divisions are also mentioned. From Al-Maskati, Bate and Bharbra (2015), additional causes of diversification discounts are pointed. In Campa and Kedia (2002) the potential endogeneity issues are studied, empirically supporting that the decision of diversification determines the discount as for a minority occasions diversification results in a premium, though in a majority of occasions leads to discounts.

(8)

8

There are also benefits documented to the application of diversification, namely tax benefits of debt financing by Lewellen (1971), lowered bankruptcy costs by Amihud and Lev (1981), the creation of an internal capital market void of information asymmetries by Williamson (1975) and Stein (1997) and economies of scope by Teece (1980).

Furthermore, synergy benefits can result from industrial diversification similar to how it can result from international diversification, as explained in the interpretation by Morck and Yeung (1998) of Caves’s (1971) internationalization theory of synergy. They theorize

synergistic benefits are realized when valuable information based assets are acquired within a firm, allowing sharing of valuable intangible assets, for which industrial diversification can be a means by allowing the exploitation of these superior skill into different business segments. Similarly, Myers (1977) pointed to the possibilities for an increase in resource allocation efficiency. This can be viewed as contradictory to the perspective presented by more recent works mentioned. Rajan et al. (2000) describe how industrial diversification could lead to improved capital allocation due to the creation of an internal capital market as well how associated costs are created. Furthermore the firm-level decision of diversification can lead to premiums as well, as shown by In Campa and Kedia’s (2002) study of potential endogeneity issues.

The existing literature appears to agree on that industrial diversification is on average negatively influencing firm value. However, value creating effects are not necessarily undermined by drawbacks as is shown by the observation of Lins and Servaes (1999) in the German market, where no diversification discount was observed and Campa and Kedia (2002), who find diversification can lead to premiums by analyzing the decision to diversify at the firm level.

2.2.1 Industrial Diversification and Innovation

Hitt et al. (1997) show that performance of internationally diversified firms is positively related to high levels of segmentation and negatively to non-segmented firms. Also, they find innovation premiums are enhanced for internationally diversified firms when segmentation is high. From the empirical evidence in Hitt et al. (1997) shows a positive relation of R&D intensity on firm value for internationally diversified firms when enhanced by industrial diversification.

(9)

9

relationship between diversification and innovation is found when management emphasizes on financial control. The application of high levels of financial controls are observed amongst U.S. diversified firms (Hoskisson and Hitt, 1988, Baysinger and Hoskisson,1989). Tallman and Li (1996) identify several reasons for why higher levels of financial controls are related to diversified firms, concerning the management’s incapability of overseeing the strategic needs of many different businesses, leading to risk-averse decisions.

2.3 Measurement of industrial diversification

To weigh the level of industrial diversification of a firm, Robins and Wiersema (2003) discuss measures of industrial diversification. They describe the entropy and concentrix indices, which consider the number of industries active in, their interrelatedness and the weight of an industry segment to the observed firm. Due to the unavailability of data on firm specific segment size, a modified index is formed which treats all accounted industry segments of a firm by equal weight, while measuring the number of industries active in and their

interrelatedness.

(10)

10 2.4 Valuation effects from Industrial Diversification

A measure to capture the effect firm value can be found with Tobin’s q, initially introduced by Kaldor (1966). In Gozzi et al. (2008) the following definition is retrieved, shown to be applied in numerous similar studies:

Tobin’s 𝑞 = 𝑇𝑜𝑡𝑎𝑙 𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 +𝐵𝑜𝑜𝑘 𝑉𝑎𝑙𝑢𝑒 𝑜𝑓 𝑑𝑒𝑏𝑡

𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡 𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 (1)

The ratio is often computed without subtracting both liability book and market value, assuming both are equal. As elaborated by Tobin (1969), the ratio compares a market price by its replacement costs. According to the efficient market hypothesis all information should be included in the price of a stock following which a higher Tobin’s q can be interpreted a signal of growth expectations by the market.

Lang and Stultz (1994) explain that, when applying q for studying effects from industrial diversification, reflects the market’s opinion on the value of diversification, which can be (partially) illusory. Therefore the Efficient Market Hypothesis of Fama (1970) must be followed, assuming that the market value is an unbiased estimated of the present value of cashflow.

Bodnar et al. (1999) measured the valuation effects of industrial diversification by a variation of Berger and Ofek (1995). This measure takes the ratio of a firm’s total capital value (market value of equity plus book value of equity minus book value of common equity) to the sum of imputed values of its industrial segments. The latter is computed by industry median ratios of total capital to sales multiplied by the level of sales for the segment. To compute the industry median ratios, single-segment domestic firms are taken. Our largest number of observations was found in Germany, of 585 highly industrial diversified observations. With over 17 times less observations, my sample size is does not suffice the application of this method.

Other studies have applied Tobins’q as dependent variable when analyzing the value influencing effects of industrial diversification. Lang and Stultz (1994) studied the relation between Tobin’s q and the degree of diversification in a cross-sectional setting. They

(11)

11

Al-Maskati, Bate and Bharbra (2015), studied industrial diversification, measured at a single point in time. They derived the impact of diversification from market-to-book valuations, by combinations of observable and computed data. Compared to Al-Maskati, Bate and Bharbra (2015). For the construction of this thesis no data is available on interfirm segment size, in assets or sales.

Authors Al-Maskati, Bate and Bharbra (2015) included in their study the impact of corporate governance on industrial diversification and the relationship between industrial diversification and firm performance. With variables such as board structure, busyness and independence, effect of institutional and block holding ownerships, shareholder activism, executive payment, CEO ownership, board size, CEO compensation and more, numerous corporate control

strength proxies were included in their models. They found that the quality of corporate governance is negatively related to the level of industrial diversification and the

(12)

12 2.5 Performance effects from Industrial Diversification

Below several models are described which measure the impact of diversification on firm performance. These are the Capital Asset Pricing Model of Sharpe (1964), from here on referred to as CAPM, and the 3-Factor model by Fama and French (1992).

The expected return of an asset or portfolio can be determined from a systematic market risk factor, the beta, according to the Capital Assets Pricing Model (CAPM) of Sharpe (1964), given below.

𝑅𝑗 = 𝛼𝑗+ 𝛽𝑗∗ 𝑅𝑚+ 𝜀𝑗 (2)

Here, 𝑅𝑗 indicates the return of firm j, 𝛽𝑗 the systematic risk of firm j, 𝑅𝑚 illustrates the return of the market and 𝜀_𝑗 the random variable or unexplained variation. In internationalization over segmented markets, the systematic risk is derived from the covariance between the firm and market returns and the change in cashflow diversity, as explained by Shapiro (1978). Commonly the firm and market returns are subtracted by the risk free rate. Equation 2 below illustrates how the 𝛽_𝑗 can be rewritten:

𝛽_𝑗 = (ρ_𝑗𝑚σ_𝑗/σ_𝑚) (3)

Here, 𝛽_𝑗 shows the beta of asset j, while ρ_𝑗𝑚 is the correlation coefficient between j and the market, σ𝑗 shows the standard deviation of j and σ𝑚 represents the market standard deviation. Cashflow diversity is captured by firm standard deviation, showing systematic risk is derived from an assets covariance with the market and by its cashflow diversity.

Building upon the CAPM, researchers have found evidence for various other factors predicting asset returns, such the Arbitrage Pricing Theory (APT) of Ross (1976). By the Efficient Market Hypothesis of Fama (1970) it is argued that all relevant information should be reflected in an assets price in an efficient capital market. However in a later study Fama and French (1992) find three particular variables, being a market, size and book-to-market factor explain returns over 90% in their sample, comparing to 70% with the CAPM.

(13)

13

𝑅 = 𝑅_𝑓+ 𝛽₃(𝑅_𝑚− 𝑅_𝑓) + 𝛽_𝑠∗ 𝑆𝑀𝐵 + 𝛽_𝑣∗ 𝐻𝑀𝐿 + 𝛼 (4)

Here, R is asset return, 𝑅_𝑚 is the return of the market portfolio, 𝑅_𝑓 the risk free rate, three beta’s represent the correlation from each variable, SMB stands for "Small (market capitalization) Minus Big" and HML for "High (book-to-market ratio) Minus Low". The subtractions within SMB and HML are computed from the historic returns of portfolios categorized by size and market valuation. In a study covering the years 1963 up to 1990, they found superior performance by firms with low book-to-market ratios and small market capitalizations. The alpha in this model shows the unexplained difference, which can be interpreted as the added value difference, or premium, when comparing portfolios.

Recently, Fama and French published a new five factor model (2015), still excluding momentum but including a profitability and investment factor. Their findings show higher returns for high operating profitability stocks and lower returns for high total asset growth stocks. The investment factor could be significant to our relationship, as possible value enhancing or reducing effects of diversification could be catalyzed when investment increases. Following the evidence proved by Rajan et al. (2000) and Scharfstein (1998) of inefficient investment over a firms different segments, a continuation of the negative effects of

(14)

14 2.5 Hypothesis development

In order to develop the hypotheses, the identification of value enhancing and reducing sources of industrial diversification is observed. Value enhancing effects being tax benefits of debt financing by Lewellen (1971), lowered bankruptcy costs by Amihud and Lev (1981), the creation of an internal capital market characterized by minimal information asymmetries (Williamson (1975) and Stein (1997) economies of scope by Teece (1980) and increased resource allocation efficiency (Myers, 1977, Rajan et al., 2000). Furthermore benefits from synergy effects are identified from interpreting Morck and Yeung (1998).

Value reducing effects identified include increased managerial complexity, from Scharfstein (1998), increased agency costs from Denis et al. (2002) , and inefficient investing from Scharfstein (1998), Rajan et al. (2000) and Scharfstein and Stein (2000).

An answer to the research question: How does industrial diversification influence firm

valuation? is sought by a two-fold study, analyzing firm value and firm performance

independently by industrial diversification. Both methods are applied for the first two hypotheses. After establishing results on the relation between diversification and firm value for our sample, hypotheses 3 and 4 will be tested by the performance method only.

Similar to the literature survey findings, I expect to find the overall effect of industrial diversification on the total sample to be value reducing, leading to the first hypothesis.

H1a: Firm value is negatively related to industrial diversification

H1b: Firm performance is negatively related to industrial diversification

The expected on average negative effect between industrial diversification and firm value suggests the relationship could be linear. However, it could be so that the benefits of industrial diversification outweigh the costs at certain levels of industrial diversification, resulting in a U-shaped or S-shaped relationship. It is unclear where an optimal trade-off can be found, if it can be identified for a national sample after consideration of control variables. Basing a prediction on previous findings, which illustrated that diversification overall negatively impacts firm value, while not neglecting the theoretical benefits, an optimal level of

(15)

15

H2a: The relationship between firm value and industrial diversification is inversed-u

shaped

H2b: The relationship between firm performance and industrial diversification is

inversed-u shaped

The following hypotheses are not tested for firm value but only for firm performance. As knowledge on the relationship between firm value and diversification within our sample is established, the performance method is viewed as sufficiently capable for examining how the underlying mechanisms affect valuation.

Following my interpretation of Morck and Yeung (1998) explanation of Caves’s (1971) internationalization theory of synergy, hypothesis H3 is developed. As elaborated in chapter 2.2, I assume industrial diversification expectably facilitates the exploitation from synergistic benefits in similar fashion to international diversification, as the potential for utilizing

valuable intangible assets is larger when a firm is present different businesses. Building on this assumption, I expect that the synergistic benefits from diversification will be enhanced when the relatedness of a firms business is higher, as valuable intangible assets, such as superior production or logistic capacities will be easier applicable. Alternatively, the potential synergy benefits could be greater when a firm is widely diversified, as it is exposed in a greater extend to unique or different circumstances. Therefor the concentration of

diversification is determined as independent variable, derived from the relatedness of a firms industries in which it is active.

H3: Firm performance is positively related to ‘narrow’ industrial diversification Based on the findings of Hitt et al. (1997), who show that performance of internationally diversified firms is positively related to innovation and high levels of segmentation, it is unclear if higher levels of R&D may also lead to higher value for segmented firms, given the potential for spillover effects and the management incapability argument of Tallman and Li (1996). The management incapability argument can lead to two mechanisms, assumption A: Segmented firms who highly invest in R&D do so less adequately do to the limited

(16)

16

Based on earlier observations of a negative relationship between diversification and

innovation in the U.S., from Baysinger and Hoskisson (1989) and Hitt et al. (1990), a negative relation can be expected, proving assumption B.

H4 is formulated to test whether or not industrial diversification provides a greater potential for exploiting valuable intangible assets created by innovation, by R&D spillovers.

H4: Firm performance is positively related to R&D intensity, for industrially

diversified firms

For industrial diversification and innovation, they observe in support for a negative relationship.

(17)

17 3 Methodology

After considering the above limitations, the hypotheses are tested in two different models. In the first part of this chapter, explanation is provided on how the dependent variable Tobin’s q can be used to capture the effects of industrial diversification on firm value, in a cross-sectional setting. With the second part of this chapter the performance is observed for firms categorized by their level of segmentation. With short-term performance as dependent variable, in monthly returns from 2012 to 2014, the Fama and French Three-factor model (1992) is applied to capture the effects of industrial diversification levels.

3.1 Industrial diversification measure

Five portfolios are formed after categorizing each firm by an industrial diversification measure, which ranks the extend of industrial diversification by the number of industries a firm is active in, and the relatedness of these industries. The literature study provided two measures to be applied for determining the level of firm industrial diversification, the entropy and concentrix index by Robins and Wiersema (2003). Due to unavailability of data regarding sales per industry segment, a modified index was applied which considerers the number of industry segments identified by four digit industry segment by NACE rev 2:

𝐼𝐷 = (#4𝑑𝑖𝑔 − 1) ∗ 1 + (#3𝑑𝑖𝑔 − 1) ∗ 2 + (#2𝑑𝑖𝑔 − 1) ∗ 3 (5)

In equation 5, ID is the Weighted Industrial Diversification Ratio. #4𝑑𝑖𝑔 is the number of unique 4-digit industries a firm is active in. #3𝑑𝑖𝑔 is the number of unique 3-digit industries a firm is active in and #2𝑑𝑖𝑔 is the number of unique 2-digit industries a firm is active in. The Weighted Industrial Diversification Ratio, from here on ID ratio, is computed by considering the number of industries a firm is active in on the 4 digit, 3 digit and 2 digit level, where different weights are given to depending on what digit level a firm is diversified. Each industry a firm is active in is assumed to carry an equal weight of assets, sales and profit due to unsufficient data. With this equation, industrial diversification is determined by considering the amount of different industries in which a firm is active in, and their relatedness. For example, the following computation is made for Volkswagen AG, based on data retrieved from Orbis. As it is active in the segments 2920, 2910, 2932, 2920, 7711 and 7712, it is diversified at the 4-,3- and 2- digit level by the values 5, 4 and 2, the Weighted Industrial Diversification Ratio equals 13.

(18)

18

the 4 digit level, as the interrelatedness is expectably lowest between different industries different on the 2 digit level, and highest when only different on the 4 digit level.

3.2 Tobin’s q method

To analyze the relationship between firm value and industrial diversification, the

methodology of ‘Tobin’s q, corporate diversification, and firm performance’ by Lang and Stultz (1994) is reviewed, leading to the following regression models.

𝑞 = 𝛼.+ 𝑏2𝑠𝑞𝑟𝑡𝑊𝐼𝐷𝑟𝑎𝑡𝑖𝑜 + 𝑏3𝑅&𝐷 + 𝑏4𝐿𝑜𝑔𝐴 + 𝜀 (6) 𝑞 = 𝛼_.+ 𝑏₂𝐷(𝐼𝐷2) + 𝑏₃𝐷(𝐼𝐷3) + 𝑏₄𝐷(𝐼𝐷4) + 𝑏₅𝐷(𝐼𝐷5) + 𝑏₆𝑅&𝐷 + 𝑏₇𝐿𝑜𝑔𝐴 + 𝜀 (7)

𝑞 = 𝛼.+ 𝑏2𝐷(𝑆) + 𝑏3𝑅&𝐷 + 𝑏4𝑙𝑜𝑔𝐵𝑉 + 𝜀 (8)

The definitions of the symbols used in the above equations are shown on the following page. Similar to Lang and Stultz (1994), the observations are ranked based on their level industrial diversification, with the square root value of the ID ratio. Unlike the ID ratio variable, of which the distribution is positively skewed, the square root value of the ID ratio appears normally distributed. With the square root of the ID ratio, from here on sqrtID ratio, Cross-sectional regression is performed for observations in the year 2016. Equation 6 will be used for regression analysis, previously applied by Lang and Stultz (1994), with added control variables for size and R&D intensity. Equation 7 and 8 build on the same model while using categories formed out of different industrial diversification models, being the portfolios formed from the ID ratio, and a dummy variable for firms segmented at the 2-digit level. In

equations 6, 7 and 8, 𝛼 is a constant of the function, 𝜀 is the error term.

The regressions will be performed on a transformation of q as well, for developing an optimal model, depending on de skewedness of the correlation of industrial diversification on q. These transformation concern the logarithm of q, q squared, the square root of q and the cube root of

q. The square of q, is selected as means for countering positive skewness (or skewness to the

right). The other three transformation aim at countering negative skewness (or skewness to the left), for realizing normal distribution. Dependent variables are also considered for transformation.

(19)

19 Table 3.1: Tobin’s q method equation variables

q: Developed from equation 1.

Tobin’s 𝑞 = 𝑇𝑜𝑡𝑎𝑙 𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 +𝐵𝑜𝑜𝑘 𝑉𝑎𝑙𝑢𝑒 𝑜𝑓 𝑑𝑒𝑏𝑡

𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡 𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 (1)

q²: The squared value of Tobin’s q.

√q: The square root value of Tobin’s q, where for negative q values a multiplication by -1 is added.

³√q: The cube root value of Tobin’s q. Log q: The logarithm of q.

√IDratio: The square root of the Weighted Industrial Diversification score.

ID(j): Portfolio j, derived from ranking the Weighted Industrial Diversification ratios and forming five portfolios. D1: non-diversified, ID2: low diversification (ratios from 1 to 3), ID3 (ratios from 4 to 6), ID4 (ratios from 7 to 11) and ID5 (ratios from 12 up to 51).

S: Portfolio of firms segmented at the 2-digit level.

R&D: R&D intensity, measured by R&D expenses divided by operating revenue. Control variable added to capture effects on q.

(20)

20 3.3 Fama French 3 factor model (1992) method

To test how firm performance affects industrial diversification, Fama and French’s 1996 ‘Multifactor explanations of asset pricing anomalies ‘ is followed. The paper builds on Fama and French’s three factor model (1992), shown in equation 1, from which excess portfolio returns can be determined.

Time series regression is performed by the following the three-factor model of Fama and French (1992) for different portfolios.

(𝑅_𝑗− 𝑅_𝑓) = α_𝑗+ 𝛽_𝑚∗ (𝑅_𝑚− 𝑅_𝑓) + 𝛽_𝑠∗ 𝑆𝑀𝐵 + 𝛽_ℎ∗ 𝐻𝑀𝐿 + ε_𝑗 (9)

Here, (𝑅_𝑗− 𝑅_𝑓) is the returns of portfolio j subtracted by the risk free rate, (𝑅_𝑚− 𝑅_𝑓) is the return of the market portfolio subtracted by the risk free rate. The three beta’s represent the correlation from market development, size and growth. SMB stands for "Small Minus Big", referring to market capitalization. HML stands for "High Minus Low", measured by the book-to-market ratio.

(21)

21

(22)

22 4. Data collection

The following sample criteria were applied: Listed firms only, headquartered in Germany, the Netherlands and Belgium, active in any primary industry, with the exception of those active in the financial industry, identified by the two digit NACE rev. 2 industries 64, 65 and 66 (64: Financial service activities, except insurance and pension funding. 65: Insurance, reinsurance and pension funding, except compulsory social security. 66: Activities auxiliary to financial services and insurance activities.)

4.1 Data collection - Tobin’s q method

Data was retrieved from the Tomson Reuters DataStream software for the variables: Total market value, Book value of debt and Total asset book value, all 2016, single point in time observations. The Orbis database provided data on: the industries in which a firm is active in, by NACE rev. 2 codes, required for determining the industrial diversification variables described in Chapter 3. The R&D intensity variable, measured by the average value of R&D expenses divided by operating revenues over the years 2011-2014. Initially the sample size was 745 for Germany, 145 for the Netherlands and 140 for Belgium, after removing observations for missing values sizes reduced to 522, 94 and 94 respectively.

4.1.1 Data collection - Fama French 3 factor model (1992) method

Data was retrieved from the Tomson Reuters DataStream software for the variables: Total Return Index (TRI), Market Value and Price-to-Book value, all monthly data from 2012 up to 2014. The Total Return Index was used to produce asset returns, determined by the formula:

𝑅_𝑡1 = TRI𝑡1−TRI𝑡0 TRI𝑡0

(10)

Here, 𝑅_𝑡1 is the return of an asset on t=1. TRI_𝑡1 is the Total Return Index on t=1 and TRI_𝑡0 is the Total Return Index on t=0. TRI was selected as it incorporates dividends payouts, which are treated to be reinvested in the firm. Price-to-Book values were inverted to create the 3-Factor model growth value. The Orbis database provided data on: Industries in which a firm is active in, by NACE rev. 2 codes, required for determining the industrial diversification

(23)

23

in the 2012-2014 dataset. In 2011 found for one firm, by screening for large changes in firm total returns.

(24)

24 4.2 Descriptive Statistics

Tobin’s q method

Table 4.1: Descriptive Statistics for Tobin’s q Regression Analysis, Dataset of German, Dutch and Belgian Listed Firms, 2016.

Germany Belgium

Sample Mean Med. Std. D. Min Max Sample Mean Med. Std. D. Min Max

q 522 0.13 0.03 0.81 -0.99 8.95 94 0.35 0.22 1.11 -0.16 10.43 q^2 522 0.59 0.00 5.81 -0.98 80.03 94 1.33 0.05 11.23 -0.03 108.75 log_q 522 -0.86 -0.75 0.65 -3.83 0.95 94 -0.73 -0.66 0.49 -2.82 1.02 cubert_q 522 -4.27 1.77 361.75 -6723.76 2946.67 94 10.20 3.40 69.94 -75.98 665.96 sqrt_q* 522 0.11 0.19 0.55 -1.00 2.99 94 -0.50 -0.47 0.36 -3.23 -0.04 ID1 308 0 1 49 0 1 ID2 77 0 1 16 0 1 ID3 75 0 1 11 0 1 ID4 27 0 1 13 0 1 ID5 35 0 1 5 0 1 DS 133 0 1 31 0 1 √WIDratio 522 0.96 1.33 0.00 7.14 94 1.07 1.31 0.00 4.24 R&D 471 2.78 8.15 0.00 72.33 90 4.20 14.83 -0.42 96.25 Log_RD 471 0.27 0.00 0.42 0.00 1.87 90 0.14 0.00 0.36 -0.24 1.99 Log_A 522 5.13 1.14 1.41 8.57 94 5.49 0.89 3.21 8.09 Netherlands

Sample Mean Med. Std. D. Min Max

q 94 0.17 0.14 0.50 -0.98 2.98 q^2 94 0.18 0.02 1.02 -0.96 8.87 log_q 94 -0.75 -0.69 0.52 -2.60 0.47 cubert_q 94 -3.29 2.43 45.86 -396.65 117.19 sqrt_q 94 -0.50 -0.45 0.28 -1.73 -0.05 ID1 41 0 1 ID2 17 0 1 ID3 17 0 1 ID4 6 0.00 1.00 ID5 13 0 1 DS 31 0 1.00 √WIDratio 94 1.41 1.49 0.00 5.74 R&D 90 2.34 6.71 -9.16 52.76 Log_RD 90 0.23 0.00 0.44 -1.01 1.73 Log_A 94 5.64 1.36 0.85 8.60

Note - ID1 represents Industrial Diversification category 1, as defined in Appendix B. The square root value of Tobin’s q is computed for negative q values by adding a multiplication with -1.

(25)

25

or cube root). q clearly shows a skewed distribution in all countries, with a tail to the right. Log of q and the square root of q, show to be closest resembling a normal distribution, based on the mean and median values are closely in the middle of the range set by the minimum and maximum values. Also, the log of R&D summary statistics show a more normal distribution as the regular R&D intensity value. Therefor it is decided to run preliminary regressions by using ‘q’, ‘Log_q’, ‘sqrt_q’ and ‘log_rd’, leaving out the variables ‘q^2’, ‘cubert_q’ and ‘R&D’ which follow a less preferable distribution for performing regression analysis. 133, 31 and 31 observations of diversified at the 2-digit level firms are found in Germany,

Netherlands and Belgium. 308, 41 and 49 observations were found with no level of

diversification, measured at the 4-digit level. By running preliminary regressions using the four optional dependent variables, the highest explanatory power is found by applying the square root of q (‘sqrt_q’) measure, which is therefore used as Tobin’s q for the regression analysis.

Table 4.2: Correlations Matrix for Tobin’s q Regression Analysis dataset, German Listed Firms, 2016.

Equation 1 sqrt_q √WIDr Log_RD Log A

sqrt_q 1

√WIDr 0.09** 1

Log_RD -0.10** -0.07* 1

Log A 0.11** 0.30* 0.14*** 1

Equation 2 sqrt_q ID2 ID3 ID4 ID5 Log_RD LogA

sqrt_q 1 ID2 0.14** 1 ID3 -0.01 -0.17*** 1 ID4 0.04 -0.10** -0.10* 1 ID5 0.05 -0.11** -0.11** -0.06 1 Log_RD -0.10** -0.08* 0.01 -0.04 -0.06 1 LogA 0.12 -0.04 0.07 0.15*** 0.25*** 0.14*** 1

Equation 3 sqrt_q DS Log_RD Log A

sqrt_q 1

DS 0.03 1

Log_RD -0.10** -0.038794 1

Log A 0.107539 0.26*** 0.14*** 1

Note: ‘***’, ‘**’, ‘*’, indicate observations significant at the 1%, 5% and 10% level.

Table 4.2 shows the correlations between the variables. There is no sign of multicollinearity

(26)

26

listed firms, that diversified firms are on average larger. R&D intensity appears to be negatively related to diversification. In line with the argument that managers of highly diversified firms often prefer financial restrictive corporate policies over funding R&D, as they often are incapable of having sufficient understanding of different industries.

Table 4.3: Correlations Matrix for Tobin’s q Regression Analysis dataset, Dutch Listed Firms, 2016.

sqrt_q 1

√WIDr 0.10 1

Log_RD 0.08 0.03 1.00

Log A 0.26** 0.31*** 0.27*** 1.000

sqrt_q 1 ID2 -0.06 1 ID3 0.15 -0.22** 1 ID4 0.09 -0.12 -0.12 1 ID5 -0.02 -0.18* -0.19* -0.1 1 Log_RD 0.08 -0.15 0.01 0.10 0.01 1 LogA 0.26** -0.23** 0.10 0.24** 0.20** 0.27*** 1

sqrt_q 1

DS 0.11 1

Log_RD 0.08 0.01 1

Log A 0.26** 0.32*** 0.27*** 1

(27)

27 Table 4.4: Correlations Matrix for Tobin’s q Regression Analysis dataset, Belgian Listed Firms, 2016.

sqrt_q 1

√WIDr 0.09 1

Log_RD -0.12 -0.13 1

Log A -0.03 -0.01 0.00 1

sqrt_q 1 ID2 -0.20** 1 ID3 0.03 -0.16 1 ID4 0.08 -0.18* -0.14 1 ID5 0.10 -0.11 -0.09 -0.1 1 Log_RD -0.12 -0.06 -0.01 -0.07 -0.08 1 LogA -0.03 -0.04 0.08 -0.07 -0.01 0.00 1

sqrt_q 1

DS 0.13 1

Log_RD -0.12 -0.10 1

Log A -0.03 -0.01 0.00 1

(28)

28 Fama French 3 factor model (1992) method

Below in Table 4.5: Distribution of ID-Scores, the distribution of industrial diversification scores over the total sample is shown, following which the sample was divided in five portfolio’s ranging from low to high industrial diversification, shown in Table 2: Observations

per ID category.

Table 4.5: Distribution of ID-Scores

ID-scores observations ID-scores observations

0 457 14 3 1 73 15 7 2 6 16 1 3 35 18 6 4 6 19 4 5 2 20 1 6 100 21 2 7 20 25 1 8 2 27 1 9 14 29 1 10 9 33 1 11 5 35 1 12 24 51 1 13 9

Table 4.6: Observations per ID category

Observations per ID category ID cat 1 ID cat 2 ID cat 3 ID cat 4 ID cat 5

ID Scores 0 1-3 4-6 7-11 12-51

Germany 355 78 79 30 43

The Netherlands 44 18 18 6 14

Belgium 58 18 11 14 6

Total 457 114 108 50 62

(29)

29 Table 4.7 Descriptive statistics for Fama French 3 factor model (1992) dataset

- Germany and Belgium. Monthly Return Descriptive Statistics, in percentage points: 1/2012-12/2014, 36 Months

Germany Belgium

Sample Mean Median Std. D. Min Max Sample Mean Median Std. D. Min Max

(30)

30 Table 4.8 Descriptive statistics for Fama French 3 factor model (1992) dataset

- The Netherlands. Monthly Return Descriptive Statistics, in percentage points: 1/2012-12/2014, 36 Months

Sample Mean Median Std. D. Min Max

RmRf 36 1.85 2.01 3.14 -4.46 7.41 SMB 36 -0.40 -0.54 1.36 -3.04 4.78 HML 36 0.12 -0.41 2.86 -7.76 6.73 ID1 36 1.38 1.70 3.48 -6.53 7.95 ID2 36 1.34 1.54 6.33 -18.54 12.78 ID3 36 3.00 3.24 5.30 -11.90 18.55 ID4 36 2.37 2.41 5.25 -8.34 15.19 ID5 36 1.03 0.98 1.59 -1.75 4.51 NS 36 1.48 1.46 3.41 -5.87 7.68 S 36 2.13 1.89 3.26 -4.66 8.63 Low R&D 36 1.46 2.16 2.98 -5.47 6.60 High RD 36 1.09 0.29 2.82 -4.74 9.15 No R&D 36 0.96 1.22 4.53 -6.15 11.75 Narrow 36 2.04 1.99 4.79 -8.43 13.43 Broad 36 1.84 1.66 3.09 -4.58 7.14 IDconD1 36 1.65 2.09 3.86 -6.82 7.96 IDconD2 36 1.44 1.70 5.91 -18.16 11.87 IDconD3 36 2.17 2.16 3.23 -4.36 8.37 manID1 36 2.10 2.67 4.56 -7.34 9.62 manID2 36 3.18 2.67 6.29 -11.89 22.22 manID3 36 2.00 1.74 3.35 -4.66 10.31

(31)

31 5. Results

5.1 Regression analysis of firm value on industrial diversification, by the Tobin’s q method

To test H1a: ‘Firm value is negatively related to industrial diversification’ regressions are performed with equation 6, of which the results are presented in Table 5.1.

Table 5.1: Tobin’s q Regressions Results with Weighted Industrial Diversification Ratio for Germany, the Netherlands and Belgium: 2016

Equation 6: √𝑞 = 𝛼.+ 𝑏2√𝑊𝐼𝐷𝑟𝑎𝑡𝑖𝑜 + 𝑏3𝐿𝑜𝑔𝑅&𝐷 + 𝑏4𝐿𝑜𝑔𝐴 + 𝜀

Germany Netherlands Belgium

√q t-stat √q t-stat √q t-stat

c -0.14 -1.25 -0.79 -6.48*** -0.46 -2.10 √WIDratio 0.02 1.10 0.00 0.22 0.02 0.79 Log_RD -0.15 -2.51** 0.01 0.15 -0.08 -1.07 LogA 0.05 2.35** 0.05 2.25** -0.01 -0.29 R² 0.03 0.07 0.02 Adj.R² 0.02 0.04 -0.01

From the model we can conclude that ’√q’ is explained by ‘Log_RD’ and ‘LogA’, by a value of -2.51 and 2.35 at the 5% significance level for Germany. This means an increase of size, by 1 unit of ‘LogA’, increase the market valuation, on average by a value of 2.35 ’√q’. In the same way, the results show that in the Netherlands market valuation is explained negatively by the constant and positively by size. No significant relationship in the Belgium model is found. The correlation coefficient of ‘√WIDratio’ is insignificant and close to zero in all countries, indicating industrial diversification does not affect market valuation on average. The adjusted R² values, 0.02, 0.04 and -0.02, show the models have, with the exception of Belgium, explanatory power similar to those produced by Lang and Stultz (1994), where adjusted R² ranged from 0.02 to 0.08, the explanatory power of the models is very low (their research was conducted in a different setting: U.S., 1980’s, 668 observations, only firms with Total Assets over US$100 million). Finding the strong correlations for Log of Assets is in line of the expectation, as Tobin's q is composition includes book value of assets. Clearly, the findings do not support H1a as the relationship appears to be neutral.

(32)

32

these perhaps have changed when arriving in 2016, where ‘√q’, ‘√WIDratio’ and ‘LogA’ are measured.

As robustness check, a second model is developed to test H1a by changing the measure of industrial diversification, in equation 8. In Table 5.2 results are shown.

Table 5.2: Tobin’s q Regressions Results with Firms Segmented at the 2-digit level, for Germany, the Netherlands and Belgium: 2016

Equation 8: √𝑞 = 𝛼.+ 𝑏2𝐷(𝑆) + 𝑏3𝐿𝑜𝑔𝑅&𝐷 + 𝑏4𝐿𝑜𝑔𝐴 + 𝜀

Germany Netherlands Belgium

c -0.64 -3.90*** -0.79 -6.41*** -0.46 -2.14** DS -0.06 -0.67 0.02 0.28 0.09 1.22 Log_RD 0.00 0.27 0.01 0.16 -0.08 -1.06 LogA 0.10 3.20*** 0.05 2.21** -0.01 -0.30 R² 0.02 0.07 0.03 Adj. R² 0.01 0.04 0.00

The model shows that ’√q’ is explained by ‘c’ and ‘LogA’, by values of -0.64 and 0.10 for Germany, significant at the 1 percent significance level and in the Netherlands by -0.79 at the 1 percent level and 0.05 at the 5 percent level. In Belgium, relative market valuation is explained only by the constant, by -0.46, at the 5 percent significance level. Similar to the primary model, the industrial diversification measure is not significantly different from zero, though negative in Germany and positive in the Netherlands and Belgium.

(33)

33 5.2 Regression analysis on the shape of the relationship of firm value on industrial

diversification, by the Tobin’s q method

For testing H2a: ‘The relationship between firm value and industrial diversification is

inversed-u shaped’, Equation 2 is applied for regression analysis of q on all of the 4 industrial

diversification portfolios of segmented firms, where ‘ID2’ includes firms with lower levels of diversification and ‘ID5’ those of highest observed levels of diversification. ‘ID1’ is left out of the regression to prevent multicollinearity issues. To test the hypothesis, the shape of the relationship between the degrees of diversification is observed.

Table 5.3: Tobin’s q Regressions Results with Firms Categorized by level of Industrial Diversification, for Germany, the Netherlands and Belgium: 2016

Equation 7: 𝑞 = 𝛼.+ 𝑏2𝐷(2) + 𝑏3𝐷(3) + 𝑏4𝐷(4) + 𝑏5𝐷(5) + 𝑏6𝑅&𝐷 +

𝑏7𝐿𝑜𝑔𝐴 + 𝜀 _Germany _Netherlands _Belgium

c -0.18 -1.58 -0.80 -6.14*** -1.85 -1.85* ID2 0.22 3.21*** 0.02 0.26 -1.80 -1.80* ID3 0.02 0.30 0.10 1.19 0.11 0.11 ID4 0.09 0.78 0.05 0.40 0.47 0.47 ID5 0.07 0.72 -0.03 -0.33 0.74 0.74 Log_RD -0.13 -2.28** 0.01 0.14 -1.19 -1.19 LogA 0.05 2.43** 0.05 2.16** -0.34 -0.34 R² 0.04 0.09 0.06 Adj. R² 0.03 0.03 0.01

The results show in Germany for all degrees of industrial diversification positive values, significant for ‘ID2’ at the 1% level, with 0.22 a correlation showing that lower levels of industrial diversification are associated with increased relative market valuation in Germany. In Belgium the opposite is observed for ‘ID2’ where √q decreases by -1.80 for ‘ID2’,

significant at the 10 percent level. Furthermore it is shown that the German model is

(34)

34

To determine whether H2a is true, the following null hypotheses is tested: ‘Firm value is

negatively related to the industrial diversification levels of low diversification (ID2), medium diversification (ID3) and medium-high diversification (ID4)’.

In Germany lower levels of diversification (ID2) show significant positive results, while medium diversification (ID3) and medium-high diversification (ID4) show positive yet insignificant results. H2a is hereby partially supported for the German market as an optimal level of diversification is shown in between no diversification and higher levels of

diversification. However the null hypothesis cannot be rejected, as no direct evidence is produced to test if the ‘ID2’ measure is significantly positive when comparing it to ‘ID5’. In the Netherlands, when ignoring the t-statistics, it is shown that only firms of the highest levels of diversification have a negative coefficient. This hints at the existence of optimal levels of industrial diversification, possibly in category 3. However no significant findings have been produced, therefore H2a is not supported for the Dutch market.

In Belgium, when ignoring t-statistics, firm value rises by each increase in level of industrial diversification, which does not hint at the existence of an inversed u-shaped relationship. Clearly, H2a cannot be supported for the Belgian market.

5.3 Primary Three-Factor Regression analysis of performance

Regression analysis is performed on equation 9, where for each nation, five industrial diversification portfolios have been formed, ranked by levels of diversification. Panel A of

(35)

35 Table 5.4: Summary Statistics and Three-Factor Regressions for Monthly Excess Returns on 15 Portfolios Formed on Industrial Diversification and country: 1/2012-12/2014, 36 Months

ID: No ID 2 3 4 High _{No ID} 2 3 4 High

Panel A: Summary Statistics

Means (in percentage points) Std. deviations (in percentage points)

GER 1.39 0.68 1.25 1.28 _1.46 2.66 2.34 3.64 4.41 3.61

NET 1.38 1.34 3.00 2.37 _1.03 3.48 6.33 5.30 5.25 1.59

BEL 2.15 0.94 1.98 1.70 _1.19 3.90 2.24 3.80 6.10 2.55

A 1.64 0.99 2.08 1.79 _1.22 3.35 3.64 4.25 5.25 2.58

Panel B: Regressions: (𝑅𝑗− 𝑅𝑓) = α𝑗+ 𝛽𝑚∗ (𝑅𝑚− 𝑅𝑓) + 𝛽𝑠∗ 𝑆𝑀𝐵 + 𝛽ℎ∗ 𝐻𝑀𝐿 + ε𝑗

α (in percentage points) t(α)

GER 0.04 0.44 -0.43 -0.14 _0.15 0.23 1.05 -1.29 -0.37 0.75 NET -0.48 -0.89 0.83 -0.08 _0.27 -1.87* -1.05 1.22 -0.13 1.44 BEL -0.03 0.06 0.64 -0.15 _0.49 -0.30 0.16 0.91 -0.15 1.11 A -0.16 -0.13 0.35 -0.12 _0.30 𝛽𝑚 t(𝛽𝑚) GER 0.89 0.31 1.14 1.18 _{1.05 16.67***} 2.44** 11.52** 10.38*** 17.28*** NET 1.07 1.21 1.40 1.17 _{0.39 12.65*** 4.36*** 6.27*** 5.81*** 6.23***} BEL 1.03 0.58 0.79 1.19 _{0.49 27.90*** 4.66*** 3.26*** 3.44*** 3.23***} A 1.00 0.70 1.11 1.18 _0.64 𝛽𝑠 t(𝛽𝑠) GER -0.18 1.09 -0.13 0.05 _0.04 -0.99 2.56** -0.39 0.13 0.18 NET 0.23 0.35 1.15 -0.74 _-0.10 1.02 0.48 1.94* -1.38 -0.60 BEL -0.02 0.36 0.47 -1.03 _0.22 -0.15 0.95 0.63 -0.98 0.47 A 0.01 0.60 0.49 -0.57 _0.05 𝛽ℎ t(𝛽ℎ) GER -0.37 -0.14 -0.36 0.51 _{0.31 -2.95***} -0.47 -1.574264 1.92* 2.17** NET -0.16 1.07 0.33 -0.05 _0.01 -1.74* 3.53** 1.34 -0.22 0.21 BEL -0.21 0.40 0.29 1.92 _{0.53 -3.65***} 2.12** 0.79 3.61*** 2.25** A -0.24 0.45 0.09 0.79 _0.28 R² s.e. GER 0.91 0.36 0.84 0.85 _0.94 0.01 0.02 0.02 0.02 0.01 NET 0.87 0.57 0.60 0.67 _0.66 0.01 0.04 0.03 0.03 0.01 BEL 0.98 0.44 0.25 0.41 _0.35 0.01 0.02 0.03 0.05 0.02 A 0.92 0.46 0.56 0.64 _0.65

(36)

36

To test H1b: ‘Firm performance is negatively related to industrial diversification’ and, the alpha of the portfolios are compared, for portfolio premiums. Panel B shows that only for the non-diversified portfolio in the Netherlands a significant value is found, at the 10 percent significance level, supporting a negative portfolio premium. Our results indicate that, after controlling for market, size and growth, Dutch non-diversified firms have an significantly lower premiums, by -0.48 percentage points per month or -5.76 per year. Herewith no evidence to accept H1b is found. Results which are closest to this prediction are found in Germany, where non-diversified firms showed a premium of 0.04 percentage points per month, however with an insignificant p-value of 0.82 (t-stat: 0.23). Ignoring the t-statistics, it can be observed that premiums are higher for the High ID portfolio compared to the No ID portfolio, in all three countries.

Comparing the No ID and High ID portfolios, the results show that in Germany the No ID portfolio is less strongly correlated to the market, with a betas of 0.89 versus 1.05. In the Netherlands and Belgium the opposite is found, with betas of 1.07 versus 0.39 and 1.03 versus 0.49, suggesting that in the smaller markets industrial diversification reduces exposure to market risk.

Exposure to size is insignificant for the No ID and High portfolios in all markets.

The growth correlation is significantly negative for all No ID portfolios, and significantly positive for all but the Dutch High portfolio. As Fama and French (1996) found that book-to-market increases excess returns, this indicates that No ID excess returns are explained by lower book-to-market values, or lower growth values. For High ID the opposite is found, their excess returns are explained by higher book-to-market values, although for the Dutch

observations insignificantly.

(37)

37 5.4 Additional tests for Three-Factor Regression analysis of performance

Industrial diversification measure of Diversified vs non-diversified

To check the robustness of the results, three alternative tests are performed. The first

robustness test replaces the Weighted Industrial Diversification (ID) measure by dividing the observation into two categories, non-diversified or diversified (measured at the 2-digit level). An overview of the regression results are provided in Appendix B. The results now show premiums of 0.06, -0.41, -0.04 for non-diversified firms and -0.03, 0.31, 0.35 for diversified firms (in the order: Germany-Netherlands-Belgium), where the values for the Netherlands are significant at the ten percent level (p:0.08 and 0.09). This shows a positive premium for diversified firms and negative premium for non-diversified firms, as diversified firms show, after controlling for market, size and growth, a monthly premium 0.72 (0.31+0.41) percent higher than non-diversified firms, equal to 8.64 percent annually. These findings directly contradict H1b: ‘Firm performance is negatively related to industrial diversification’ for the Netherlands.

CAPM regressions

In the second robustness check, the initial performance test is altered by removing the size and growth factors from the regression, resulting in an equation similar to Sharpe’s Capital Assets Pricing Model:

(𝑅_𝑗− 𝑅_𝑓) = α_𝑗 + 𝛽_𝑗∗ (𝑅_𝑚− 𝑅_𝑓) + ε_𝑗 (12)

An overview of the regression results are provided in Appendix A, a summary is provided in

(38)

38 Table 5.5: CAPM Regressions Premiums for Monthly Excess Returns on 15 Portfolios

Formed on Industrial Diversification and Country: 1/2012-12/2014, 36 Months

Regressions: (𝑅𝑗− 𝑅𝑓) = α𝑗+ 𝛽𝑗∗ (𝑅𝑚− 𝑅𝑓) + ε𝑗

GER 0.31 0.20 -0.17 -0.45 _-0.04 1.85** 0.53 -0.60 -1.37 -0.23 NET -0.51 -0.68 0.90 -0.09 _{0.23 -1.97**} -0.65 1.16 -0.14 1.50

BEL -0.21 0.52 0.11 0.10 _0.10 -1.47 1.23 1.52 0.87 2**

A -0.14 0.01 0.28 -0.15 _0.10

Table 5.5 shows that in Germany, the No ID portfolio produced a premium of 0.31 per month,

significant at the 5 percent level, in opposite of an insignificant premium of -0.04 for the High portfolio. Hereby evidence is found in support of H1b: ‘Firm performance is negatively

related to industrial diversification’, for the German market, as non-diversified firms have an

unexplained 0.31 percent positive return per month. For both the Netherlands and Belgium opposite evidence is found. While both show negative premiums for non-diversified firms, this is only significant for the Dutch firms at the 5 percent level, with monthly premium of -0.51 percent. The High portfolios show positive premiums, only significant for Belgian firms, at the 5 percent level, with a monthly premium of 0.10%. Regarding H1b: ‘Firm performance

is negatively related to industrial diversification’, evidence is found suggesting an opposite

relationship for the Netherlands and Belgium.

Single industry analysis: the Manufacturing Section

A third robustness check is performed by rerunning regressions within a single industry segment, to control for differences amongst the countries caused by the observation

(39)

39 Table 5.6: Summary Statistics and Three-Factor Regressions for Monthly Excess Returns on 15 Manufacturing Portfolios Formed on Industrial Diversification and Country: 1/2012-12/2014, 36 Months

ID: No ID 2 & 3 _{4 & High} No ID 2 & 3 4 & High Regressions: (𝑅𝑗− 𝑅𝑓) = α𝑗+ 𝛽𝑚∗ (𝑅𝑚− 𝑅𝑓) + 𝛽𝑠∗ 𝑆𝑀𝐵 + 𝛽ℎ∗ 𝐻𝑀𝐿 + ε𝑗

GER 0.04 -0.44 0.25 0.23 -1.38 1.04

NET -0.22 0.76 0.25 -0.50 0.99 0.86

BEL 0.02 0.42 -0.15 0.13 0.64 -0.15

A -0.05 0.25 0.12

In Table 5.6 it is shown that within the manufacturing segment, no significant premiums have been found, similar to the initial regression where only for the No ID portfolio in the Dutch market the premium is significantly negative at the 10 percent level. Ignoring the t-statistics, the No ID portfolios have lower premiums than the High ID portfolios in Germany and the Netherlands, in opposite of what was hypothesized, while in Belgium the opposite is found. In summary, the third robustness test provides no evidence to support or reject H1b: Firm

performance is negatively related to industrial diversification. 5.5 Secondary Three-Factor Regression analysis of performance

Shape of the relationship between performance and industrial diversification

To test H2b: ‘The relationship between firm performance and industrial diversification is

inversed-u shaped‘, the five ID portfolio premiums described in chapter 5.3 and 5.4 are

reconsidered. First the results from equation 1 are reviewed, in Table5.7.

Table 5.7: Premiums from Three-Factor Regressions for Monthly Excess Returns on 15 Portfolios Formed on Industrial Diversification and country: 1/2012-12/2014, 36 Months

Regressions: (𝑅𝑗− 𝑅𝑓) = α𝑗+ 𝛽𝑚∗ (𝑅𝑚− 𝑅𝑓) + 𝛽𝑠∗ 𝑆𝑀𝐵 + 𝛽ℎ∗ 𝐻𝑀𝐿 + ε𝑗

GER 0.04 0.44 -0.43 -0.14 _0.15 0.23 1.05 -1.29 -0.37 0.75

NET -0.48 -0.89 0.83 -0.08 _0.27 -1.87* -1.05 1.22 -0.13 1.44

BEL -0.03 0.06 0.64 -0.15 _0.49 -0.30 0.16 0.91 -0.15 1.11

A -0.16 -0.13 0.35 -0.12 _0.30

(40)

40

Due to low t-statistic values, no evidence can be derived from the observed premiums. Ignoring the t-statistics, in Germany the highest premiums are found for the 2-ID portfolio (0.44 and 0.15 percent), suggesting that lower levels of industrial diversification, with a weighted ID score from 1 to 3, realized the highest returns after controlling for market, size and growth effects. In the Netherlands and Belgium, the highest premiums are found at 4-6 weighted ID scores (0.83 and 0.64 percent). The findings therefor do suggest that lower to medium levels of industrial diversification perform best, depending on market, which in line with H2b, however evidence is not found.

Performance and industrial diversification concentration

For the third hypothesis, H3: ‘Firm performance is positively related to ‘narrow’ industrial

diversification’, regressions are run on equation 1 for portfolios consisting only of firms who

are industrially diversified and are ranked by their level of concentration in industrial

diversification. A ID concentration ratio is computed by dividing the number of unique 2 digit industries a firm is active in by the number of unique 4 digit industries. A lower value

indicates a greater concentration of industrial diversification within related industries, a higher value indicates industrial diversification is spread over unrelated industries, where a value of 1 represents a diversified firm that is active in industries which differ at the 2-digit level. The portfolios are formed as follows, ‘Narrow’ consists of highly concentrated diversified firms, measured by a ratio up to 0.5 of ID concentration ratio. ‘Broad’ is composed of diversified firms with a ratio larger than 0.5, representing diversified firms active in less related industries. In Table 5.8 the regressed premiums and t-statistics are shown, in Appendix A a complete overview of the results are provided.

Table 5.8: Premiums from Three-Factor Regressions for Monthly Excess Returns on 6

Portfolios Formed on Industrial Diversification Concentration and country: 1/2012-12/2014, 36 Months

P: Narrow Broad Narrow Broad

Regressions: (𝑅𝑗− 𝑅𝑓) = α𝑗+ 𝛽𝑚∗ (𝑅𝑚− 𝑅𝑓) + 𝛽𝑠∗ 𝑆𝑀𝐵 + 𝛽ℎ∗ 𝐻𝑀𝐿 + ε𝑗

GER 0.32 -0.24 1.36 -1.56

NET -0.20 0.03 -0.38 0.32

BEL 0.23 0.21 0.66 1.03

A 0.12 0.01

(41)

41

In Table 5.8 the premiums are compared for firms who are narrowly diversified (Narrow) to firms broadly diversified (Broad). All observations are insignificant at the 10 percent level. Only in Germany the premiums hint at the existence of the expected relationship, where monthly premiums were computed 0.56 percent higher for narrow diversified firms.

As secondary test for H3, different portfolios were computed to rerun the regression (IDD1: non-diversified, IDD2: diversified at the 4-digit level only, IDD3: diversified at the 2-digit level). The regressions result are presented in Appendix A. No significant premiums are observed for firms in the IDD2 category. A premium of 0.33 percent is observed, significant at the 10 percent level, for Dutch firms which are diversified at the 2-digit level, suggesting diversification into unrelated industries created excess returns over the analyzed timeframe.

Performance and R&D intensity amongst diversified firms

To test for H4: ‘Firm performance is positively related to R&D intensity, for industrially

diversified firms’ the Three-Factor model of Fama and French (1992) is reviewed, for a

different composition of portfolios. From only the diversified firms, observations are ranked based on their average R&D intensity over the years 2010 to 2014, resulting in three

portfolios, ‘Low R&D’, ‘High R&D’ and ‘No R&D’, with respective R&D intensity values of; larger than 0 up to 1.5, values larger than 1.5 and values equal to 0.

In Table 5.9 the premiums computed from regressing equation 9 on the portfolios are shown. More regression details are provided in Appendix A.

Table 5.9: Premiums from Three-Factor Regressions for Monthly Excess Returns on 9 Diversified Portfolios Formed on R&D Intensity and Country: 1/2012-12/2014, 36 Months

P: Low R&D High R&D No R&D _{Low R&D} High R&D No R&D Regressions: (𝑅𝑗− 𝑅𝑓) = α𝑗+ 𝛽𝑚∗ (𝑅𝑚− 𝑅𝑓) + 𝛽𝑠∗ 𝑆𝑀𝐵 + 𝛽ℎ∗ 𝐻𝑀𝐿 + ε𝑗

GER -0.74 0.26 _-0.05 -1.83* 1.12 -0.23

NET -0.21 0.51 _-0.20 -1.07 0.95 -0.26

BEL 0.05 0.28 _0.53 0.47 0.39 0.08

A -0.30 0.35 _0.09

The regression results show, at the 10 percent significance level, that German firms with lower R&D intensity have a negative premium, of -0.74 percent per month. Although

(42)

42 intensity, for industrially diversified firms’ is provided for the German market. Although