1
Country Effects on European
Enterprise Valuation Multiples
Master thesis
MSc Business Administration - Finance
University of Groningen
Faculty of Economics and Business
Groningen, August 2011
Author:
Bernd A. Kohlleppel
Student number:
1396455
Thesis Supervisor:
dr. ing. N. Brunia
Second Supervisor:
dr. A.G. Schertler
2
Country Effects on European
Enterprise Valuation Multiples
Bernd A. Kohlleppel
1Abstract
This study investigates valuation differences between companies in the 27 countries of the European Union, in the period 2001-2010. Regression analyses on enterprise and equity valuation multiples find, that valuation differences between multiples of 10 to 14 countries cannot be explained by fundamental factors. Valuation errors of international and domestic peer groups are examined for the enterprise multiples. The peer groups are selected with two methods. The systematic method selects peer groups based on fundamental variables and the industry method selects companies based on industry
classification. When the industry method is used, significantly lower valuation error and standard deviation is observed for the international peer groups. Also, significantly lower valuation error is
observed for the industry method compared to the systematic method, for most of the years and measures. It can be concluded that there is a significant advantage of using an international peer group selected with the industry method. Furthermore, lower valuation error and standard deviation is observed, when increasing the number of comparable companies in the peer group from five to ten.
Keywords: Valuation, multiples, country differences, enterprise value, equity value
JEL classifications: F30, G12
1
1.
Introduction
Practitioners in the field of corporate valuation typically use domestic peer groups to value companies (Bhojraj and Lee [2002]). This practice is motivated by, for example, a lack of sufficient domestic comparable companies and differences in accounting practices, the legal system and the culture between countries. Due to the implementation of an international accounting standard and the increase in political and economic integration within the European Union of the last decade, these differences within the EU may have decreased. If this is the case, practitioners could choose an international peer group. Thereby, a selection from more companies could be made and comparable companies more similar to the target company2 may be selected. To test the viability of using international peer groups, this study investigates valuation differences between companies from the 27 countries of the European Union (EU27) in the period 2001-2010. The existence of country differences in valuation multiples is determined by comparing the valuation errors and dispersion of these errors, between international and domestic peer groups.
Corporate valuation methods can be divided into direct and indirect methods, with the direct methods being for example, the Discounted Cash Flow (DCF) analysis. A DCF analysis requires forecasts of for example sales, profit margins and investments, for several years. The notion of an efficient market implies that this information is already incorporated into stock prices. Indirect methods are designed to combine this notion with the use of accounting-based market multiples, to make corporate valuation less time-consuming. However, Koller et al (2010) conclude that a well-executed multiples valuation takes time as well. Yet, they promote the use of multiples valuations, because different methods provide additional insights in the value of a company. Valuations of companies make use of a tracking portfolio based on comparable companies. In a DCF-valuation the expected cash flows are discounted with the expected rate of return of the tracking portfolio. A valuation based on multiples uses the market value of the tracking portfolio to calculate the value of the company. From this perspective, multiples can be used as substitutes for comprehensive valuations. In addition, multiples valuations are used to put the value obtained from a direct approach into a current market perspective. Practitioners use the multiples valuation method in seasoned equity offerings, IPO-valuations, leveraged buyout transactions and other M&A activities.
However, little research has been executed into the question regarding the effects of including foreign companies in the peer group. The main research question in this study is: Does a domestic peer group lead to lower valuation errors in the multiples valuation approach than an international peer group for
2 countries within the EU27? The valuation multiples under investigation are both enterprise (enterprise value/invested capital and enterprise value/earnings) and equity (price/book and price/earnings) multiples. I also investigate which independent variables in the regression are significant in determining the multiple and how large their effect is. Furthermore, the valuation errors of the enterprise multiples are investigated. International and domestic peer groups are selected with two different methods. First, the systematic method follows the approach of Bhojraj and Lee (2002) by creating a systematic multiple3 based on fundamental variables. This multiple is used to select the closest comparable companies of a target. However, the assumption of linearity and the use of proxies for the fundamental factors in the analyses, may create new errors. Therefore, a second selection method is employed, where peer groups are selected based on the industry classification of the target company, following Alford (1992), Liu et al. (2002, 2007) and Schreiner and Spremann (2007). The median, mean and harmonic mean multiples of the peer groups are calculated, to investigate valuation error resulting from the use of different measures of central tendency. Finally, the research will address the effects on valuation error and its dispersion, of including ten instead of five comparable companies in the peer group of the target company.
This study finds, that the EV/EBIT multiple outperforms the EV/IC multiple, in all years and for all measures, both selection methods and peer groups. Also, the pooled regression analyses show, that differences between multiples of ten to fourteen countries cannot be explained by fundamental factors. The average cross-sectional R2 values are between 5.8% and 11.6%. However, selecting peer groups based on industry classification leads to a significant outperformance of the international compared to the domestic peer groups, in all years and for all measures and both multiples. When the peer groups are selected with the systematic multiple, lower valuation error (insignificant in most years) is observed for the domestic compared to the international peer group, in most years. Furthermore, the industry selection method has a significantly lower valuation error and standard deviation compared to the systematic method, in almost all years and peer groups. This proves that there are no valuation differences between companies from countries within the EU27, that are unaccounted for by the industry classification of a company. The systematic selection method only outperforms the industry method with the harmonic mean measure of the domestic peer group of the EV/EBIT multiple and the average measure of both peer groups of the EV/IC multiple. In general, the lowest valuation errors are observed by using the harmonic mean measure. However, this measure also has the highest standard deviation for both multiples, peer groups and selection methods. The mean measure has the lowest standard deviation percentage in all years and for all measures and peer groups. Finally, increasing the peer group from five to ten companies, leads to a reduction in the pooled average absolute valuation errors of 1.5% to 1.9% (EV/EBIT) and 1.2%
3 to 7.3% (EV/IC). The average percentage decrease in dispersion is comparable for both multiples and lies between 5.8% and 8.2%. It can be concluded that including more companies in the peer group leads to lower valuation errors and lower dispersion. Prior literature finds a decrease in dispersion when the number of companies in the peer groups is increased.
2.
Literature Review
Several approaches are used within the field of corporate valuation. These approaches can be divided into direct and indirect methods, with the direct methods consisting of for example the well-known DCF-analysis. This method requires forecasts of financial information, which is difficult to estimate accurately. Practitioners use valuation multiples to triangulate the results obtained from a direct approach, and put them into a current market perspective. Demirakos et al (2004) investigate the use of different valuation techniques in the United Kingdom and find that most analyst reports are based on the multiples valuation approach. In addition, Asquith et al (2005) find that 99% of the analyst reports within their sample (1997-1999), are based on multiples valuation. Table 1 (page 7) presents an overview of prior literature.
2.1 Types of Multiples
The value of a company is equal to a multiple of an accounting measure of that company. A multiples valuation bases the multiple on a set of comparable companies. Multiplying the peer group multiple and the accounting measure of the target company, gives the value of the target company.
Valuation multiples are based on the following formula (Koller et al [2010]):
4 The use of cash flow multiples is motivated by the assumption that reported cash flow is the best available proxy for the future cash flow, that is used in a DCF-analysis, and that cash flow multiples are less
susceptible to manipulation by management (Liu et al [2002]). Earnings multiples are mostly based on Earnings Before Interest Taxes (EBIT), Earnings Before Interest Taxes and Amortization (EBITA), Earnings Before Interest Taxes Depreciation Amortization (EBITDA) or Net Income (NI). Capital multiples are based on the invested capital (IC) of the company and sales multiples are based on the sales of the company. The cash flow, earnings and sales multiples are based on the profit and loss statement, whereas capital multiples are based on the balance sheet of a company. Liu et al (2002) find, for their US sample between 1987 and 2000, that sales multiples significantly outperform cash flow multiples and earnings multiples outperform sales multiples, in terms of valuation error.
The numerator of the multiple can be either the enterprise or equity value of the company, referred to respectively as enterprise and equity multiples. Enterprise value is equal to the market value of equity plus the value of debt minus cash. Corporate finance practitioners prefer enterprise multiples, because the cash flow of a company is obtained with both equity and debt. Bhojraj and Lee (2002) find lower
valuation error for the Price/Book multiple (P/B) compared to the enterprise value/sales multiple (EV/Sales), for their US sample. Schreiner and Spremann (2007) find that price-multiples have 23% lower median absolute valuation error compared to enterprise multiples. Liu et al (2002) also find that equity multiples outperform enterprise multiples. However, Hermann and Richter (2003) conclude that enterprise multiples outperform equity multiples. Koller et al. (2010) identify two drawbacks with the equity value/earnings multiple (P/E): (1) it is affected by the capital structure of the company and (2) it includes non-operating items, such as restructuring charges and write-offs. The capital structure also affects enterprise multiples, because it can change the Weighted Average Cost of Capital (WACC). An increase in debt leads to an increase in tax shields, if there is sufficient taxable income. This will lower the WACC and increase the value of the company. However, the increase in debt also leads to an increase in risk to equity holders, which will increase the WACC. Bhojraj et al (2003) compare the accuracy of enterprise and equity multiples for the countries of the G7 within the period 1990-2000. They find that the EV/Sales multiple outperforms the P/B multiple. In addition, Liu et al (2007) and Schreiner and
Spremann (2007) find that earnings multiples outperform multiples based on the book value of equity or
5 sales. Tu (2010) finds that equity multiples outperform enterprise multiples and that EBIT-multiples are superior to other earnings and cash flow multiples, in assessing the target value.
Another important distinction between multiples is whether forward-looking or backward-looking
information is used for the denominator of the multiple. Forecasts made by analysts for a number of years ahead, are used as forward-looking information. Liu et al (2002) find that forward-looking multiples lead to lower valuation errors compared to backward-looking multiples. This result is confirmed by Liu et al (2007) and Schreiner and Spremann (2007). Tu (2010) investigates several different types of multiples and examines the improvement of valuation error when using forward- instead of backward-looking multiples, for a sample of European countries in the period 1999-2009. He concludes that forward-looking multiples outperform all other multiples except for the sales multiple. Contrary to this, Henschke and Homburg (2009) conclude that it is more important to use the best fundamental variables rather than the best type of multiple.
Next, a distinction can be made between trading and transaction multiples. Trading multiples are based on observed market prices of companies, whereas transaction multiples are based on actual transaction prices that have been paid for companies as part of an acquisition or a merger. When investigating transaction multiples, companies do not need to be listed. Transaction multiples are higher than trading multiples, since corporate transactions change ownership, control, and management structure. During favorable market conditions, the premium that is required in corporate transactions can be as large as fifty percent (Tu [2010]). Kaplan and Ruback (1995) investigate a US sample of highly leveraged transactions between 1983 and 1989 and compare the values obtained from a DCF-valuation and a multiples valuation with the actual transaction values. They find that a multiples valuation based on comparable companies
underestimates value compared to the actual transaction value. When peer groups are selected using the industry classification, the company value comes closer to the paid transaction price. However, the DCF-approach remains a more accurate valuation technique. Kim & Ritter (1999) investigate the use of multiples of comparable companies to value IPO‟s and find that forward-looking multiples outperform backward-looking multiples, similar to Kaplan and Ruback (1995).
6 group based on the fundamental variables does not decrease the valuation error more, than a peer group based on 3-digit SIC-codes. Several authors have implemented a regression approach, to assess the valuation errors of different fundamental variables and for example industry classification. Bhojraj and Lee (2002) implement a regression model for their US sample in the period 1982-1998, and base the selection of comparable companies on the regression results. First, they combine industry classification and eight fundamental variables, by subtracting the median industry value for a variable from the value of each firm. In this manner, they include industry effects indirectly in the selection of comparable
companies. Then, they create a systematic multiple by multiplying the coefficients from the regressions with the industry-adjusted values of each variable for each company. The six companies with the closest systematic multiple are included in the peer group. Finally, the systematic multiple is used to forecast three-to-five year-ahead multiples in another regression model. Bhojraj and Lee (2002) conclude that the systematic multiples outperform (based on the R2) the standard multiples based on industry classification in forecasting three-to-five year-ahead multiples. Hermann and Richter (2003) use a regression model to assess valuation differences between multiples based on industry classification and four fundamental variables. They investigate enterprise and equity multiples and find that industry classification has no additional contribution in selecting comparable companies. Henschke and Homburg (2009) also utilize a regression approach in examining capital and earnings multiples, in the period 1985-2004 for a US dataset. First, they calculate the valuation errors of the peer group to the target. Then, they use six fundamental variables to explain the errors in a regression analysis. They conclude that it is more important to choose the correct fundamental variables, as opposed to using the correct multiple and that the industry
classification is unimportant in selecting peer groups when the selection is based on fundamental factors.
7
Sample years M ultiples Comparables selected
based on Outliers Sample requirements Sample description
Country
effects Results Alford (1992) 1978, 1982,
1986 P/E
Industry, firm size,
earnings growth No information
EBIT is available and positive, fiscal year ends in December
1636 firms in 1978, 1591 firms in 1982, 1471 firms in 1986 N/A
Industry membership or a combination of risk and earnings growth are good proxies for comparable company selection Baker and Ruback (1999) 1995 unclear M ultiple No information
M V-equity, book value of debt, revenue, EBITDA and EBIT are available
225 S&P500 firms N/A EBITDA outperforms the EBIT and Sales multiple Bhojraj and Lee (2002) 1982-1998 EV/Sales & M /B 'Warranted multiple'
Top & bottom 1% observations on each multiple and variable is deleted
US firms with M arket Cap> $100 mln 3515 firms from the NYSE, NASDAQ, AM EX in 2000 N/A
Warranted multiple outperforms straightforward methods based on industry classification
Bhojraj and Lee (2003) 1990-2000 EV/Sales & P/E & forward P/E
& M /B 'Warranted multiple'
Top & bottom 3% observations on each multiple and variable is deleted
Earnings are available and positive 26,626 firm years from G7 country firms No
EV/S multiples outperform M /B multiples in terms of valuation accuracy. Growth and R&D information enhances explanatory power, but leverage does not Dittmann and Weiner (2005) 1993-2002 EV/EBIT OECD, country, region
Top & bottom 1% observations on each multiple and variable is deleted
Total assets, EBIT and total debt are
available and positive 67,433 firm years Yes
UK and US firms are comparable in valuation and differ from EU valuation
Henschke and Homburg
(2009) 1985-2004 P/E & M /B
Industry, size, leverage, growth, return on equity, R&D
Top & bottom 1% observations on each multiple and variable is deleted
Book value > $10 mln and net sales are
positive 24,308 firm years N/A
Valuation error based on industry-based comparable company selection can be explained by differences in financial ratios
Hermann and Richter (2003) 1997-1999
EV/EBEITDAAT & EV/EBIAT & EV/IC & EV/Sales & P/E & M /B
Industry, growth, leverage, reinvestment rate, return on equity
No information No information
524 US and 830 EU firms, 645 in 1997, 665 in 1998, 664 in 1999
N/A EVmultiples lead to more accurate results than equity multiples. EV/Earnings outperform EV/IC Kaplan and Ruback (1995) 1980-1989 EV/EBITDA Industry, transaction No information
Transaction value > $100 mln and industry classification is available, management obtains equity-stake
51 highly levered transations N/A DCF performs best, followed by comparable transaction valuation and lastly multiple valuation
Kim and Ritter (1999) 1992-1993 P/E & M /B & P/Sales Industry
P/E ratios above 100 are maximised at 100, M /B ratios above 10 are maximised at 10
Positive earnings per share and book
value per share 190 IPO's N/A
Valuation accuracy is higher for old firms than for young firms. EV/Sales accuracy improves when multiples are corrected for profitability and leverage
Lie and Lie (2002) 1998
EV/EBITDA & EV/EBIT & EV/Sales & EV/Book value & P/E & forward P/E
Industry No information No information 8,621 firms from Compustat database N/A
The estimation errors are negative in mean and EBITDA multiles outperform EBIT multiples in terms of accuracy Liu et al (2002) 1987-2001
EBITDA/P & CF/P & Dividend/P & Sales/P & EBITDA/EV & Sales/EV
Industry
Top & bottom 1% observations on each multiple and variable is deleted
Price > $2 and multiples and value
drivers are positive 19,879 US firm years N/A
M ultiples based on forward looking estimates outperform EBITDA and M /B valuations, EBITDA and M /B valuations dominate estimates based on CF Liu et al (2007) 1982-1999 EBITDA/P & M /B & B/P &
Sales/P Country, industry
Top & bottom 1% observations on each multiple and variable is deleted
M ultiples are positive
Firms from Australia, Canada, France, Germany, Hong Kong, Japan, South Africa, Taiwan, UK and US
Yes
M ultiples based on forward looking estimates outperform trailing multiples. There are significant differences among countries in terms of valuations and pricing errors Richter (2005) 1996-1998 P/E & M /B
Growth, industry, leverage, reinvestment rate, return on equity
No information No information 332 US firms, 210 German firms and 455 other EU firms N/A
Valuations based on value-driver-comparables outperform simple SIC-code classified peer group valuations Ros (2010) 1998-2008 EV/Sales & EV/EBIT &
EV/EBITDA
Country, industry, sales, sales growth, profit margin
Top & bottom 1% observations on each multiple and variable is deleted
M ultiples are positive 33,987 EU firm years Yes Country is a significant variable in estimating multiples in the EU, EV/Sales has the highest R-squared
Schreiner and Spremann
(2007) 1996-2005 P/E & M /B & P/Sales Industry No information
M arket Cap > $200mln and net debt is positive and multiples must be positive
Dow Jones STOXX 600 & S&P500 N/A
Equity value multiples outperform EV multiples, knowledge related multiples outperform traditional ones, forward looking multiples outperform trailing multiples Tu (2010) 1999-2009 24 multiples based on cashflow
and earnings
Industry, return on assets, size
Top & bottom 1% observations on each multiple and variable is deleted
Market cap > €20mln, Price > €1.5 26,487 EU firm years N/A
Forward-looking multiples outperform trailing multiples, equity and EBIT multiples are preferred, optimal peer group consists of 5 companies
Table 1 Overview of Prior Literature
8 2.2 Literature on International Differences in Multiples Valuation
There must be reasons for countries to differ from each other, so that valuation differences between countries exist. Gray (1980) finds that there are significant differences in accounting practices on profits, between France, Germany and the UK. He identifies differences in tax and consolidation accounting, the legal framework and disclosure practices, to be the main reasons for the significant differences. La Porta et al (1998) examine legal rules with respect to the level of protection of shareholders and creditors. Differences in this level of protection and the quality with which the legal rules are enforced, lead to differences in profits measurement and as such in valuation differences using the multiples valuation approach. They conclude that countries with weaker investor protection have less developed debt and equity markets. In less developed markets there is less stringent control and liquidity, which leads to less accurate pricing of stock and debt. Furthermore, La Porta et al (1998) find that civil law (for example Germany, France, Scandinavia) assigns weaker legal rights to investors than common law (for example US and UK) does. In addition, Ball et al (1999) conclude that accounting income within countries with a civil law system experiences significantly more timeliness due to income conservatism. This means that income is acknowledged later due to more conservative practices in calculating accounting income. Choi and Meek (2008) present eight factors that have a significant influence on the development of accounting systems. These factors are: sources of finance, legal system, taxation, political and economic ties,
inflation, level of economic development, educational level and culture. Accounting within equity-based systems is designed to facilitate investors who assess cash flows and risks in countries with strong equity markets (US/UK). On the other hand, accounting within credit-based systems focuses on creditor
protection through conservative earnings measures. Of course, the increasing importance of a uniform accounting standard (IFRS) should reduce international differences between companies. However, Choi & Meek (2008) note that a uniform accounting standard is a too easy solution for a complex problem, especially for smaller firms, because of its complexity and disclosure requirements. Also, they expect that cultural differences and the past development of national accounting systems are factors that will remain to separate countries from an accounting perspective.
9 together with the industry median. This might be a reason for their result that valuation differences between countries do not exist. Also, Bhojraj et al (2003) do not investigate valuation error. Instead, they assess the accuracy of multiples by comparing their predictive power in forecasting a company‟s three-to-five year-ahead multiples. Second, Ditmann and Weiner (2005) investigate differences between multiples in fifteen European countries and the USA, for the period 1993-2002. They find that only for companies from the UK, the USA, Denmark and Greece a domestic peer group should be used. There is no clear trend of valuation error over time. Furthermore, Ditmann and Weiner (2005) only use the industry classification and return on assets as variables in selecting comparable companies. Finally, Ros (2010) finds that there are significant differences between multiples from different countries. He also performs a regression analysis for fifteen European countries in the period 1998-2009 and concludes that the country-variable is significant in estimating multiples. Furthermore, he finds that the country-effect is significant for three, five and eight (EV/EBIT, EV/Sales and EV/EBITDA, respectively) countries. He concludes that some of the differences between multiples from different countries can be explained by fundamental factors.
10
3.
Data and Methodology
3.1 Sample Data
This section describes the sample and data. The data included in this study are obtained from the Orbis2 database. This database contains 73,163,779 companies, of which 28,528,740 companies have their office registered in one of the 27 countries currently part of the EU4. Only companies with a NACE rev.2 industry classification code are selected, of which financial firms are excluded due to the differences in valuing these companies (Lie and Lie [2002]), leaving 22,774,373 companies. Furthermore, only
companies that have a stock listing are included in the sample, because the market capitalization is used in calculating the valuation multiples. This leaves 7,049 companies. The financial information for the period 2001-2010 is retrieved from Thompson‟s DataStream, which contains 5,206 companies from the list of 7,049 companies.
The multiples are calculated as follows. The enterprise value (EV) is defined as the market capitalization (WC08001) + Debt (WC03255) – Cash & cash equivalents (WC02001). Earnings before interest and taxes (EBIT) (WC18191), sales (WC01001), assets (WC02999) and net income (WC01706) are directly retrieved from DataStream. Price is defined as the market capitalization of the company. Invested capital (IC) is calculated by adding the book value of equity (WC03501) to debt and subtracting cash. All the variables are backward-looking even though prior literature has established that forward-looking information is preferred. However, forward-looking financial information is unavailable in Thompson‟s DataStream and therefore not used in this study.
In accordance with Bhojraj and Lee (2002), Dittmann and Weiner (2005) and Henschke and Homburg (2009), the top and bottom 1% observations of all multiples and variables is deleted. Most of the prior literature poses additional restrictions on the data. For example, Bhojraj and Lee (2002) exclude
companies with a market capitalization lower than $100 mln and Henschke and Homburg (2009) require the book value of equity to be larger than $10 mln. In this study, the enterprise value, invested capital, price, book value of equity, EBIT and net income (NI) must be positive, because otherwise the valuation multiples are negative (Liu et al [2002], Schreiner and Spremann [2007], etc.). As a consequence, the average multiple will be biased upwards. To solve this problem, the average multiples are adjusted downward by deleting multiples larger than 30 5 . The regression has also been executed including multiples larger than 30. However, then the coefficients of the fundamental and dummy variables are
4
Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Germany, Denmark, Estonia, Spain, Finland, France, UK, Greece, Hungary, Ireland, Italy, Lithuania, Luxemburg, Latvia, Malta, The Netherlands, Poland, Portugal, Romania, Sweden, Slovenia, Slovakia.
11 EV/IC P/B EV/EBIT P/NI
Number of observations 27,355 33,875 19,618 19,596 Average 2.35 2.66 11.80 13.96 Median 1.38 1.59 10.76 13.28 Harmonic Mean 0.72 0.95 2.57 5.78 Standard Deviation 3.19 3.36 6.21 6.62 Minimum 0.03 0.005 0.001 0.001 Q1 0.89 0.87 7.40 9.01 Q3 2.40 3.05 15.17 18.35 Maximum 29.97 29.98 30.00 30.00 Skewness 4.24 3.63 0.73 0.36 Kurtosis 25.80 20.37 3.15 2.49 Jarque-Bera 0.000* 0.000* 0.000* 0.000*
Table II
Descriptive Statistics of the Valuation Multiples This table presents the descriptive statistics of each multiple. EV is defined as market capitalization + debt – cash & cash equivalents. IC is defined as book value of equity + debt – cash & cash equivalents. P(rice) is defined as market capitalization and B as book value of common equity. NI is the abbreviation for Net Income. The p-value of the Jarque-Bera
test is presented, with * indicating significance at the 5%-level.
extremely high (in some cases >100). This leads to problems in the next step of the methodology, where these coefficients are used to select companies.
Table II presents the descriptive statistics of each multiple that is investigated in this study. In addition to the average, the median and harmonic mean are also presented, to compare the effect of using different measures of peer group multiples. The median values per type of multiple are within the range observed in prior literature (Lie and Lie [2002] and Ros [2010]). The high Jarque-Bera statistic of each multiple indicates a non-normal distribution.
12 P/B EV/EBIT P/NI Austria 1.44 10.48 12.75 Belgium 1.37 11.50 12.71 Bulgaria 0.93 10.62 9.78 Cyprus 0.49 10.04 8.55 Czech Republic 0.91 6.82 11.37 Germany 1.59 9.90 13.53 Denmark 1.38 10.54 13.06 Estonia 1.51 15.92 14.53 Spain 1.98 11.82 13.60 Finland 1.80 10.69 14.15 France 1.66 10.54 13.39 United Kingdom 1.75 10.43 13.17 Greece 1.06 12.21 13.74 Hungary 1.04 8.76 10.40 Ireland 2.33 11.61 13.98 Italy 1.55 11.01 14.71 Lithuania 1.00 15.03 11.19 Luxemburg 1.47 12.31 12.94 Latvia 0.55 8.51 8.10 Malta 1.59 14.81 19.10 The Netherlands 1.85 10.23 12.92 Poland 1.67 11.40 13.57 Portugal 1.32 13.00 13.01 Romania 0.80 10.84 10.03 Sweden 2.12 11.17 13.54 Slovenia 1.01 13.73 13.03 Slovakia 0.50 7.29 8.88 Test statistic 1,044.48 3,597.76 3,181.83 p-value 0.000* 0.000* 0.000* 0.000* 0.78 1.74 0.92 0.70 Kruskal-Wallis test 1,785.20 1.28 0.79 1.35 1.60 1.29 1.14 1.50 1.19 0.98 1.71 1.32 1.43 1.41 1.22 2.33 1.23 1.45 1.42 EV/IC 1.12 1.20 1.05 0.56 0.77 Table III
Median Multiples per Country
This table shows the median multiples per country for the period 2001-2010. The Kruskal-Wallis test indicates the relevance of the country classification in
explaining multiples. * indicates significance at the 5%-level.
13 Median Average 2001 0,305 0,415 2002 0,315 0,424 2003 0,305 0,422 2004 0,282 0,396 2005 0,264 0,395 2006 0,263 0,386 2007 0,272 0,412 2008 0,33 0,483 2009 0,31 0,479 2010 0,286 0,438
Table IV
Median and Average Leverage Rate per Year for the Period 2001-2010
countries increases on average by 11.3%. The leverage rate of the „older‟ EU countries remains approximately the same when 2001 and 2010 are compared. Secondly, economic prospects clearly deteriorate from 2007 onwards. However, unlike the leverage rate, profitability and growth are normalized by averaging the last three years. This leads to a smaller downfall of these variables from 2007 to 2010.
Graph 1
14 Fundamental Drivers of Value
To learn more about the intuition behind valuation multiples, a closer look at the drivers of value of a company needs to be taken. Koller et al (2010) present the „Key value driver formula‟ (3), which combines the fundamental drivers of value: growth, profitability and cost of capital.
Where: EV = Enterprise Value
Noplat = Net operating profit less adjusted taxes
ROIC = Return On Invested Capital WACC = Weighted Average Cost of Capital
EBITA = Earnings Before Interest, Taxes and Amortization T = corporate income tax rate
Based on (3) the following can be obtained:
where EBI represents Earnings Before Interest and equals Noplat after amortization. Since EBI is not directly available from Thompson„s DataStream it has to be calculated separately. However, this leads to a reduction in the number of multiples due to a lack of information on amortization. Therefore, EBIT (Earnings Before Interest and Taxes) is used, which is directly available from Thompson‟s Datastream. Now, the numerator on the right side of equation (4) has to be multiplied by (1-T), to bring taxes back into both sides of the equation. Table V gives an overview of the equations of the different types of valuation multiples and the variables that determine the multiples. The equations of the other multiples in Table V are also based on (3).
15
expected sign expected sign expected sign
r mean last 3 year return
on operating assets +
mean last 3 year return on
equity
+ mean last 3 year return on
equity assets +
mean last 3 year
return on equity +
g mean last 3 year sales
growth +
mean last 3 year sales
growth
+ mean last 3 year sales growth + mean last 3 year
sales growth +
k last year leverage rate + last year
leverage rate - last year leverage rate +
last year leverage
rate
-T NA NA corporate income tax rate +/- NA
size last year sales - last year sales - last year sales - last year sales
-industry NACE rev.2 industry
code
+/-NACE rev.2
industry code +/- NACE rev.2 industry code
+/-NACE rev.2
industry code
+/-country country of registered
office address
+/-country of registered office address
+/- country of registered office
address
+/-country of registered office
address
+/-year year in which multiple
is recorded
+/-year in which multiple is
recorded
+/- year in which multiple is
recorded +/-year in which multiple is recorded +/-Variables expected sign g k g r IC EV g k g r B P g k r g T EBIT EV 1 * 1 g k r g NI P 1
Table V
The Definitions of the Multiples, the Fundamental and Dummy Variables
This table shows the definitions of the valuation multiples and the fundamental and dummy variables that are used. The expected sign of each variable is also included in the table. The signs are based on
16
EV/IC P/B EV/EBIT P/NI
Test statistic 1,013.14 274.15 808.36 646.52 p-value 0.000* 0.000* 0.000* 0.000* Test statistic 175.95 628.50 42.60 136.25 p-value 0.000* 0.000* 0.000* 0.000* Test statistic 34.14 9.92 16.12 14.59 p-value 0.000* 0.357 0.064 0.103 Test statistic 2,108.24 3,715.02 500.41 417.95 p-value 0.000* 0.000* 0.000* 0.000* Test statistic 3,873.66 6,057.83 918.55 614.16 p-value 0.000* 0.000* 0.000* 0.000* Leverage Size Growth Roe Roa Table VI
Results of the Kruskal-Wallis Test on the Fundamental Variables Each fundamental variable is divided into 10 classes. Significant p-values indicate that there are significant differences between the groups per variable and multiple. A regression is executed, to establish which variables are significant in explaining multiples. The four fundamental variables and the dummy variables included in the regression are discussed next.
Leverage Rate and Size
The leverage rate equals the debt/equity ratio. In accordance with Bhojraj and Lee (2002) and Hermann and Richter (2003), the leverage rate is used to proxy for risk (cost of capital, k). The expected signs presented in Table V, are based on the direct and indirect effect of that variable on the multiple. The direct effect is that an increase of the leverage rate increases the cost of equity capital and hence decreases the equity multiples. On the other hand, an increase of the leverage rate decreases the cost of capital of the enterprise (WACC) and increases the enterprise multiples. The indirect effect is that the decrease of the cost of capital of the enterprise might be (partly) offset by an increase of the cost of debt. Generally, higher levels of debt are associated with higher costs of debt. Gebhardt et al (2001) find that companies with a higher leverage rate have a higher implied cost of equity capital. Bhojraj and Lee (2002) find a negative sign of the leverage rate for each type of multiple. A positive sign of the leverage rate is observed for the enterprise multiple, in the paper by Bhojraj et al (2003). They find that the sign differs over time for the equity multiples. Ros (2010) also includes last year‟s sales as a measure of size and finds size to be significant at the 5%-level in all of his regressions. Therefore size has been added to the
17 Table VII
Descriptive Statistics of Growth, Profitability, Leverage and Size in the Period 2001-2010
Growth is defined as sales growth, Roa and Roe as return on assets and return on equity (all last three year averages). Size and leverage represent last year‟s sales and leverage rate. *
indicates significance at the 5%-level.
Growth Leverage Size (in € mln) Roa Roe
Average 20.54% 42.81% 1370 10.27% 5.89% Median 8.54% 29.22% 104 11.36% 9.24% Harmonic Mean 2.74% 3.10% 9 na na Standard Deviation 54.22% 48.25% 4.76 79.43% 42.04% Q1 0.30% 11.04% 23.37 1.82% 0.11% Q3 21.61% 56.70% 534.43 23.85% 19.38% Skewness 5.89 2.66 6.32 -1.30 -2.03 Kurtosis 49.97 13.31 50.15 31.32 21.20 Jarque-Bera 0.000* 0.000* 0.000* 0.000* 0.000* Growth and Profitability
As can be seen from Table V, the proxies for growth and profitability are respectively, sales growth and return on operating assets (roa, used with enterprise multiples) or return on equity (roe, used with equity multiples). These variables are also used by Ros (2010), whereas Bhojraj and Lee (2002) use earnings growth as a measure of growth. Both sales and earnings growth are used by Hermann and Richter (2003), who find that both measures lead to the same results. The formula in (3) assumes growth and profitability (ROIC) to be constant (Koller et al [2010]). This means that any excess returns on assets are expected to exist forever. Therefore, the variables profitability and growth are normalized by averaging the values of the last three years, comparable to Ros (2010). This adjustment assures that the variables become more reasonable estimates of future growth (Koller et al [2010]). Additional financial information for the years 1998-2000 is used to calculate these three year averages. Hermann and Richter (2003) average the values of each variable of the last four years, whereas Bhojraj and Lee (2002) do not normalize their variables. The corporate income tax rates have been obtained from the OECD-database. The sign of growth is expected to be positive in general. However, the values of profitability and cost of capital have important implications for the effect on the multiple of a change in growth. For example, if an increase in growth is obtained by a decrease in profitability, the effect of growth on the multiple can be negative. However, prior literature finds growth to have a positive sign for every multiple (Ros [2010], Bhojraj and Lee [2002] and Hermann and Richter [2003]). A similar argument can be made for profitability, where prior literature finds positive and negative signs for different multiples in different years. It is expected here, that
18 Dummy Variables
In order to include and examine industry-effects in explaining valuation multiples, dummy variables based on the NACE rev.2 industry classification system6, are included in the regression. Ros (2010) also uses the NACE rev. 2 classification system, whereas Dittmann and Weiner (2005), Bhojraj and Lee (2002) and Hermann and Richter (2003) use the SIC classification system. The inclusion of the industry dummy variables in the regression, is merely to control for the industry-effects found by Alford (1992) and others. This study follows Liu et al (2002) and Ros (2010) by including a two-digit industry classification, which results in ten broadly defined industry classes. Increasing the number of industry classes leads to a reduction in the number of companies that represent an entire industry and the number of companies from which the closest peer group can be selected. Contrary to this study, Bhojraj and Lee (2002) do not include industry dummies in their regression, but subtract the industry median from five of their eight explanatory variables for each company. However, since most countries have a few main industries, part of the country-effect may be subtracted, by using that approach. This might be an explanation for the „interesting fact‟ of Bhojraj & Lee (2002), that the original multiple of the comparable companies
(selected based on the systematic multiple) has more explanatory power than the systematic multiple itself. The Kruskal-Wallis test in Table VIII shows that a classification based on industry has a significant effect on multiples. Appendix B1 presents the median valuation multiples per industry.
Next, dummy variables for every sample year are included in the regression to capture time-effects. The problem of perfect multicollinearity arises when all dummy variables are included in the regression. This problem makes the estimation of coefficients impossible (Brooks [2008]). To avoid this problem, one industry dummy variable, one year dummy variable and one country dummy variable are excluded. To some extend valuation multiples are affected by interest rates and inflation, which change over time (Koller et al (2010). To eliminate the time-effects, 10 yearly (cross-sectional) regressions are also performed. Appendix B2 presents the median valuation multiples per year. Table VIII indicates that multiples differ significantly over time. Finally, dummy variables for each of the 27 countries of the EU, except for Malta, are included in the regression. These variables are most important, since they determine how much of the valuation multiples of each company can be explained by the country the head office of that company is registered in.
6
19
EV/IC P/B EV/EBIT P/NI
Test statistic 414.47 342.20 380.11 131.88 p-value 0.000* 0.000* 0.000* 0.000* Test statistic 635.65 3,551.29 319.65 138.83 p-value 0.000* 0.000* 0.000* 0.000* Industry Year
Table VIII
Results of the Kruskal-Wallis Test
Each fundamental variable is divided into 10 classes. Significant p-values indicate that there are significant differences between the groups for each variable and multiple.
3.2
Methodology
The main hypothesis in this paper is whether the country, the head office of a company is located in, has a significant effect on the valuation of that company. Tables VI and VIII have already established that multiples significantly differ between companies classified by profitability, leverage, industry and time. A panel regression analysis is executed in order to establish, which fundamental and dummy variables are significant and how large their impact is in explaining valuation multiples over the entire dataset.
Therefore, a regression is performed including the fundamental variables: sales growth, leverage rate, sales, return on operating assets (for enterprise multiples) and return on equity (for equity multiples), and dummy variables for each, except for one, industry, year and country. This results in the following regression equation:
White‟s cross-section standard errors are used to control for possible heteroscedasticity. To combine both cross-sectional and time-series information, a panel estimation technique is used with the valuation multiple as the dependent variable. Correlations between the independent variables that are used in the same regression are below 0.08 (Appendix C).
Next, cross-sectional regressions are executed to eliminate time-effects. The coefficients of the significant variables with the appropriate sign are used to calculate a systematic multiple. Ros (2010) bases his conclusions regarding the country-effect solely on the regression analysis. However, there are two reasons for including another step in this study. Firstly, the regression approach assumes that the relation between the variables and the multiples is linear, whereas equation (5) is not linear at all. Secondly, the variables in equation (5) are in expected terms, but proxies are used in the regressions since the expected values are unavailable. These arguments support potential deviations between the results from the regressions and valuation errors. The systematic multiple is used to find the closest yearly comparable companies of each
20 firm. Bhojraj and Lee (2002) also perform cross-sectional regressions and use the coefficients to calculate a systematic multiple. Since the systematic multiple is based only on significant variables in explaining multiples, it is easy to see that the actual multiple should equal the systematic multiple and an error term:
(6)
The coefficients of the significant variables, with the appropriate sign in the regression analyses, are multiplied by the actual values of the variables of each company. Coefficients of significant dummy variables are simply added to the systematic multiples of the relevant companies. The regressions are also performed without the insignificant variables. However, this does not change the coefficients of the remaining variables. Then, each company‟s closest comparable companies are selected by searching for the closest systematic multiples. However, this methodology could lead to a potential problem by not selecting the actual closest comparable companies. For example, if there are two companies and the systematic variable is calculated based on two fundamental variables (a and b) with coefficients of 2 and -2. The financial information for a and b of the first company are both 1 and for the second company both 5. In this case, the systematic multiple of both companies is 0 (-2*1+2*1 and -2*5+2*5), which means that they are identical to the target company. However, per variable the first company is closer to the target in absolute terms, which suggests a difference between the first and second company relative to the target company, that is not captured by the methodology employed here. Henschke and Homburg (2009) calculate the absolute differences of their variables of each company to the target and add these together. However, they use these differences as independent variables in a regression analysis to explain the valuation error. Based on the definition in (6) and the fact that prior literature does not reject the methodology, this study follows Bhojraj and Lee (2002) in calculating a systematic multiple.
The systematic multiple is calculated for the enterprise multiples, because these multiples are used most by practitioners. Besides selecting comparable companies based on the systematic multiple, they are also selected based solely on the industry classification. Alford (1992) and Dittmann and Weiner (2005) also compare the results of the industry selection method with results of a fundamental selection method.
21 systematic multiple industry multiple
international xx xx
domestic x x
Table IX
Comparison of Groups and Selections
This table shows the tests that are performed for the EV/IC and EV/EBIT multiples. The international peer group is compared to the domestic peer group for both selection methods (x‟s and xx‟s).
The selection methods are compared for both peer groups (white and grey area)
selection methods
peer groups
assigns less weight to outliers. All three measures are calculated for the peer groups to find out which measure results in the lowest valuation error. The median is used by Alford (1992) and the harmonic mean by Bhojraj and Lee (2002) and Tu (2010). It is important to note that the systematic multiple is only used to select comparable companies.
Tu (2010) concludes that a peer group consisting of five companies is optimal, whereas adding additional companies comes at a cost of adding more noise to the distribution. In addition, Henschke and Homburg (2009) find that a peer group consisting of four to eight companies is preferred and Bhojraj and Lee (2002) use the six closest peers. Cooper and Cordeiro (2008) select comparable companies based on the growth rate and find the optimal number of companies to be five. This study selects five comparable companies.
The nonparametric Wilcoxon signed rank test is performed, to test if there are significant differences between the valuation errors (calculated with the median peer group multiple) of the international and domestic peer groups. Paired samples t-tests are executed to test for differences between the valuation errors (calculated with the mean and harmonic mean peer group multiple) of the international and domestic peer groups. Furthermore, the valuation errors are estimated by subtracting the multiple of the peer group (for the measures: median, average and harmonic mean) from the target multiple. These valuation errors are divided by the actual target multiples to obtain the „valuation error percentage‟. The valuation error percentages and the dispersion of these valuation errors are combined, to assess the valuation accuracy of the peer groups and selection methods for all measures. However, it is easy to see that the absolute valuation error has to be calculated for the peer groups selected with the industry
22 expected observed expected observed expected observed expected observed
Growth + 0 + 0 + 0 + + Leverage + + - 0 + + - -Profitability + - + + + - + 0 Size - - - + - 0 - + R2 EV/EBIT P/NI 0.080 0.070 0.071 0.047 EV/IC P/B Table X
Expected and Estimated Signs of the Fundamental Variables in the Pooled Regression Analysis The expected signs are based on the direct and indirect effect of the variable on the multiple. + and – indicate positive and negative significant signs at the 5%-level. 0 indicates insignificance of the variable.
4.
Results
This study is designed to investigate country differences in valuing companies within the 27 countries of the EU. To accomplish this objective, a regression analysis is executed. First a general regression is run for each multiple. This analysis controls for growth, profitability, risk, size, industry classification, the country with the registered head office of the company and the year the multiple is recorded in.
Table X presents the expected signs (from Table V) and the estimated signs of the fundamental variables from the pooled regression analysis. The negative sign of profitability for the enterprise multiples are rather odd, because a higher profitability should generally lead to higher multiples. For the pooled regression, the sign could be explained by the development of the enterprise multiples and the return on assets over time, as discussed before. However, Table XI (Panel A and B) presents the coefficients of the cross-sectional regressions and shows that profitability remains to have a negative sign for 7 of 10 years for the EV/IC multiple and for all years of the EV/EBIT multiple. A negative sign of profitability is also observed by Bhojraj and Lee (2002) for their enterprise multiple in all years and Ros (2010) finds a negative sign for profitability for the EV/EBIT multiple in his pooled regression.
The R2 values in Table X are between 4.7% and 8.0%. Table XI (Panel C) shows the average of the cross-sectional R2, in order to compare the explanatory power of the model between the multiples. Averaging the cross-sectional R2 values, increases the R2 of the enterprise multiples and the P/NI multiple. Ros (2010) finds an average R2 of 5.9% for their EV/EBIT multiple and Bhojraj and Lee (2003) report an average R2 of 34% for the P/B and 28% for the P/NI multiples. Damodaran (Website) finds even higher R2 values (44% for P/B, 58% for EV/IC and 29% for P/NI). However, he selects different variables for each multiple, that have the highest explanatory power and he uses forward-looking financial information. The sample period used in this study, has substantial overlap with Ros (2010). The inclusion of
23 Table XI
Coefficients of Fundamental Variables in the Cross-sectional Regressions
Panel A present the values for the EV/IC multiple and Panel B for the EV/EBIT multiple. Panel C presents the average R2 per multiple. Bold-marked values are significant at the 10%-level. * means that the value is significant and has the appropriate sign and is therefore used in the calculation of the systematic multiple.
Panel A EV/IC 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 growth 0.093 0.040* -0.096 -0.023 -0.066 -0.010 0.064* 0.042* -0.070 0.116 leverage -0.987 -1.410 -0.774 -0.884 0.423* -0.545 -0.887 0.299* 0.554* -0.579 profitability -0.110 -0.255 1.222* 0.909* -0.115 0.123 -0.047 -0.020 -0.195 -0.160 size -0.034* -0.035* -0.026 -0.006 -0.004* -0.003* -0.011 -0.003 -0.011 -0.009 Panel B EVEBIT 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 growth 0.949* 0.369 0.901* 0.136 0.197 0.081* -0.205 0.353 -0.220 0.035 leverage 1.570* 1.630* 2.272* 1.877* 2.575* 2.485* 2.061* 0.883* 1.237* 2.286* profitability -0.459 -0.761 -0.961 -1.293 -0.808 -0.306 -0.845 -1.038 -0.558 -0.386 size -0.021 0.126 0.064 -0.073* 0.038 0.002 -0.044 -0.066* -0.081* -0.031 tax rate -0.225 -0.189 -0.990 0.164 0.344 0.154 -0.826 0.069 0.162 0.189
Panel C EV/IC P/B EV/EBIT P/NI
average R2 0.077 0.058 0.116 0.071
multiples outperform the equity multiples. Hermann and Richter (2003) find similar results, whereas Bhojraj and Lee (2002), Schreiner and Spremann (2007) and Tu (2010) find that equity multiples outperform enterprise multiples. Although the R2 values presented here are relatively low, the variables included in the regression analyses make sense from an economic point of view. Also, prior literature uses similar variables (Bhojraj and Lee [2002], Bhojraj et al [2003], Ros [2010]).
The Tables XIIa and XIIb show the signs of the significant dummy variables in the pooled regression analyses. Table XIIa indicates that there are significant country-effects for 10 to 14 countries depending on the type of multiple. The actual coefficients are included in Appendix D. The results indicate that 14 of 27 countries have a significant sign for the EV/IC and the EV/EBIT multiple, whereas the P/B and P/NI multiple have a significant sign for, respectively 11 and 10 countries. Ros (2010) finds 3 of 15 significant country-effects for the EV/EBIT multiple, whereas Dittmann and Weiner (2005) find significant
24
EV/IC P/B EV/EBIT P/NI EV/IC P/B EV/EBIT P/NI
Austria 0 - - 0 Administration, healthcare - - -
-Belgium 0 0 - - Agriculture, mining - - -
-Bulgaria 0 - 0 - Arts, entertainment, - - -
-Cyprus - - - - Construction, wholesale trade - - -
-Czech Rep. 0 - - - Electricity, water supply - - -
-Denmark + 0 - 0 ICT, real estate - - -
-Estonia + - 0 0 Chemicals, machinery - - -
-Finland + 0 - 0 Food, clothing, paper 0 - -
-France + 0 - 0 Transportation, storage - - - 0
Germany 0 0 - 0 2002 + - + -Greece 0 - 0 - 2003 + - - + Hungary 0 0 0 - 2004 + + - + Ireland + 0 0 0 2005 - + - + Italy 0 0 - 0 2006 - + + + Lithuania 0 - 0 - 2007 - + + + Luxemburg + 0 0 0 2008 0 - + -Latvia 0 - 0 - 2009 + - + + Netherlands + 0 - 0 2010 + - + + Poland 0 0 - 0 Portugal 0 - 0 0 Romania - - 0 -Sweden + 0 0 0 Slovenia - 0 0 0 Slovakia - - - -Spain + 0 - 0 UK + 0 - 0 Total +/- 14 11 14 10 Table XIIb
Signs of the Industry and Year Dummy Variables in the Pooled Regression Analysis
+ indicates a positive significant sign. – indicates a negative significant sign and 0 indicates an insignificant sign (at the
5%-level). The „professional and administrative‟ industry classification and the year 2001 dummy variables are
excluded to avoid multicollinearity. Table XIIa
Signs of the Country Dummy Variables in the Pooled Regression Analysis + indicates a positive significant sign. – indicates a negative significant sign and 0 indicates an insignificant sign (at the
5%-level). Malta is excluded to avoid multicollinearity.
25
EV/IC systematic multiple
int dom int dom int dom int dom int dom int dom
2001 -0.006 -0.012 0.084 0.064 -0.148 -0.142 0.002 -0.007 0.290 0.267 -0.154 -0.154 2002 -0.006 * 0.003 0.099 * 0.071 -0.202 * -0.214 0.024 0.023 0.502 0.486 -0.134 -0.154 2003 -0.021 -0.004 0.114 0.107 -0.181 -0.169 -0.000 -0.000 0.317 0.294 -0.153 * -0.146 2004 -0.004 -0.019 0.068 0.047 -0.177 -0.173 0.002 0.011 0.286 0.283 -0.137 * -0.128 2005 -0.005 0.000 0.090 0.097 -0.154 -0.155 0.006 -0.013 0.236 0.227 -0.153 -0.146 2006 0.002 -0.015 0.081 0.071 -0.119 -0.138 -0.002 0.016 0.250 0.229 -0.161 * -0.144 2007 -0.006 0.003 0.074 0.074 -0.140 * -0.129 0.007 -0.000 0.257 0.235 -0.146 -0.154 2008 0.009 -0.008 0.065 0.054 -0.121 -0.121 -0.011 * 0.007 0.243 0.212 -0.192 * -0.136 2009 0.000 -0.000 0.080 0.058 -0.109 -0.097 -0.005 0.010 0.205 0.215 -0.160 * -0.124 2010 -0.006 -0.004 0.058 0.067 -0.105 -0.090 0.000 * -0.006 0.210 0.186 -0.115 * -0.105 standard deviation 0.628 0.618 0.610 0.602 0.623 0.613 2.190 2.199 2.288 2.285 2.158 2.158
M edian M ean Harmonic mean M edian M ean
Table XIIIa
Average Valuation Error Percentage of the International and Domestic Peer Groups, that are Selected With the Systematic Multiple
The Wilcoxon signed rank test is used to compare the median valuation error of the international and domestic peer groups. The paired samples t-test is used for the mean and harmonic mean valuation error. The peer groups consist of 5 companies. * indicate that the valuation errors of the peer groups differ significantly at the 5%-level for the year and measure. Bold-marked values show per measure which peer group has a lower valuation error for that year and
measure. The last row shows the average standard deviation percentage of the valuation error.
Panel A EV/EBIT systematic Panel B
multiple
26
EV/IC systematic multiple
int dom int dom int dom int dom int dom int dom
2001 0.399 0.409 0.384 0.392 0.367 0.388 0.529 0.514 0.697 0.682 0.495 0.464 2002 0.416 * 0.421 0.435 * 0.449 0.427 * 0.443 0.665 0.670 1.152 1.059 0.530 0.532 2003 0.403 0.459 0.435 0.464 0.433 0.415 0.565 0.522 0.822 0.748 0.516 * 0.486 2004 0.394 0.388 0.408 0.388 0.388 0.384 0.492 0.542 0.658 0.706 0.462 * 0.462 2005 0.387 0.379 0.413 0.400 0.396 0.390 0.469 0.457 0.630 0.592 0.454 0.432 2006 0.352 0.348 0.369 0.363 0.363 0.358 0.463 0.457 0.622 0.590 0.457 * 0.433 2007 0.337 0.351 0.349 0.341 0.353 * 0.339 0.511 0.474 0.642 0.616 0.464 0.455 2008 0.355 0.332 0.362 0.344 0.356 0.345 0.492 * 0.459 0.636 0.574 0.490 * 0.424 2009 0.354 0.335 0.357 0.338 0.334 0.323 0.487 0.435 0.587 0.571 0.449 * 0.419 2010 0.334 0.329 0.336 0.346 0.345 0.327 0.431 * 0.407 0.549 0.520 0.409 * 0.393 standard deviation 0.407 0.400 0.379 0.376 0.439 0.431 1.957 1.965 1.901 1.907 2.004 2.006 Table XIIIb
Average Valuation Error Percentage of the International and Domestic Peer Groups, that are Selected With the Systematic Multiple
The Wilcoxon signed rank test is used to compare the absolute median valuation error of the international and domestic peer groups. The paired samples t-test is used for the mean and harmonic mean valuation error. The peer groups consist of 5 companies. * indicate that the valuation errors of the peer groups differ significantly at the 5%-level for the year and measure. Bold-marked values show per measure, which peer
group has a lower valuation error for that year. The last row shows the average standard deviation percentage of the valuation error.
Panel A EV/EBIT systematic
multiple Panel B
M edian M ean Harmonic mean M edian M ean Harmonic mean
Table XIIIb presents the absolute valuation error percentage. The valuation errors are higher, compared to Table XIIIa, whereas the standard deviation percentage is lower. The differences between the three measures are small, but the harmonic mean valuation errors are lowest for both multiples. This is in accordance with Baker and Ruback (1999) who find that the harmonic mean outperforms the median and average in terms of valuation error using multiples. However, the dispersion of the peer group multiple is lowest, indicating that the mean peer group multiples is more appropriate to use. Again, lower valuation errors and less dispersion is observed for the EV/EBIT multiple. Furthermore, the valuation error
percentage of the domestic peer group is lower in 6 to 8 years (EV/EBIT). This indicates that a peer group selection based on the systematic multiple is best executed domestically, for the EV/EBIT multiple.
27
int dom int dom int dom int dom int dom int dom
2001 0.309 * 0.357 0.355 * 0.392 0.313 * 0.349 0.436 * 0.443 0.734 * 0.917 0.377 * 0.391 2002 0.309 * 0.386 0.355 * 0.416 0.342 * 0.399 0.449 * 0.494 0.825 * 1.054 0.397 * 0.409 2003 0.339 * 0.399 0.398 * 0.449 0.365 * 0.486 0.438 * 0.500 0.886 * 1.008 0.392 * 0.400 2004 0.326 * 0.381 0.358 * 0.405 0.351 * 0.468 0.400 * 0.457 0.706 * 0.846 0.361 * 0.377 2005 0.348 * 0.370 0.387 * 0.401 0.434 * 0.706 0.374 * 0.393 0.711 * 0.803 0.369 * 0.439 2006 0.305 * 0.329 0.343 * 0.358 0.348 * 0.492 0.379 * 0.410 0.707 * 0.800 0.385 * 0.450 2007 0.262 * 0.301 0.303 * 0.320 0.331 * 0.406 0.414 * 0.436 0.755 * 0.897 0.397 * 0.427 2008 0.270 * 0.304 0.305 * 0.326 0.317 * 0.445 0.372 * 0.452 0.703 * 0.876 0.398 * 0.481 2009 0.269 * 0.299 0.303 * 0.333 0.295 * 0.463 0.379 * 0.442 0.653 * 0.755 0.353 * 0.487 2010 0.292 * 0.321 0.327 * 0.347 0.316 * 0.366 0.373 * 0.411 0.684 * 0.746 0.361 * 0.380 standard deviation 0.385 0.385 0.342 0.37 0.466 0.525 2.093 2.162 1.834 1.912 2.197 2.260
M edian M ean Harmonic mean M edian M ean
Table XIV
Average Absolute Valuation Error Percentage of the International and Domestic Peer Groups, Based on Industry Classification
The Wilcoxon signed rank test is used to compare the median of the international and domestic peer groups. The paired samples t-test is used for the mean and harmonic mean. The peer groups consist of 5 companies. * indicates
that the valuation error of the peer groups differ significantly at the 5%-level per year and measure. Bold-marked values show per measure which peer group has the lowest valuation error. The last row shows the average standard
deviation percentage of the valuation error.
Panel A EV/EBIT industry
selection Panel B EV/IC industry selection
Harmonic mean
This means that when the comparable company selection is based on the industry, the international peer group has a significantly lower valuation error, than the domestic peer group. An explanation for the result can be that the number of comparable companies within the same industry is larger in the
international peer group, so that the number of appropriate comparable companies increases. This result contradicts Ros (2010) and Liu et al (2007), who conclude that international differences between countries exist and that a valuation based on multiples should control for this fact. The result also contradicts Liu et al (2007), who find significant differences between countries. However, they investigate countries from different continents, whereas this study restricts itself to the EU27.
Furthermore, Table XIV shows that the median peer group multiple has the lowest valuation errors for the EV/EBIT multiple and the harmonic mean for the EV/IC multiple. Comparable to Table XIIIa/b, the mean peer group multiple has the least dispersion for both multiples.
28
syst ind syst ind syst ind syst ind syst ind syst ind
2001 0.399 * 0.309 0.384 * 0.355 0.367 * 0.313 0.409 * 0.357 0.392 * 0.392 0.388 * 0.349 2002 0.416 * 0.309 0.435 * 0.355 0.427 * 0.342 0.421 * 0.386 0.449 * 0.416 0.443 * 0.399 2003 0.403 * 0.339 0.435 * 0.398 0.433 * 0.365 0.459 * 0.399 0.464 * 0.449 0.415 * 0.486 2004 0.394 * 0.326 0.408 * 0.358 0.388 * 0.351 0.388 * 0.381 0.388 * 0.405 0.384 * 0.468 2005 0.387 * 0.348 0.413 * 0.387 0.396 * 0.434 0.379 * 0.370 0.400 * 0.401 0.390 * 0.706 2006 0.352 * 0.305 0.369 * 0.343 0.363 * 0.348 0.348 * 0.329 0.363 * 0.358 0.358 * 0.492 2007 0.337 * 0.262 0.349 * 0.303 0.353 * 0.331 0.351 * 0.301 0.341 * 0.320 0.339 * 0.406 2008 0.355 * 0.270 0.362 * 0.305 0.356 * 0.317 0.332 * 0.304 0.344 * 0.326 0.345 * 0.445 2009 0.354 * 0.269 0.357 * 0.303 0.334 * 0.295 0.335 * 0.299 0.338 * 0.333 0.323 * 0.463 2010 0.334 * 0.292 0.336 * 0.327 0.345 * 0.316 0.329 * 0.321 0.346 * 0.347 0.327 * 0.366 standard deviation 0.407 0.381 0.379 0.342 0.439 0.466 0.4 0.385 0.375 0.361 0.431 0.525
Table XV
Average Absolute Valuation Error Percentage of the Selection Methods: Systematic Multiple and Industry Classification, for the International and Domestic Peer Groups
The Wilcoxon signed rank test is used to compare the median absolute valuation error of the international and domestic peer groups. The paired samples t-test is used for the mean and harmonic mean. The peer
groups consist of 5 companies. * means that the absolute valuation error of the peer groups differ significantly at the 5%-level for that year and measure. Bold-marked values show per measure which peer group has the lowest valuation error. The last row shows the average percentage standard deviation of the
valuation error. Panel A EV/EBIT international
peer group Panel B
EV/EBIT domestic peer group
M edian M ean Harmonic mean M edian M ean Harmonic mean
valuation errors. Furthermore, lower valuation errors are observed when the harmonic mean measure is used, whereas the median measure returns lower valuation errors for the EV/EBIT multiple.
Next, the selection methods are investigated more closely, by comparing the valuation errors of both methods for both peer groups. Table XV presents this comparison for the EV/EBIT multiple and Table XVI for the EV/IC multiple. Table XV shows that the valuation error percentages of the international peer groups, based on industry selection (Panel A), are significantly lower in all years and for all measures. The same is observed for the domestic peer group (Panel B) for the median and mean measures. These results confirm Alford (1992), who finds that a company selection based on fundamental variables does not improve valuation compared to a selection based on industry classification. Also, this result
![Leverton, B.J.T. 1990. Records of Natal, III, August 1835 - June 1838. [Book review]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACH5BAEAAAAALAAAAAABAAEAAAICRAEAOw==)