Peer Group Selection for Multiple Valuation in a Global Setting

(1)

Peer Group Selection for Multiple Valuation in a Global Setting

A thesis presented by Lars Emil Frandsen (92450),

Master of Science in Finance and Accounting (cand.merc.fir)

&

Kasper Juellund (92576),

Master of Science in Applied Economics and Finance (cand.merc.aef)

Copenhagen Business School Supervisor: Daniel Wekke Proebst

May 15, 2019

Number of pages(Characters): 109 (185,848)

(2)

(3)

Resum´ e

Dette speciale har til form˚al at undersøge, hvordan man opn˚ar den mest præcise værdiansættelse ved brug af multipler. Der vil være fokus p˚a at teste udvælgelsen af sammenlignelige selskaber i et globalt perspektiv. Endvidere vil det blive un- dersøgt, om man ved brug af fundamentale udvægelseskriterier i form af tekniske m˚al for profitabilitet,vækst og risiko kan opn˚a en mere præcis værdiansættelse i forhold til brugen af den traditionelle og ofte anvendte industritilgang, hvor sammenlignelige selskaber udvælges inden for samme industri. M˚aden, hvorp˚a resultater vil blive testet, er ved at m˚ale præcisionsgraden af en estimeret multipel i forhold til den reelle værdi af et selskab. Der vil derfor blive estimeret multipler for alle selskaber hvert ˚ar ved brug af forskellige undvægelsesmetoder, for derefter at kunne sammenligne præcisionsgraden af værdiansættelserne. Datagrundlaget for dette speciale er det globale indeks ’S&P Global 1200 Index’ for ˚arene 2007-2017. Dette er konstitueret af syv forskellige indeks, som naturligt kan opdeles i fem forskellige regioner. Tidligere studier har ikke fundet resultater, der entydigt støtter den ene tilgang frem for den anden. Dog indikeres det, at en kombination af den fundamentale tilgang inden for en industri giver en højere præcisionsgrad af værdiansættelsen.

Resultaterne af tidligere studier er begrænset til at undersøge selskaber fra et eller f˚a lande. Vores resultater indikerer at præcisionsgraden ikke kan forbedres ved at udvide universet af sammenlignelige selskaber til et globalt perspektiv. Tværtimod indikerer vores resultater, at man opn˚ar en højerere præcisionsgrad, n˚ar man be- grænser universet af potentielle sammenlignelige selskaber til regioner, hvilket gør sig gældende uanset, hvilken metode der anvendes til udvælgelsen af sammenlignelige selskaber. Endvidere viser resulterne, at der ikke er noget entydigt resultat for, at hverken industitilgangen eller den fundamentale tilgang er bedst til udvælgelse af sammenlignelige selskaber. Dog indikerer resultaterne, at kombinationen af begge metoder generelt set, p˚a tværs af flere multipler, forbedrer præcisionsgraden.

(4)

1 Introduction

For a society, it is generally important that assets and companies can be priced accurately, as this enables optimal employment of capital. Several methods are available for valuing companies, the most applied method being the market approach, also known as peer group analysis, or multiples-based valuation. Peer group analysis utilises multiples of comparable companies to determine the value of the analysed company. One criterion for being able to value companies through peer groups accurately consists of the ability to find similar companies with respect to the drivers of multiples. The valuation of two similar companies will, with perfect information, trade for the same price. Within peer group selection, three main schools of thought exist. First, the most applied method in practice is the industry peer group selection method, which employs industry classification to identify comparable companies. Second, the fundamental value driver approach identifies peers based solely on the value drivers behind multiples, which can be decomposed into three main factors, namely, profitability, growth, and risk. The third, and the one followed by the most recent study within multiples-based valuation conducted by Lee, Ma, &

Wang (2015), identify peers based on internet searches. See Section 3 for further description of the previous literature.

The extensive use of multiples for valuation purpose makes this a relevant topic for investigation. In a global context, the globalisation trend has incited increased economic interdependency across regions and has further led to accounting standards converging towards common standards (Young & Zeng, 2015). The major part of large cap companies operates internationally, and in so doing, they become exposed to risks beyond the country-specific. Thus, we find it relevant and interesting to test how applying a global universe affects valuation accuracy. Within the field of peer group selection, this thesis examines both the industry method and the fundamental method. Several previous studies have tested which of these two methods is most accurate, but no clear pattern supports one over another. To relate previous studies to a global setting, both methods are thus applied.

(7)

This thesis aims to contribute specific information concerning the impact of ex- panding the data sample to a global setting. To this end, we implicitly test how peers should be selected to most accurately value companies. This further implies performing several combinations and tests in order to obtain an overview of what should be considered to obtain the most accurate company valuation.

The thesis is structured by our research question, which is supported by five hypotheses as well as some delimitations that, altogether, set the framework for the rest of the thesis. Section 2provides a theoretical framework to introduce multiples as a valuation approach and explain the relationship between discounted cash flow valuation approaches and multiples. Section 3 presents a literature review to develop an overview and build a solid foundation for this thesis by presenting previous literature in the field of identifying comparable companies. Section 4describes and discusses the data and methodology applied in this thesis. Once the basis for the thesis has been fully established, Section 5 details our empirical findings and the robustness checks to test the strength of our findings. Section 6continues by explaining our findings and relating them to previous literature. This section also details the limitations of our findings, suggests subjects for future research, and comments on the implications for practitioners in applying our findings. Section 7 concludes this thesis.

(8)

1.1 Research Question

Previous literature on multiples-based valuations found ambiguous results when comparing peer groups identified by the traditional industry approach and the fundamental value driver approach. However, several studies indicate that selection methods that combine industry affiliation and fundamental value drivers increase accuracy. Those studies primarily use domestic samples for their research. Thus, the first part of this study investigates if the same results apply in a global setting using the S&P Global 1200 Index as a sample. Furthermore, it investigates if the estimation of market value can be more precise by applying the fundamentals approach within an industry. Lastly, it tests whether an intra-regional setting has an impact on the valuation accuracy. The following main question sets up the general outline for this thesis.

How can peers most accurately be selected to estimate the market value of listed companies using a multiples-based approach?

To set up an efficient research design and be able to interpret the findings of this thesis, the following sub-question regarding previous literature is answered before testing our hypotheses.

How are previous studies on peer group selection for multiple-based valuations de- signed, and what did they find?

Five hypotheses are tested in order to structure the empirical analysis and answer the main question. Hypothesis 1 and Hypothesis 2 aim to test whether existing findings apply in a global setting.

Hypothesis 1: The selection of peer groups based on fundamental value drivers reduces the median valuation error compared to traditional industry peer groups.

Hypothesis 2: The selection of peers within industries based on fundamental value drivers reduces the median valuation error compared to the traditional industry peer

(9)

groups and peer groups based on fundamental value drivers.

Serra & F´avero (2018) found differences in multiples for comparable firms across regions, which Hypothesis 1 and Hypothesis 2 do not take into consideration. There- fore, it is relevant to test the selection methods in a regional setting. Furthermore, it makes sense as it relates our findings from Hypothesis 1 and Hypothesis 2 to previous studies concerning industry-based and fundamental-based peer group selection methods. However, this thesis adds additional insights regarding the selection methods’ applicability in several regions rather than just a single national environment.

Formally, this is tested by answering Hypothesis 3 and Hypothesis 4.

Hypothesis 3: The selection of peers within regions based on the industry approach reduces the median valuation error compared to a cross-regional setting.

Hypothesis 4: The selection of peers within regions based on fundamental value drivers (i.e. profitability, growth, and risk in combination) reduces the median valuation error compared to a cross-regional setting.

Lastly, it is relevant to combine all of the selection methods to clarify whether this further improves valuation accuracy. This might be the case due to differences in country-specific risk and the previous findings indicating that industry combined with fundamentals increases accuracy. To test this combination, Hypothesis 5 has been formulated.

Hypothesis 5: Combining industry, region, and fundamental value drivers (i.e.

profitability, growth, and risk in combination) can further reduce the median valuation error.

(10)

Figure 1: Overview of Hypothesis Structure

Note: The arrows indicates which selection methods that each hypothesis intend to test the difference in valuation errors between.

1.2 Delimitations

This section gives an overview of the delimitations made in relation to answering the research question and the subsequent hypotheses.

The thesis is limited in scope and considers only the multiples-based valuation method. Moreover, it analyses the multiples EV/Sales, EV/EBITDA, EV/EBIT, P/E, and P/B which is similar to the multiples applied in Knudsen, Kold, & Plen- borg (2017). Therefore, this thesis does not intend to test which multiple leads to the most accurate valuation. A robustness test is performed by applying forward multiples. The selection variables for the fundamentals-based method is based on the theoretical decomposition of multiples and previous literature. The dataset is limited to the S&P 1200 Global Index during the period 2007–2017. This index is based solely on large cap companies; therefore the thesis does not test the differences in valuation errors across small cap, mid cap, and large cap.

No effort is made to test which peer group size is optimal. The peer group size for

(11)

the fundamentals-based method is limited to 10 peers and at least five peers for industry. The argumentation behind these choices is elaborated on in Section 4.2 and4.4. Moreover, a robustness check is performed to illustrate the implications of this delimitation. Furthermore, the predicted multiples are calculated by applying the harmonic mean. This delimitation means we do not test which method of averaging procedure is most precise.

The normalisation of financial statements is limited to the ‘Bloomberg Adjusted Figures’ which normalises all abnormal expenses/non-recurring items in the financial statements, such as legal settlements, restructuring costs, divestment of assets, and so on. This delimitation means that no further adjustments are made to nor- malise the financial statements. Liu, Nissim, & Thomas (2002b), Alford (1992), and Bhojraj & Lee (2002) support the argument that normalised earnings provide more accurate valuations compared to earnings including extraordinary earnings.

Moreover, no adjustments are made to compensate for differences in accounting policies. Lie (2002) and Young & Zeng (2015) found that there is empirical evidence that accounting differences affect the accuracy of multiples and thereby bias the valuation if no action is made to align this. However, this has not been implemented due to the comprehensiveness of doing so on a large dataset, such as the one considered in this thesis.

In general, this thesis is limited to only answering the main question, the sub- question, and the following hypotheses presented in the research question.

(12)

2 Theoretical Framework

The purpose of this section is to introduce multiples as a valuation approach and explain the relationship between discounted cash flow valuation approaches and multiples-based approaches. This relationship shows under which conditions it would be justified to assume identical multiples among a set of companies (i.e. a peer group). The derived properties will provide the foundation for further discussions on which fundamental value drivers would be optimal to use when identifying such peer groups.

First, the theoretical properties will be derived mathematically for both equity- based and enterprise value-based multiples. Second, a review of empirical evidence on drivers of multiples will be provided. The purpose is to set up some criteria that ensure the best starting point for our empirical findings.

2.1 Theoretical Drivers of Multiples

In this section, it will be shown that under certain assumptions multiples are positively related to profitability and growth, while they are negatively related to risk.

First, properties of equity-based multiples will be derived. Second, properties of enterprise value-based multiples will be derived. Third, the direction of the relationships between multiples and profitability, growth and risk will be explained.

For the sake of simplicity, it is assumed that dividends and cash flows are growing at a constant rate in perpetuity. This assumption provides a simple constant setup in which the factors can be interpreted as the expected constant factor in the long run.

Some theoretical insights are overlooked in a constant growth model; therefore, several researchers have derived expressions that do not assume constant growth (e.g. Fairfield, 1994; Penman, 1996). However, the simple constant setup should be sufficient for the purpose of showing a basic intuition in the relationship between the variables and the multiples.

(13)

2.1.1 Drivers of Equity-based Multiples

The equity-based multiples are derived from the dividend discount model (Gordon, 1962). Assuming a constant dividend growth rate in perpetuity,gD, the model can be expressed as

Pt= Dt+1

r_e−g_D, (1)

where P is the market value of equity, D is the amount of dividends, r_e is the required return on equity. Replacing dividends with net earnings (E) multiplied by payout ratio (PayOut) yields

Pt= Et+1×P ayOut

r_e−g_D . (2)

Next, replacing net earnings (E) with return on equity (ROE) multiplied by book value of equity (B) yields

P = ROE×B×P ayOut

r_e−g_D . (3)

The retention rate (RR), defined as the share of net earnings that is reinvested in the company, is equal to ROE multiplied by gD. Thus, replacing the payout ratio with (1 – RR) and dividing the equation by B yields the following expression for the P/B multiple:

P

B = ROE×(1−RR) re−gD

= ROE−gD

re−gD

. (4)

Finally, multiplying the denominator on both sides in Equation 4 by ROE yields this expression for the P/E multiple:

(14)

P

E = ROE−g_D re−gD

× 1

ROE. (5)

From Equations 4 and 5, it is evident that profitability (ROE), growth (gD) and risk (r_e) drives the equity-based multiples in theory.

2.1.2 Drivers of Enterprise Value-based Multiples

The enterprise value-based multiples are derived from the discounted cash flow model. Assuming a constant growth rate of free cash to firm in perpetuity, gFCFF, the model can be expressed as

EV_t= F CF F_t+1

r_{W ACC} −g_{F CF F}, (6)

whereEV is the enterprise value, FCFF is the free cash flow to firm,r_WACC is the weighted average cost of capital.

Given that the reinvestment rate, RIR, equals (Change in net working capital + Capital expenditures - Depreciation & amortisation)/NOPAT, by replacing FCFF with net operating profit after tax (NOPAT) multiplied by (1 –RIR), the following expression is obtained:

EV = N OP AT ×(1−RIR) rW ACC−gF CF F

. (7)

Since the reinvestment rate (RIR) is equal to return on invested capital (ROIC) multiplied by g_FCFF, replacing NOPAT with ROIC multiplied by invested capital (IC) and dividing both sides of the equation by invested capital yields the EV/IC multiple:

(15)

EV

IC = ROIC×(1−RIR) rW ACC −gF CF F

= ROIC−g_{F CF F} rW ACC−gF CF F

. (8)

NOPAT is equal to earnings before interest and tax (EBIT) times (1 - corporate tax rate (TR)). Thus, multiplying the denominator on both sides in Equation 8 by ROIC yields the following expression for the EV/EBIT multiple:

EV

EBIT = ROIC−g_{F CF F} rW ACC−gF CF F

× 1

ROIC ×(1−T R). (9) Given that the depreciation & amortisation rate (DR) is defined as depreciation &

amortisation/EBITDA, EBIT can be expressed as earnings before interest, tax, depreciation and amortisation (EBITDA) times (1 - DR). Replacing EBIT in Equation 9 with this expression yields the following expression for the EV/EBITDA multiple:

EV

EBIT DA = ROIC−gF CF F

r_{W ACC} −g_{F CF F} × 1

ROIC ×(1−T R)×(1−DR). (10) Since EBIT-margin is defined as EBIT/Sales, EBIT can be expressed as Sales multiplied by EBIT-margin. Replacing EBIT in Equation 9 with this expression, and multiply both sides by EBIT, yields the following expression for the EV/Sales multiple:

EV

Sales = ROIC−g_{F CF F}

r_{W ACC}−g_{F CF F} × 1

ROIC ×(1−T R)×EBIT-margin. (11) As can be seen from Equations 9, 10 and 11 profitability (ROIC), growth (g_FCFF) and risk (rWACC) also drives the enterprise value-based multiples in theory.

(16)

2.1.3 Relationships between Multiples and Profitability, Growth and Risk

Having derived expressions for both equity-based and EV-based valuation multiples, it can be observed that both groups of multiples depends on profitability (in terms of ROE and ROIC), growth (in terms of g_Dand g_FCFF) and risk (in terms of r_e and r_WACC) in the same way. Let p denote profitability,g denote growth andr denote risk.

In this generalised setting, multiples based on balance sheet-based value drivers (i.e.

P/B and EV/IC) relate to the drivers in the following way:

M ultiple_BS = p−g

r−g, (12)

while income statement-based value drivers (i.e. P/E, EV/EBIT, EV/EBITDA and EV/Sales) relate to the drivers in the following way:

M ultipleIS = p−g r−g ×1

p. (13)

Now assume the following restrictions of the relations to avoid breaking the logical positive sign of the valuation multiples: 1) Both the balance sheet-based value drivers (book value of equity or invested capital) and the income statement-based value drivers (Net earnings, EBIT, EBITDA or Sales) need to be positive. 2) Growth cannot exceed either profitability (i.e. ROE or ROIC) or the required rate of return (i.e. re and r_WACC).

From the restrictions and the expressions of the multiples, the relationship between the multiples and profitability, growth and risk can be shown formally by using derivatives. First, the partial derivatives of Equation 12 and 13 with respect to profitability (p) are taken:

(17)

∂M ultiple_BS

∂p = 1

r−g >0 (14)

∂M ultiple_IS

∂p = 1

r−g × g

p² >0 (15)

Note that the restrictions imply that p is positive, since earnings and balance sheet items are restricted to be positive. Furthermore, if one assume that the retention rate and reinvestment rate is positive (which is reasonable to assume in the long run), then growth cannot be negative. Under these conditions, both partial derivatives are positive. Thus, multiples should be increasing functions of profitability.

Next, the partial derivatives of Equation 12 and 13 with respect to growth (g) are:

∂M ultiple_BS

∂g = p−r

(r−g)² >0 (16)

∂M ultipleIS

∂g = p−r (r−g)² ×1

p >0 (17)

Now, it is clear that the partial derivatives with respect to growth can only be positive, if profitability (p) exceeds the required rate of return (r). In most cases this holds true, since reinvestments in a company would otherwise not have been made. Therefore, it is fair to say that growth is generally positively related to valuation multiples. However, it is important to note that in cases where p<rthe relationship is negative.

Next, the partial derivatives of Equation 12 and 13 w.r.t risk (r) are:

∂M ultiple_BS

∂r = g−p

(r−g)² <0 (18)

(18)

∂M ultiple_IS

∂g = g−p (r−g)² ×1

p <0 (19)

Due to the restriction,g<p, these partial derivatives will always be negative. Thus, the relationship between the valuation multiples and risk is negative.

2.2 Empirical Evidence on Drivers of Multiples

Having explained the theoretical relationship between valuation multiples and profitability, growth and risk, the purpose of this section is to shortly review empirical evidence in the literature for the theoretical properties to hold. Most studies in this subject focus on equity-based multiples; however, due to the similarity in dependence on profitability, growth, and risk one could argue that evidence on equity-based multiples also applies to EV-based multiples.

In terms of profitability, Fairfield (1994) and Penman (1996) found that P/B correlated with return on equity (both using a sample of US firms). Penman (1996) also tested, but found no significant correlation between P/E and return on equity.

Hansen, Mouritsen, & Plenborg (2003) confirmed these findings using a sample of Danish firms. Penman (1996) argued this was because return on equity was strongly serially correlated and predicted future profitability on which the P/B was based.

However, because a given level of P/E can be associated with alternative combinations of current and expected future return on equity, current return on equity is not (unconditionally) a good indicator of P/E.

Several empirical studies have also examined the relationship between trading multiples and growth. In the literature, growth has been measured both as realised growth and future growth (both as implied expected growth from analysts’ earnings estimates and ex-post realised growth). Beaver & Morse (1978) (based on a sample of US firms), Fairfield (1994) (also based on a sample of US firms) and Hansen et al. (2003) (based on a sample of Danish firms) found a negative relation between

(19)

P/E and trailing realised one-year growth in earnings. At first sight these findings are surprising in relation to the theory; however, the theoretical properties are as known based on growth in the future rather than realised growth. Thus, these findings might prove that historical growth does not reflect future growth.

Beaver & Morse (1978) and Fairfield (1994) also studied the relation between P/E and ex post earnings growth. Beaver & Morse (1978) found a positive relationship between the two that becomes insignificant after two years. Fairfield (1994) found that firms with high P/E multiples tend to experience higher earnings growth in the following years. However, after five years the difference in earnings growth between initial high P/E and low P/E groups was insignificant. Harris & Marston (1994) measure growth as the mean of financial analysts’ forecasts of five-year earnings growth. They find that growth has a more significant relation to P/B than risk (measured by beta).

In the empirical studies on drivers of multiples, beta has been applied as a measure of risk. Beaver & Morse (1978) found an insignificant relation between P/E multiples and beat, while Harris & Marston (1994) found that, when growth is controlled for, beta has a significant negative relation to P/B multiples.

(20)

3 Literature Review

To develop an overview and build a solid foundation for this thesis, this section aims to present previous literature from the field of identifying comparable firms for valuation purposes, which describes the branch of literature to which this thesis contributes. First, an overview of research designs is provided. This methodological review forms the basis for considerations about the research design of this thesis.

Next, a review of empirical findings in this field is presented.

In a broader context, this subject also relates to studies on other implementation issues when applying multiples for valuation purposes. For a review of the broader literature Plenborg & Pimentel (2016) provided an overview of discussions concerning 1) Choice of comparable firms (which this section also presents in detail), 2) accrual vs. cash flow-based multiples, 3) reported vs. forward multiples, 4) mea- surement of averages, 5) accounting differences, 6) normalisation of earnings, 7) impact of firm size, and 8) illiquidity discount and control premium.

3.1 Review of Research Designs

In terms of valuation accuracy measures, the studies have typically either 1) calculated and compared valuation errors between actual and predicted multiples or 2) used the predicted multiples as the independent variable in a regression with the actual multiples as dependent variables.

In the setup using valuation errors, Dittmann & Maug (2008) demonstrated that the choice of error measure is critical. This also represents a general issue in research on implementation issues when applying multiples. Dittmann & Maug (2008) considered the following error measures:

Absolute percentage error =| Vˆ −V

V | (20)

(21)

Logarithmic error = ln Vˆ

V (21)

where ˆV is the predicted valuation and V is the actual valuation observed in the market. Mathematically, percentage errors penalise overvaluations more than undervaluations. While undervaluations exceeding 100% are impossible, overvaluations are not limited, and will, in several cases, be more extreme than 100%. In contrast, logarithmic errors create more symmetric distributions of valuation errors because, for each overvaluation, there exists an undervaluation of equal absolute size. Statistically, logarithmic error distributions are closer to satisfying the nor- mality assumptions often made for statistical inference.

Out of the 14 studies on horse races of multiples and discounted cash flow valuation methods considered by Dittmann & Maug (2008) in their paper, nine utilised percentage errors, while another five employed logarithmic errors. When applying a valuation error approach, valuation accuracy is ultimately evaluated by the central tendency (i.e. mean and median) and dispersion (i.e. variance, standard deviation, and interquartile range) of valuation errors.

In the regression setup, the predicted multiples for each method under consideration is regressed on the actual trading multiples. Accuracy is measured by the model’s ability to explain cross-sectional variance in the actual multiples.

Within the field of study on peer group selection, percentage errors are most widely used (Alford, 1992; Cheng & McNamara, 2000; Dittmann & Weiner, 2005; Nel et al., 2014; Knudsen et al., 2017). No studies have applied logarithmic errors, while Bho- jraj & Lee (2002), Bhojraj et al. (2003) and Lee et al. (2015) applied the regression setup.

Peer group selection methods based on fundamentals generally follow a ranking approach to identify the closest peers in terms of one or more fundamentals. For fundamentals as single factors, Alford (1992) and Dittmann & Weiner (2005) defined

(22)

the closest 2% of firms in terms of the applied selection variable as peers. Other studies have utilised a fixed number of peers that are closest in terms of the given fundamental. Bhojraj & Lee (2002), for instance, used four to six peers, while Cheng

& McNamara (2000) and Knudsen et al. (2017) employed a default number of six peers Nel et al. (2014), meanwhile, used firms with a given fundamental within a spread of +/- 30% from the firm under consideration as peers.

When two fundamental factors are combined, Alford (1992) and Dittmann & Weiner (2005) took the 14% closest firms in terms of the first fundamental, and then the 14% closest firms in that sub-sample in terms of the second fundamental.¹ In order to be able to include an unlimited number of fundamental factors in the selection of peers, Knudsen et al. (2017) introduced the sum of absolute rank differences (SARD) approach and applied this for combinations of up to five different fundamental factors.

Once peer groups have been identified, two approaches have been employed in this branch of literature when averaging the multiples of firms in the peer group. Alford (1992) and Cheng & McNamara (2000) utilised the median, Bhojraj & Lee (2002), Dittmann & Weiner (2005), Nel et al. (2014), and Knudsen et al. (2017) employed the harmonic mean.

Another important part of the research design that might affect findings is sample selection. Most studies use a sample of US firms. Alford (1992), Cheng & McNamara (2000), and Bhojraj & Lee (2002) employed a wide sample of US firms in several indices, while Lee et al. (2015) and Knudsen et al. (2017) applied S&P Composite 1500. Additionally, Dittmann & Weiner (2005) used both US and European firms.

Finally, Nel et al. (2014) investigated a sample of South African firms as the only contribution to the literature on emerging markets.

Equity-based multiples have been widely employed in the literature on peer group

1Note that 14% times 14% is approximately 2% as used by both studies for single factor fundamentals.

(23)

selection. P/E was used by Alford (1992), Cheng & McNamara (2000), Lee et al.

(2015), and Knudsen et al. (2017), while P/B was also used by Cheng & McNamara (2000), Lee et al. (2015) and Knudsen et al. (2017) as well as Bhojraj & Lee (2002).

However, enterprise value-based multiples have also been employed. EV/EBIT was employed by Dittmann & Weiner (2005) and Knudsen et al. (2017), while EV/Sales was employed by Bhojraj & Lee (2002), Lee et al. (2015) and Knudsen et al. (2017).

Nel et al. (2014) employed a total of 16 different multiples.

3.2 Empirical Evidence on Peer Group Selection

The multiples-based valuation literature has emphasised the importance of being able to identify peers that are truly comparable to the target company. Truly comparable companies must match the expected future cash flows and the riskiness of those cash flows. Dissimilar companies in the peer group will produce biased valuations. Despite this, the number of studies within this branch of valuation literature remains fairly limited. Compared to other valuation approaches, such as discounted cash flow methods, the multiples approach is justified due to its less time-consuming nature. Because of this, finding the right peer group becomes a trade-off between finding the most comparable firms and the effort needed to do so. As a result, several researchers have conducted studies concerning how to most efficiently identify comparable companies. Generally, three different schools of thought on peer group selection have evolved (Plenborg & Pimentel, 2016). The first group of researchers argues that peer group selection should be based on industry classification (Alford, 1992; Cheng & McNamara, 2000). The second school of researchers claims that industry selection should be based solely on fundamental value drivers, which involves similar characteristics in terms of profitability, growth, and risk (Bhojraj & Lee, 2002; Dittmann & Weiner, 2005; Knudsen et al., 2017). The third, and fairly new, school of thought is based on the work of Lee et al. (2015), who argued that peer group selection should be based on web-search traffic patterns.

(24)

3.2.1 Evidence Supporting Industry-based Peer Group Selection

In one of the first contributions to the literature on peer group selection, Alford (1992) found that industry affiliation constituted the most efficient individual selection variable for peer group selection. The industry selection, which was based on four-digit SIC codes, was compared to the fundamental selection variables ROE and total assets. Furthermore, all of the variables were tested combined in pairs. The combination of industry and ROE was found to be statistically insignificant from industry alone, while all other combinations performed significantly worse. His sample consisted of firms from NYSE, ASE, and OTC, and was tested in a cross-sectional analysis based on P/E in the years 1978, 1982, and 1986.

Cheng & McNamara (2000) also evaluated the valuation accuracy of comparable companies selected based on industry membership, total assets, and ROE, as well as combinations of industry membership with total assets and with ROE. However, they evaluated on the basis of the P/E, the P/B, and a combined P/E-P/B benchmark. Cheng & McNamara (2000) confirmed that using industry constitutes an effective method for identifying peers. However, they also found that a combination of industry and ROE yielded even better results. This was true when applying both P/E and P/B multiples. Under the combined P/E-P/B benchmark, industry membership was found to yield the lowest valuation errors across all selection methods.

The sample applied by Cheng & McNamara (2000) consisted of all US firms for which data was available in the 1992 Industrial Compustat database. The sample was evaluated for the period of 1973 to 1992.

3.2.2 Evidence Supporting Fundamentals-based Peer Group Selection

Bhojraj & Lee (2002) developed a systematic approach for selecting comparable firms based on fundamental selection variables. They considered proxies for profitability (Profit-margin, R&D-to-Sales, Return-on-net-operating-assets (RNOA) and

(25)

ROE), growth (expected growth in earnings), and risk (book leverage). However, Bhojraj & Lee (2002) also implemented an alternative procedure when identifying peers based on fundamentals. Specifically, they regressed all of their fundamental selection variables on the valuation multiples in their study (i.e. EV/Sales and P/B) to obtain a fitted model with coefficients for each fundamental selection variable.

Based on this fitted model, they calculated a ‘warranted multiple’ to identify similar companies. They concluded that applying this fundamentals-based approach out- performs other frequently used methods, such as selections according to industry and a combination of industry and size. Bhojraj & Lee (2002) utilised a large sample of US companies from NYSE, AMEX, and NASDAQ for the 1982–1998 period.

Dittmann & Weiner’s (2005) work extended the literature’s geographical scope by utilising a sample of European and US firms (16 countries in total). Within this sample, they examined the accuracy of industry classification, total assets, and return-on-assets (ROA), both individually and in pairs. Their selection approach was similar to Alford (1992) when selecting on the basis of both single factors and factor combinations. However, unlike Alford (1992), Dittmann & Weiner (2005) evaluated the selection methods on the basis of the EV/EBIT multiple and found that ROA offers substantial improvements over industry as a peer group selection variable. Although, specifically for US- and UK-based firms, the combination of total assets and ROA constituted the most accurate selection parameter. In summary, the findings of Dittmann & Weiner (2005) supported the use of fundamental selection variables. Dittmann and Weiner’s (2005) findings relating to cross-border peer group selection are further summarised inSection 3.2.3.

By developing a systematic and flexible method (the SARD approach) for combining several fundamental selection variables, Knudsen et al. (2017) tested the combination of 1) ROE, 2) Debt/EBIT, 3) size, 4) implied EBIT growth, and 5) EBIT-margin against GICS6 industry classification. In their analysis, the fundamental selection variables were added one by one in the order as mentioned above.

This analysis was carried out based on a sample that included constituents in the

(26)

S&P Composite 1500 Index² for each year in the 1995–2014 period. Using this approach for EV/Sales, EV/EBIT, P/E, and P/B multiples, Knudsen et al. (2017) found that the combination of fundamental selection variables significantly increased the valuation accuracy relative to the industry-based approach. Furthermore, they tested the combination of fundamental selection variables within GICS6 industries and likewise found significantly better results compared to the solely industry-based approach. However, they did not test for significance in differences between the fundamentals-based method and the combination of fundamentals and industry.

3.2.3 Evidence on Cross-Border Peer Group Selection

Only a couple of studies have examined the external validity of literature-proposed selection methods outside of the US, namely, Bhojraj et al. (2003); Dittmann &

Weiner (2005); and Serra & F´avero (2018). Bhojraj et al. (2003) argued that the traditional choice of equity analysts to focus on comparing firms from the same country is challenged by economic development. In their introduction, they stated the following:

Problems associated with international accounting diversity and country- specific risk factors are often perceived as insurmountable obstacles that prohibit meaningful comparison of firms from different countries. Pow- erful economic forces now compel us to rethink this problem. With increased global competition, many large- to mid-sized corporations now operate in multiple countries. Even domestic firms find their competi- tors are increasingly likely to be foreign. In this environment, corporate managers frequently need to evaluate their firm’s performance in relation to that of a foreign competitor. At the same time, firms are increasingly cross-listing in foreign exchanges and investors are venturing beyond do-

2The S&P Composite 1500 Index consists of 500 large cap companies (S&P 500), 400 mid cap companies (S&P MidCap 400), and 600 small cap companies (S&P SmallCap 600).

(27)

mestic borders in search of attractive opportunities. As global markets continue to integrate, the demand for analytical tools that facilitate comparison of firms from different countries has also increased.

To test this, Bhojraj et al. (2003) employed the ’warranted multiple’ approach also used by Bhojraj & Lee (2002), which is explained in Section 3.2.2. Likewise, the fundamental selection variables included profitability (Profit-margin, R&D-to- Sales, Return-on-net-operating assets (RNOA), and ROE), growth (expected growth in earnings), and risk (book leverage). To test the relation to industry-specific and country-specific factors, the ‘warranted multiple’ approach was also performed inter- nally in the same two-digit SIC industry and the same country. This approach was applied to a sample of companies from G7 countries³ during the years 1990 to 2000.

The applied multiples consisted of EV/Sales, P/B, P/E, and one-year forward P/E.

In contrast with the paper’s motivation, they found that the fundamentals-based approach was significantly improved when industry- and country-specific factors were incorporated.

As mentioned inSection 3.2.2, Dittmann & Weiner (2005) also examined the optimal geographical context for the peer group selection of the European and American companies in their sample. For most European companies, they found that comparable companies from the European Union member states yielded the best forecasts.

In contrast, for the UK and the US, they determined that comparable companies should be chosen from the same country only. Dittmann & Weiner (2005) further argued that one of the explanations for this result could be that the organisation of financial markets in the UK and the US differs from that of continental Europe.

Additionally, for the specific cases of Danish and Greek companies, it was found to be most optimal to choose peers domestically.

Finally, in a recent study, Serra & F´avero (2018) examined the variability in multiples when peer groups are selected between Brazilian and American companies.

3G7 countries include Canada, France, Germany, Italy, Japan, the United Kingdom (UK), and the United States (US).

(28)

Using multilevel models (hierarchical linear model and additive crossed random- effects model), they concluded that significant variability existed within firms from the same industry, as well as significant variability between Brazil and the United States. The same tests were applied to firms grouped based on fundamentals. Based on these tests using fundamentals (ROE, Market Cap, Historical growth, and Beta), Serra & F´avero (2018) concluded that part of the variability shifted from the variability within firms from the same industry, and the variability between countries to the variability between clusters of companies with similar fundamentals, which limits the negative effect of selecting peers across borders. For the study, Serra & F´avero (2018) applied a sample of Brazilian companies listed on Bovespa and American companies listed on NYSE for the four years from 2011 to 2014. The analysis was conducted on the basis of eight different valuation multiples: EV/Sales, EV/Gross Profit, EV/EBITDA, EV/EBIT, EV/Assets, P/E, P/FCFE, and P/B.

In summary, the literature on peer group selection across borders has generally found that peer groups yield more accurate valuations when peers are selected domestically, or at least regionally. However, Serra & F´avero (2018) found that between Brazil and the US, the negative effect of cross-border peer groups remained limited for fundamentals-based peer groups relative to industry-based peer groups.

3.2.4 Evidence on Other Aspects of Peer Group Selection

This section summarises the empirical findings of studies that did not perform comparable examinations of industry-based and fundamentals-based peer group selection methods, but still, offer interesting insights concerning the discussion of peer group selection. Eberhart (2001) investigated the relationship between the amount of information provided by a company’s comparables (measured by industry membership) and stock volatility. Nel et al. (2014) studied different combinations of fundamental selection variables using a sample of South African companies. Finally, Lee et al. (2015) tested peer group selection based on frequency in co-searched web

(29)

traffic.

First, Eberhart (2001) utilised a sample of US-based companies listed on NYSE and NASDAQ for the period of 1986 to 1995 in order to investigate the relationship between the amount of information provided by a company’s comparables (measured by industry membership). This was accomplished by means of a simple regression model. Using alternative measures of information provided by comparable companies, Eberhart (2001) consistently found a negative and significant relation between each information measure and stock volatility.

Nel et al. (2014) examined peer company selection based on profitability (measured by ROE), risk (measured by total assets), growth (measured by historical revenue growth), and pairwise combinations of these factors. Thus, the considered selection methods were based exclusively on fundamentals, and so did not consider the relative performance to industry-based peer group selection. Using a sample of South African companies, Nel et al. (2014) provided the only contribution to the literature on fundamentals-based peer group selection from an emerging markets perspective.

The study focused on the 2001–2010 period. Evaluating on the basis of as many as 16 different valuation multiples, they determined that ROE constituted the best performing selection variable of the three when the variables are used as single factors. On average, across multiples, they found that a combination of ROE and historical growth in revenue led to the most accurate valuations. In general, it was found that combinations of selection variables increased valuation accuracy compared to using variables independently. Thus, Nel et al. (2014) suggested that each variable included information that was not captured by the other variables, which naturally lends towards a test of the combination of all of them. However, due to the limited depth of the South African sample, they remained unable to perform this test.

Despite the findings and the general practice of using industry, fundamentals, or a combination, Lee et al. (2015) developed an original idea concerning peer group

(30)

selection. Specifically, Lee et al. (2015) argued that new techniques should be developed, since the current service and knowledge-based economy challenges the notion of what constitutes an industry, and furthermore affects the accounting variables employed to measure fundamentals. As a response to this challenge, they developed an approach based on the web traffic passing through the EDGAR search engine.⁴ They assumed that the EDGAR users were co-searching for firms that they perceived as similar. A target company’s peer group is defined as the 10 most co-searched companies. However, when the predicted multiple is found based on the identified search-based peer group, two procedures are considered. One takes the simple average of the multiples of the 10 peers, while the other is traffic-weighted.

Lee et al. (2015) utilised companies in the S&P 1500 Composite Index for each year between 2008 and 2011 as their sample. Based on P/B and EV/Sales, the two search-based predicted multiples outperformed peer groups based on GICS6 industries. For P/E, the result was insignificant. However, Lee et al. (2015) further noted that a lack of data availability makes the approach unsuitable for practitioners.

4EDGAR, the Electronic Data Gathering, Analysis, and Retrieval system, performs automated collection, validation, indexing, acceptance, and forwarding of submissions by companies and others who are required by law to file forms with the U.S. Securities and Exchange Commission (SEC).

(31)

4 Data and Methodology

This section describes and discusses the data sample and the methodology applied in this thesis. To begin, a motivation behind the choice of the data sample is presented. This is followed by a description of how the industry peers are identified, and furthermore, how peers are identified based on the fundamental value approach.

An overview of selection methods follows. Moreover, an examination of how to company valuation estimation is performed. Subsequently, a description of how to evaluate valuation errors follows. Lastly, an illustration of the research design is presented (see Figure 3).

4.1 Selection of Data Sample

For each year, our sample consists of the firms comprising the S&P Global 1200 In- dex. This consists of the S&P 500 Index (USA), the S&P/TSX 60 Index (Canada), the S&P Latin America 40 Index (Brazil, Chile, Colombia, Mexico, Peru), the S&P/TOPIX 150 Index (Japan), the S&P Asia 50 Index (Hong Kong, Korea, Sin- gapore, Taiwan), the S&P/ASX 50 Index (Australia), and the S&P Europe 350 Index.

The motivation behind utilising a global index can be found in the ambition to extend the field of study to a global setting. Earlier studies within this field have not employed global indexes but have instead remained limited to either US equities (Alford, 1992; Cheng & McNamara, 2000; Bhojraj & Lee, 2002; Lee et al., 2015; Knudsen et al., 2017) or South African equities (Nel et al., 2014). Dittmann

& Weiner (2005) applied a European sample but used US companies as a refer- ence point, while Bhojraj et al. (2003) applied a sample of companies from the G7 countries.

Concerning this, the chosen time period is set to 11 years. It is essential to be aware of the implications of using a limited number of years, such as obtaining

(32)

too few observations to generate robust statistical results and cyclical fluctuations.

By including the years 2007–2017, the data sample includes periods with financial distress as well as bullish markets and thus can be used to provide robust findings across varying economic situations. When applying an 11-year period, the gross number of observations reaches 13,001, which must be seen as representative for statistical purposes.

The sample data was obtained from Bloomberg’s database. Valuations were carried out each year on March 31^st. This date was chosen strategically because the financial year of the majority of firms in the sample follows the calendar year. Thus, the like- lihood that the annual report was published is heightened, while the next quarterly report cannot have been published. Knudsen et al. (2017) applied the same logic for their data sample. One observation for each of the constituents included in the S&P Global 1200 Index at March 31^steach year is included.⁵ Financial market data was recorded this date at closing.⁶ To reflect the market participants’ knowledge, the latest four quarterly reports were used for income statement figure, and the last quarterly report was used for balance sheet figures.⁷ If quarterly reports were not available, the two most recent semi-annual reports were used. Finally, the median of equity analysts’ consensus estimates for income statement figures for the upcoming two annual reports was recorded at March 31^steach year from Bloomberg’s estimate database. All data points used in Bloomberg to obtain the sample are presented in Appendix A.1.3.

An issue of using a data sample that is limited to 1,200 companies could include the identification of perfect peers, as some might have better peers outside the sample. Furthermore, it could be argued that using a more extensive data sample would improve the conditions for peer selection, and thereby limit valuation biases.

5Firms that are included with multiple share classes has been limited to one observation per year.

6Market cap observed directly, while enterprise value is defined as Market Cap + Prefered Equity & Minority interests + Net Debt (Net Debt will be defined inSection 4.4.1).

7Bloomberg Adjusted figures, which are adjusted for abnormal earnings-related items, is applied.

(33)

However, using the S&P Global 1200 Index includes approximately 70% of the global market capitalisation.

Another issue of applying a global index is the fact that accounting policies differ across different regions, which can bias valuation accuracy. However, this issue is not solely cross-regional, as accounting differences can appear within regions as well.

Earlier studies have found that adjusting for accounting differences provides better results, and if ignored, it can distort the valuations’ accuracy (Lie, 2002; Young &

Zeng, 2015). No adjustments for accounting differences has been performed in this thesis. This could potentially bias our findings, not just across regions, but on a domestic level as well. Unfortunately, it is difficult to perform a robustness check to capture this, and therefore, this constitutes a weakness in the data quality in the sample applied in this thesis.

The total gross number of observations is 13,001, but this is limited further based on two main reasons. First, if a company possesses negative financial values, it is not meaningful to calculate multiples, and therefore, these observations are excluded. As this thesis operates with several multiples, the total number of excluded observations is determined in relation to the specific multiple. Second, companies within the financial industry lack an enterprise value which makes it impossible to obtain results for EV-based multiples. This explains a significant drop going from equity-based multiples (i.e. P/E and P/B) to enterprise value-based multiples (i.e. EV/Sales, EV/EBITDA, and EV/EBIT).

4.2 Peer Group Identification based on Industry

This section provides an overview and discussion concerning how peer groups are identified within the industry peer group method. The purpose is to establish certain criteria to ensure the best starting point for our empirical findings.

To classify industry peers, the Bloomberg Industry Classification System (BICS) is

(34)

applied. This is composed of five levels where level one offers the broadest industry classification and level five provides the most specific industry classification. Level one possesses a two-digit code, and every time a level is added, two digits are added to the parent code. Level one divides a company into a sector, level two into an industry group, level three into an industry, level four into a sub-industry, and lastly, level five into a segment.

In some cases dividing the data sample into different industries results in a limited number of potential peers. This can cause each peer to exert a considerably significant impact on the predicted multiple, and thereby distort the valuation accuracy.

To avoid such biases, a minimum requirement of five companies for a single peer group is implemented. Earlier studies have demonstrated that employing a minimum of five peers minimises the issue of each peer’s multiple exerting too large an impact for the peer group (Liu, Nissim, & Thomas, 2002b). Alford (1992) suggested a procedure of ‘jumping’ to a higher industry level when the required number of peers is not met. However, to avoid biases from differences in the industry details, this thesis does not implement this procedure. Thus, observations with less than five industry peers are excluded.

There is no specific size of each peer group, as the only criterion is a minimum of five peers. As an example, for the P/E multiple, 9,777 observations are split into 89 different industry peer groups, with the largest peer group possessing 73 peers.

Furthermore, 75% of peers possess a peer group size between 5 and 25 peers. This example offers quite an accurate picture for other multiples as well, with the only difference being that the number of observations for EV-based multiples does not include companies within the financial industry because of their lack of enterprise values.

Table 1 presents the total number of gross observations on different BICS levels, as well as the average net observations when excluding companies within financial industries, negative financial value driver observations, and lastly, removing obser-

(35)

vations with less than five peers.

Table 1: BICS Level Observations

Level Name Peers left Peers left, avg.

BICS2 Sector 13,001 10,793

BICS4 Industry Group 13,001 10,515

BICS6 Industry 13,001 8,917

BICS8 Sub-industry 8,459 2,898

BICS10 Segment 3,110 341

As Table 1 indicates, a trade-off exists between choosing the most detailed industry classification and retaining as many observations as possible. To keep the most detailed level of peers and retain enough observations, this thesis applies level three (‘BICS6’ going forward) for industry classifications. This means that all peer groups for the industry method identify peers on a BICS6 level in this thesis. BICS6 includes 153 different industries, which is seen to be a detailed classification. This is supported by the fact that BICS6 is also the most accurate BICS level for three out of five multiples. For the last two multiples the improvements in accuracy are limited (see Appendix A.1.4).

To ensure the process of implementing rules as mentioned above for the industry peer group selection method, an algorithm was utilised to identify peer groups. First, the algorithm removes all observations with negative earnings. Second, it divides the full data sample into 11 independent data samples by year (2007–2017). Following this, it groups companies into a peer group based on BICS6. Lastly, it excludes all observations that do not possess at least five peers. This limits the gross data sample to an average of 68.6% across all multiples.

(36)

4.3 Peer Group Identification based on Regions

This section offers an examination of how the data sample is distributed across regions. The purpose is to provide the reader with an understanding of how the data sample has been grouped into regions, which is relevant information for investigating Hypotheses 3, 4, and 5.

The data sample contains companies from 39 different countries, which is divided into the seven different indexes that form the S&P Global 1200 Index. The indexes provide a natural split for regions, but some indexes are merged in regards to the region split. The regions consist of 1) Asia-Pacific ex. Japan, 2) Japan, 3) North America, 4) Latin America, and 5) Europe. In asset management, Japan is often separated from Asia-Pacific when grouping regions into indexes (e.g. MSCI AC Asia Pacific ex Japan Index), and thus it possesses its own regional group. The S&P 500 Index and S&P/TSX 60 Index are merged because these represent the US and Canadian indexes, which are both part of North America. Furthermore, the S&P Asia 50 Index and S&P/ASX 50 Index are merged as Asia-Pacific. The S&P/TOPIX 150 Index only contains Japanese equities and is classified as the Japan region. The S&P Latin America 40 Index is classified as Latin America. Lastly, the S&P Europe 350 Index is classified as Europe. For the full list of countries for each region, please see Appendix A.1.1.

Clear differences can be observed in the number of companies per region, with North America being the largest with 560 companies, followed by Europe in second with 350 companies, Japan in third with 150 companies, Asia-Pacific in fourth with 100 companies, and lastly, Latin America in fifth with 40 companies. This sets a natural limitation on the potential number of peers within some regions when analysing Hypotheses 3, 4, and 5. This is discussed further in Section 6.2.

(37)

4.4 Peer Group Identification based on Fundamentals

When identifying peers based on fundamental value drivers, this thesis adopts the SARD approach developed by Knudsen et al. (2017). The SARD approach is similar to the clustering algorithm known as the ‘Manhattan distance’. Under this approach, each company is ranked on a set of fundamental selection variables relative to all other companies in the sample. A target company’s peer group is then identified as the companies possessing the smallest sum of absolute rank differences across the applied selection variables. One advantage of the SARD approach is that it is not restricted by the number of variables that can be applied to identify peers or the number of observations available. Another advantage is that the SARD approach is flexible; it can be used in combination with other approaches, including the industry classification approach. Moreover, under the SARD approach, the selection variables may be tailored to fit the needs of any desired multiple, resulting in a more accurate valuation estimate. Finally, the SARD approach is both intuitive and easy to apply (Knudsen et al., 2017).

To identify companies that are similar in terms of the applied variables, consider a matrix of sum rank differences between each company:

SARDi,j =wX× |rX,i−r_X,j |+wY× |rY,i−rY,j |+ ... +wZ× |rZ,i−rZ,j |, (22)

where SARD is the sum of absolute rank differences between firmi and firmj,r_X,i is the rank of firm i in terms of variable X, rX,j is the rank of firm j in terms of variableX, and so on. Furthermore,w_X is the weight placed on variableX, and so on. The weights must sum to 100%.

Thus, once this matrix has been constructed, the peer group is defined as the n firms with the lowest SARD value, wheren describes the desired number of peers.

(38)

To illustrate the idea of the SARD approach, assume that the sample only consists of the eight firms in Table 2 , with two variables used for peer identification, namely, ROIC and expected growth in EBIT.

Table 2: Fundamentals and Ranks of Companies used in Example

Company ROIC E[4EBIT] r_ROIC r_E[4EBIT]

Sanofi SA 7.3% 5.8% 2 5

Komatsu Ltd 7.7% 40.4% 3 8

Burberry Group plc 38.9% -5.6% 8 2

Carlsberg A/S 8.2% 5.9% 4 6

Walt Disney Company 12.5% 0.2% 6 3

General Motors Company 6.2% -7.6% 1 1

Henkel AG 13.2% 6.2% 7 7

Kroton Educacional SA 8.4% 0.4% 5 4

First, all firms are ranked on the basis of each variable. Second, a matrix of absolute rank difference for combinations of companies for each selection variable is constructed. Third, the matrices of absolute rank differences are summed to form the matrix of SARD values, as illustrated in Table 3.

Table 3: SARD Value and Ranks for Example

Walt General Kroton

Rank direction =↓ Sanofi Komatsu Burberry Carlsberg Disney Motors Henkel Educacional

Sanofi 4 (2) 9 (6) 3 (1) 6 (5) 5 (1) 7 (6) 4 (3)

Komatsu 4 (2) 11 (7) 3 (1) 8 (7) 9 (6) 5 (2) 6 (6)

Burberry 9 (7) 11 (7) 8 (6) 3 (2) 8 (4) 6 (5) 5 (4)

Carlsberg 3 (1) 3 (1) 8 (4) 5 (3) 8 (4) 4 (1) 3 (2)

Walt Disney 6 (5) 8 (5) 3 (1) 5 (5) 7 (2) 5 (2) 2 (1)

General Motors 5 (4) 9 (6) 8 (4) 8 (6) 7 (6) 12 (7) 7 (7)

Henkel 7 (6) 5 (3) 6 (3) 4 (4) 5 (3) 12 (7) 5 (4)

Kroton Educacional 4 (2) 6 (4) 5 (2) 3 (1) 2 (1) 7 (2) 5 (2) For example, the SARD value between Sanofi and Komatsu is|2−3|+|5−8|= 4.

Finally, the desired number of peers is chosen based on the lowest SARD value; the ranks of the SARD values are displayed vertically in Table 3. For instance, if the desired number of peers is four, then the Komatsu peer group would be 1) Carlsberg,

(39)

2) Sanofi, 3) Henkel, and 4) Kroton Educacional. These peers represent the most similar firms within this limited universe in terms of ROIC and expected EBIT growth.

Based on the findings of Cooper & Cordeiro (2008), who concluded that adding more than 10 peers to the peer group does not significantly increase accuracy, this thesis applies 10 as the default number of peers for strictly fundamentals-based peer group selection methods.

4.4.1 Choice of Fundamental Selection Variables

As proven theoretically inSection 2, the main drivers of multiples are profitability, growth, and risk. Other studies have also employed these measures for peer group selection for valuations (Cheng & McNamara, 2000; Bhojraj & Lee, 2002; Nel et al., 2014; Knudsen et al., 2017). The definition of each selection variable is displayed in Table 4. Table 5 provides an overview of selection variables applied for each multiple.

Table 4: Definition of Fundamental Selection Variables Variable Definition

Profitability

ROIC NOPAT⁸ / Average Invested Capital⁹ ROE Net Income / Average Book Value of Equity EBIT-margin EBIT / Sales

Growth

E[4EBIT] (E[EBITt+2] / E[EBITt+1]) - 1

E[4E] (E[Net Incomet+2] / E[Net Incomet+1]) - 1 Risk

Net Debt/EBIT Net Debt¹⁰ / EBIT

Net Debt/Equity Book Value of Equity / EBIT

8NOPAT is calculated as EBIT×(1 - Corporate tax rate in country incorporated)

9Invested Capital is defined as Book Value of Equity + Net Debt.

10Net Debt is calculated as Short and Long Term Debt - Cash and Marketable Securities + Net Deferred Tax Liabilities + Accrued Taxes + Allowance For Doubtful Accounts + Pension Liabilities + Net Derivatives & Hedging Liabilities.

(40)

In Section 2, ROE was proven to be directly linked to equity-based multiples, while ROIC was proven to be directly linked to enterprise value-based multiples.

Thus, this thesis applies ROE as a proxy for profitability when employing P/E and P/B and applies ROIC as a proxy for profitability when employing EV/Sales, EV/EBITDA, and EV/EBIT. Several previous studies have applied ROE within peer group selection methods for equity-based multiples (Alford, 1992; Cheng &

McNamara, 2000; Bhojraj & Lee, 2002; Nel et al., 2014; Knudsen et al., 2017). Fur- thermore, Knudsen et al. (2017) also applied ROE for EV-based multiples. Bhojraj

& Lee (2002), meanwhile, applied ROIC for EV/Sales. Finally, this thesis includes the EBIT-margin as a selection variable for EV/Sales, since it was proven to be an important determinant of this multiple (See Section 2.1.2).

In terms of growth, Beaver & Morse (1978), Fairfield (1994), and Hansen et al.

(2003) found a negative relation between historical growth and multiples. Thus, this thesis implements measures for expected future growth based on equity analysts’

consensus estimates in Bloomberg’s estimate database. To avoid any impact from earning normalisation, the expected future growth is defined by the change between the first and second forecast year. For equity-based multiples, the expected growth proxy is measured on the basis of net income, while the expected growth proxy for EV-based multiples is measured on the basis of EBIT. In previous studies, Bhojraj

& Lee (2002) applied expected growth in net income, while Knudsen et al. (2017) applied expected growth in EBIT.

As proxies for risk, this thesis applies Net Debt/EBIT to measure a company’s payback capacity and applies Net Debt/Book Value of Equity to measure financial leverage. Net Debt/EBIT has been widely used in credit analysis (Petersen, Plen- borg, & Kinserdal, 2017) and was applied for peer group selection by Knudsen et al. (2017), while Debt/Book Value of Equity was applied by Bhojraj & Lee (2002).

(41)

Table 5: Overview of Selection Variables for each Multiple

Profitability (Weight: 33%) Growth: (Weight: 33%) Risk: (Weight: 33%)

Multiple ROIC ROE EBIT-margin E[4EBIT] E[4E] Net Debt/EBIT Net Debt/Equity

EV/Sales X X X X X

EV/EBITDA X X X X

EV/EBIT X X X X

P/E X X X X

P/B X X X X

4.5 Overview of Selection Methods

With each of the individual selection methods explained, the purpose of this section is to provide a short overview of the different peer group selection methods, including combinations of the methods explained above. The sample is divided into sub-samples of each year, from 2007 to 2017. Within each of those yearly sub-samples, peers are identified using the selection methods explained be- low. Next, valuations are developed based on the following valuation multiples:

EV/Sales, EV/EBITDA, EV/EBIT, P/E, and P/B. The one-year forward multiples for EV/Sales, EV/EBITDA, EV/EBIT, and P/E are applied as a robustness check. Please note that the selection variables for the fundamentals-based approach vary across multiples, as explained inSection 4.4.

Each of the selection methods is defined as follows:

Industry refers to an algorithm that selects all companies from the same industry according to the BICS6 code. If this results in less than five peers, the observation is excluded.

Profitability denotes an algorithm that selects the 10 companies with the lowest SARD value using the profitability selection variable(s) defined for the applied mul-

Peer Group Selection for Multiple Valuation in a Global Setting